01-database-procs.tcl

An API for managing database queries.

Location:
packages/acs-tcl/tcl/01-database-procs.tcl
Created:
15 Apr 2000
Author:
Jon Salz <jsalz@arsdigita.com>
CVS Identification:
$Id: 01-database-procs.tcl,v 1.5 2024/10/21 15:49:22 gustafn Exp $

Procedures in this file

Detailed information

ad_column_type (public, deprecated)

 ad_column_type [ -dbn dbn ] table_name column_name
Deprecated. Invoking this procedure generates a warning.

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Parameters:
table_name (required)
column_name (required)
Returns:
'numeric' for number type columns, 'text' otherwise Throws an error if no such column exists.
Author:
Yon Feldman (yon@arsdigita.com) DEPRECATED: it is unclear what the purpose of this proc is. For instance, on a Linux/Postgres installation, ad_column_type acs_objects object_type -> 'numeric'. When things should happen based on the column type, maybe a better approach is to rely on more complete or consistent api, or on the information schema.
See Also:
  • db_column_type, https://wikipedia.org/wiki/Information_schema

Partial Call Graph (max 5 caller/called nodes):
%3 ad_log_deprecated ad_log_deprecated (public) db_column_type db_column_type (public) ad_column_type ad_column_type ad_column_type->ad_log_deprecated ad_column_type->db_column_type

Testcases:
No testcase defined.

callback::subsite::parameter_changed::impl::acs-tcl (private)

 callback::subsite::parameter_changed::impl::acs-tcl \
    -package_id package_id -parameter parameter -value value

Implementation of subsite::parameter_changed for acs-tcl.

Switches:
-package_id (required)
the package_id of the package the parameter was changed for
-parameter (required)
the parameter name
-value (required)
the new value
See Also:

Partial Call Graph (max 5 caller/called nodes):
%3

Testcases:
No testcase defined.

db_0or1row (public)

 db_0or1row [ -dbn dbn ] [ -cache_key cache_key ] \
    [ -cache_pool cache_pool ] [ -subst subst ] statement_name sql \
    [ -bind bind ] [ -column_array column_array ] \
    [ -column_set column_set ]

Performs the specified SQL query. If a row is returned, sets variables to column values (or a set or array populated if -column_array or column_set is specified) and returns 1.

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-cache_key (optional)
Cache the result using given value as the key. Default is to not cache.
-cache_pool (optional, defaults to "db_cache_pool")
Override the default db_cache_pool
-subst (optional, defaults to "all")
Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands
-bind (optional)
bind variables, passed either as an ns_set id, or via bind value list
-column_array (optional)
array to be populated with values
-column_set (optional)
ns_set to be populated with values
Parameters:
statement_name (required)
name of the SQL query
sql (required)
SQL query to be executed
Returns:
1 if variables are set, 0 if no rows are returned. If more than one row is returned, throws an error.

Partial Call Graph (max 5 caller/called nodes):
%3 test_db__0or1row db__0or1row (test acs-tcl) db_0or1row db_0or1row test_db__0or1row->db_0or1row test_db__1row db__1row (test acs-tcl) test_db__1row->db_0or1row test_db__caching db__caching (test acs-tcl) test_db__caching->db_0or1row db_exec db_exec (public) db_0or1row->db_exec db_qd_get_fullname db_qd_get_fullname (public) db_0or1row->db_qd_get_fullname db_with_handle db_with_handle (public) db_0or1row->db_with_handle acs::test::require_package_instance acs::test::require_package_instance (public) acs::test::require_package_instance->db_0or1row acs_admin::check_expired_certificates acs_admin::check_expired_certificates (private) acs_admin::check_expired_certificates->db_0or1row acs_mail_lite::get_address_array acs_mail_lite::get_address_array (private) acs_mail_lite::get_address_array->db_0or1row acs_mail_lite::imap_conn_set acs_mail_lite::imap_conn_set (private) acs_mail_lite::imap_conn_set->db_0or1row acs_mail_lite::inbound_cache_hit_p acs_mail_lite::inbound_cache_hit_p (private) acs_mail_lite::inbound_cache_hit_p->db_0or1row

Testcases:
db__caching, db__0or1row, db__1row

db_1row (public)

 db_1row [ -subst subst ] [ args... ]

A wrapper for db_0or1row, which produces an error if no rows are returned.

Switches:
-subst (optional, defaults to "all")
Returns:
1 if variables are set, otherwise an exception is thrown.
See Also:

Partial Call Graph (max 5 caller/called nodes):
%3 test_acs_object_procs_test acs_object_procs_test (test acs-tcl) db_1row db_1row test_acs_object_procs_test->db_1row test_acs_user__ban_approve acs_user__ban_approve (test acs-tcl) test_acs_user__ban_approve->db_1row test_acs_user__registered_user_p acs_user__registered_user_p (test acs-tcl) test_acs_user__registered_user_p->db_1row test_ad_context_bar_multirow ad_context_bar_multirow (test acs-tcl) test_ad_context_bar_multirow->db_1row test_ad_proc_permission_permission_p ad_proc_permission_permission_p (test acs-tcl) test_ad_proc_permission_permission_p->db_1row Class ::acs::SiteNode Class ::acs::SiteNode (public) Class ::acs::SiteNode->db_1row Class ::xo::db::Class Class ::xo::db::Class (public) Class ::xo::db::Class->db_1row Class ::xo::db::CrClass Class ::xo::db::CrClass (public) Class ::xo::db::CrClass->db_1row Class ::xo::db::CrFolder Class ::xo::db::CrFolder (public) Class ::xo::db::CrFolder->db_1row acs::SiteNode instproc properties acs::SiteNode instproc properties (protected) acs::SiteNode instproc properties->db_1row

Testcases:
test_apm_parameter__register, test_apm_package_instance__new, apm_version_names_compare, apm__test_info_file, site_node_get_children, acs_user__registered_user_p, acs_user__ban_approve, parameter_register_test, parameter__check_procs, community_cc_procs, person_procs_test, party_procs_test, db__1row, util_http_json_encoding, ad_context_bar_multirow, acs_object_procs_test, ad_proc_permission_permission_p

db_abort_transaction (public)

 db_abort_transaction [ -dbn dbn ]

Aborts all levels of a transaction. That is if this is called within several nested transactions, all of them are terminated. Use this instead of db_dml "abort" "abort transaction".

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.

Partial Call Graph (max 5 caller/called nodes):
%3 test_db__transaction db__transaction (test acs-tcl) db_abort_transaction db_abort_transaction test_db__transaction->db_abort_transaction db_state_array_name_is db_state_array_name_is (private) db_abort_transaction->db_state_array_name_is db_with_handle db_with_handle (public) db_abort_transaction->db_with_handle db_transaction db_transaction (public) db_transaction->db_abort_transaction forum::message::new forum::message::new (public) forum::message::new->db_abort_transaction packages/acs-subsite/www/admin/rel-segments/constraints/new.tcl packages/acs-subsite/ www/admin/rel-segments/constraints/new.tcl packages/acs-subsite/www/admin/rel-segments/constraints/new.tcl->db_abort_transaction

Testcases:
db__transaction

db_abort_transaction_p (private)

 db_abort_transaction_p [ -dbn dbn ]
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.

Partial Call Graph (max 5 caller/called nodes):
%3 db_transaction db_transaction (public) db_abort_transaction_p db_abort_transaction_p db_transaction->db_abort_transaction_p db_state_array_name_is db_state_array_name_is (private) db_abort_transaction_p->db_state_array_name_is db_with_handle db_with_handle (public) db_abort_transaction_p->db_with_handle

Testcases:
No testcase defined.

db_bind_var_quoted_p (private)

 db_bind_var_quoted_p sql bind_start_idx bind_end_idx
Parameters:
sql (required)
bind_start_idx (required)
bind_end_idx (required)

Partial Call Graph (max 5 caller/called nodes):
%3 db_bind_var_substitution db_bind_var_substitution (public) db_bind_var_quoted_p db_bind_var_quoted_p db_bind_var_substitution->db_bind_var_quoted_p db_get_quote_indices db_get_quote_indices (private) db_bind_var_quoted_p->db_get_quote_indices

Testcases:
No testcase defined.

db_bind_var_substitution (public)

 db_bind_var_substitution sql [ bind ]

This proc emulates the bind variable substitution in the PostgreSQL driver. Since this is a temporary hack, we do it in Tcl instead of hacking up the driver to support plsql calls. This is only used for the db_exec_plpgsql function.

Parameters:
sql (required)
bind (optional)

Partial Call Graph (max 5 caller/called nodes):
%3 test_db_bind_var_substitution db_bind_var_substitution (test acs-tcl) db_bind_var_substitution db_bind_var_substitution test_db_bind_var_substitution->db_bind_var_substitution db_bind_var_quoted_p db_bind_var_quoted_p (private) db_bind_var_substitution->db_bind_var_quoted_p ns_dbquotevalue ns_dbquotevalue db_bind_var_substitution->ns_dbquotevalue db_exec_lob_postgresql db_exec_lob_postgresql (private) db_exec_lob_postgresql->db_bind_var_substitution db_exec_plpgsql db_exec_plpgsql (private) db_exec_plpgsql->db_bind_var_substitution ds_collect_db_call ds_collect_db_call (public) ds_collect_db_call->db_bind_var_substitution

Testcases:
db_bind_var_substitution

db_blob_get (public)

 db_blob_get [ -dbn dbn ] [ -subst subst ] statement_name sql \
    [ args... ]

PostgreSQL only.

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-subst (optional, defaults to "all")
Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands
Parameters:
statement_name (required)
sql (required)

Partial Call Graph (max 5 caller/called nodes):
%3 cr_write_content-lob cr_write_content-lob (private) db_blob_get db_blob_get cr_write_content-lob->db_blob_get search::content_get search::content_get (private) search::content_get->db_blob_get ad_arg_parser ad_arg_parser (public) db_blob_get->ad_arg_parser db_driverkey db_driverkey (public) db_blob_get->db_driverkey db_exec_lob db_exec_lob (private) db_blob_get->db_exec_lob db_qd_get_fullname db_qd_get_fullname (public) db_blob_get->db_qd_get_fullname db_qd_replace_sql db_qd_replace_sql (public) db_blob_get->db_qd_replace_sql

Testcases:
No testcase defined.

db_blob_get_file (public)

 db_blob_get_file [ -dbn dbn ] statement_name sql [ args... ]
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.

TODO: This proc should probably be changed to take a final file argument, only, rather than the current args variable length argument list. Currently, it is called only 4 places in OpenACS, and each place args, if used at all, is always "-file $file". However, such a change might break custom code... I'm not sure. --atp@piskorski.com, 2003/04/09 11:39 EDT

Parameters:
statement_name (required)
sql (required)

Partial Call Graph (max 5 caller/called nodes):
%3 test_apm_tarballs apm_tarballs (test acs-tcl) db_blob_get_file db_blob_get_file test_apm_tarballs->db_blob_get_file ad_arg_parser ad_arg_parser (public) db_blob_get_file->ad_arg_parser db_driverkey db_driverkey (public) db_blob_get_file->db_driverkey db_exec_lob db_exec_lob (private) db_blob_get_file->db_exec_lob db_qd_get_fullname db_qd_get_fullname (public) db_blob_get_file->db_qd_get_fullname db_with_handle db_with_handle (public) db_blob_get_file->db_with_handle apm_extract_tarball apm_extract_tarball (public) apm_extract_tarball->db_blob_get_file content::revision::export_to_filesystem-lob content::revision::export_to_filesystem-lob (private) content::revision::export_to_filesystem-lob->db_blob_get_file packages/acs-subsite/www/shared/portrait-bits.tcl packages/acs-subsite/ www/shared/portrait-bits.tcl packages/acs-subsite/www/shared/portrait-bits.tcl->db_blob_get_file packages/photo-album/www/album-export.tcl packages/photo-album/ www/album-export.tcl packages/photo-album/www/album-export.tcl->db_blob_get_file publish::write_multiple_blobs publish::write_multiple_blobs (private) publish::write_multiple_blobs->db_blob_get_file

Testcases:
apm_tarballs

db_boolean (public)

 db_boolean bool

Converts a Tcl boolean (1/0) into a SQL boolean (t/f)

Parameters:
bool (required)
Returns:
t or f

Partial Call Graph (max 5 caller/called nodes):
%3 test_db_boolean db_boolean (test acs-tcl) db_boolean db_boolean test_db_boolean->db_boolean auth::sync::job::end_get_document auth::sync::job::end_get_document (public) auth::sync::job::end_get_document->db_boolean auth::sync::job::start auth::sync::job::start (public) auth::sync::job::start->db_boolean group::member_p_not_cached group::member_p_not_cached (private) group::member_p_not_cached->db_boolean lang::message::edit lang::message::edit (private) lang::message::edit->db_boolean lang::message::register lang::message::register (public) lang::message::register->db_boolean

Testcases:
db_boolean

db_bounce_pools (public)

 db_bounce_pools [ -dbn dbn ]
Switches:
-dbn (optional)
The database name to use. Uses the default database if not supplied.
Returns:
Call ns_db bouncepool on all pools for the named database.

Partial Call Graph (max 5 caller/called nodes):
%3 apm_package_install_data_model apm_package_install_data_model (private) db_bounce_pools db_bounce_pools apm_package_install_data_model->db_bounce_pools db_available_pools db_available_pools (public) db_bounce_pools->db_available_pools

Testcases:
No testcase defined.

db_column_exists (public)

 db_column_exists [ -dbn dbn ] table_name column_name
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Parameters:
table_name (required)
column_name (required)
Returns:
1 if the row exists in the table, 0 if not.
Author:
Lars Pind <lars@pinds.com>

Partial Call Graph (max 5 caller/called nodes):
%3 test_acs_subsite_attributes acs_subsite_attributes (test acs-subsite) db_column_exists db_column_exists test_acs_subsite_attributes->db_column_exists db_string db_string (public) db_column_exists->db_string attribute::delete attribute::delete (public) attribute::delete->db_column_exists packages/schema-browser/www/column-comments-2.tcl packages/schema-browser/ www/column-comments-2.tcl packages/schema-browser/www/column-comments-2.tcl->db_column_exists

Testcases:
acs_subsite_attributes

db_column_type (public)

 db_column_type [ -dbn dbn ] [ -complain ] table_name column_name
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-complain (optional, boolean)
throw an error when datatype is not found
Parameters:
table_name (required)
column_name (required)
Returns:
the Oracle Data Type for the specified column.
-1 if the table or column doesn't exist.
an error if table or column doesn't exist and -complain flag was specified
Author:
Yon Feldman <yon@arsdigita.com>

Partial Call Graph (max 5 caller/called nodes):
%3 test_datamodel__acs_attribute_check datamodel__acs_attribute_check (test acs-tcl) db_column_type db_column_type test_datamodel__acs_attribute_check->db_column_type db_string db_string (public) db_column_type->db_string ad_column_type ad_column_type (public, deprecated) ad_column_type->db_column_type plpgsql_utility::generate_function_signature plpgsql_utility::generate_function_signature (public) plpgsql_utility::generate_function_signature->db_column_type plpgsql_utility::table_column_type plpgsql_utility::table_column_type (public, deprecated) plpgsql_utility::table_column_type->db_column_type xo::db::require proc default xo::db::require proc default xo::db::require proc default->db_column_type

Testcases:
datamodel__acs_attribute_check

db_columns (public)

 db_columns [ -dbn dbn ] table_name
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Parameters:
table_name (required)
Returns:
a Tcl list of all the columns in the table with the given name.
Author:
Lars Pind <lars@pinds.com>
Changelog:
yon@arsdigita.com 20000711 changed to return lowercase column names

Partial Call Graph (max 5 caller/called nodes):
%3 test_datamodel__acs_attribute_check datamodel__acs_attribute_check (test acs-tcl) db_columns db_columns test_datamodel__acs_attribute_check->db_columns db_foreach db_foreach (public) db_columns->db_foreach

Testcases:
datamodel__acs_attribute_check

db_compatible_rdbms_p (public)

 db_compatible_rdbms_p db_type
Parameters:
db_type (required)
Returns:
1 if the given db_type is compatible with the current RDBMS.

Partial Call Graph (max 5 caller/called nodes):
%3 tsearch2::build_query tsearch2::build_query (private) db_compatible_rdbms_p db_compatible_rdbms_p tsearch2::build_query->db_compatible_rdbms_p db_type db_type (public) db_compatible_rdbms_p->db_type

Testcases:
No testcase defined.

db_dml (public)

 db_dml [ -dbn dbn ] [ -subst subst ] statement_name sql [ args... ]

Do a DML statement.

args can be one of: -clobs, -blobs, -clob_files or -blob_files. See the db-api doc referenced below for more information.

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-subst (optional, defaults to "all")
Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands
Parameters:
statement_name (required)
sql (required)
See Also:

Partial Call Graph (max 5 caller/called nodes):
%3 test_apm__test_info_file apm__test_info_file (test acs-tcl) db_dml db_dml test_apm__test_info_file->db_dml test_db__transaction db__transaction (test acs-tcl) test_db__transaction->db_dml test_db__transaction_bug_3440 db__transaction_bug_3440 (test acs-tcl) test_db__transaction_bug_3440->db_dml ad_arg_parser ad_arg_parser (public) db_dml->ad_arg_parser db_driverkey db_driverkey (public) db_dml->db_driverkey db_exec db_exec (public) db_dml->db_exec db_qd_get_fullname db_qd_get_fullname (public) db_dml->db_qd_get_fullname db_with_handle db_with_handle (public) db_dml->db_with_handle Class ::xo::Authorize Class ::xo::Authorize (public) Class ::xo::Authorize->db_dml aa_log_final aa_log_final (private) aa_log_final->db_dml aa_log_result aa_log_result (public) aa_log_result->db_dml aa_run_testcase aa_run_testcase (private) aa_run_testcase->db_dml acs::test::user::delete acs::test::user::delete (public) acs::test::user::delete->db_dml

Testcases:
apm__test_info_file, db__transaction, db__transaction_bug_3440

db_driverkey (public)

 db_driverkey [ -handle_p handle_p ] dbn

Normally, a dbn is passed to this proc. Unfortunately, there are one or two cases where a proc that needs to call this one has only a db handle, not the dbn that handle came from. Therefore, they instead use -handle_p 1 and pass the db handle. Hmm, as of 2018, it seems that in most cases, db_driverkey is called with a handle.

Switches:
-handle_p (optional, defaults to "0")
Parameters:
dbn (required)
Returns:
The driverkey for use in db_* API switch statements.
Author:
Andrew Piskorski <atp@piskorski.com>
Created:
2003/04/08

Partial Call Graph (max 5 caller/called nodes):
%3 test_create_form_with_form_instance create_form_with_form_instance (test xowiki) db_driverkey db_driverkey test_create_form_with_form_instance->db_driverkey db_with_handle db_with_handle (public) db_driverkey->db_with_handle Class ::acs::SiteNode Class ::acs::SiteNode (public) Class ::acs::SiteNode->db_driverkey Class ::acs::SiteNodesCache Class ::acs::SiteNodesCache (public) Class ::acs::SiteNodesCache->db_driverkey acs::SiteNode instproc get_children acs::SiteNode instproc get_children (public) acs::SiteNode instproc get_children->db_driverkey acs::SiteNodesCache instproc flush_cache acs::SiteNodesCache instproc flush_cache (public) acs::SiteNodesCache instproc flush_cache->db_driverkey acs::db::require_dc acs::db::require_dc (private) acs::db::require_dc->db_driverkey

Testcases:
create_form_with_form_instance

db_exec (public)

 db_exec [ -subst subst ] type db statement_name pre_sql [ ulevel ] \
    [ args... ]

A helper procedure to execute a SQL statement, potentially binding depending on the value of the $bind variable in the calling environment (if set).

Switches:
-subst (optional, defaults to "all")
Parameters:
type (required)
db (required)
statement_name (required)
pre_sql (required)
ulevel (optional, defaults to "2")

Partial Call Graph (max 5 caller/called nodes):
%3 test_db__0or1row db__0or1row (test acs-tcl) db_exec db_exec test_db__0or1row->db_exec test_db__1row db__1row (test acs-tcl) test_db__1row->db_exec test_xowiki_test_cases xowiki_test_cases (test xowiki) test_xowiki_test_cases->db_exec db_driverkey db_driverkey (public) db_exec->db_driverkey db_qd_replace_sql db_qd_replace_sql (public) db_exec->db_qd_replace_sql ds_collect_db_call ds_collect_db_call (public) db_exec->ds_collect_db_call db_0or1row db_0or1row (public) db_0or1row->db_exec db_dml db_dml (public) db_dml->db_exec db_exec_plsql db_exec_plsql (public) db_exec_plsql->db_exec db_list db_list (public) db_list->db_exec db_list_of_lists db_list_of_lists (public) db_list_of_lists->db_exec

Testcases:
db__0or1row, db__1row, xowiki_test_cases

db_exec_lob (private)

 db_exec_lob [ -ulevel ulevel ] type db statement_name pre_sql [ file ]

A helper procedure to execute a SQL statement, potentially binding depending on the value of the $bind variable in the calling environment (if set).

Switches:
-ulevel (optional, defaults to "2")
Parameters:
type (required)
db (required)
statement_name (required)
pre_sql (required)
file (optional)

Partial Call Graph (max 5 caller/called nodes):
%3 db_blob_get db_blob_get (public) db_exec_lob db_exec_lob db_blob_get->db_exec_lob db_blob_get_file db_blob_get_file (public) db_blob_get_file->db_exec_lob db_write_blob db_write_blob (public) db_write_blob->db_exec_lob db_driverkey db_driverkey (public) db_exec_lob->db_driverkey

Testcases:
No testcase defined.

db_exec_lob_oracle (private)

 db_exec_lob_oracle [ -ulevel ulevel ] [ -subst subst ] type db \
    statement_name pre_sql [ file ]

A helper procedure to execute a SQL statement, potentially binding depending on the value of the $bind variable in the calling environment (if set).

Switches:
-ulevel (optional, defaults to "2")
-subst (optional, defaults to "all")
Parameters:
type (required)
db (required)
statement_name (required)
pre_sql (required)
file (optional)

Partial Call Graph (max 5 caller/called nodes):
%3 ad_file ad_file (public) db_qd_replace_sql db_qd_replace_sql (public) ds_collect_db_call ds_collect_db_call (public) db_exec_lob_oracle db_exec_lob_oracle db_exec_lob_oracle->ad_file db_exec_lob_oracle->db_qd_replace_sql db_exec_lob_oracle->ds_collect_db_call

Testcases:
No testcase defined.

db_exec_lob_postgresql (private)

 db_exec_lob_postgresql [ -ulevel ulevel ] [ -subst subst ] type db \
    statement_name pre_sql [ file ]

A helper procedure to execute a SQL statement, potentially binding depending on the value of the $bind variable in the calling environment (if set). Low level replacement for db_exec which emulates blob handling.

Switches:
-ulevel (optional, defaults to "2")
-subst (optional, defaults to "all")
Parameters:
type (required)
db (required)
statement_name (required)
pre_sql (required)
file (optional)

Partial Call Graph (max 5 caller/called nodes):
%3 ad_file ad_file (public) db_bind_var_substitution db_bind_var_substitution (public) db_qd_replace_sql db_qd_replace_sql (public) ds_collect_db_call ds_collect_db_call (public) db_exec_lob_postgresql db_exec_lob_postgresql db_exec_lob_postgresql->ad_file db_exec_lob_postgresql->db_bind_var_substitution db_exec_lob_postgresql->db_qd_replace_sql db_exec_lob_postgresql->ds_collect_db_call

Testcases:
No testcase defined.

db_exec_plpgsql (private)

 db_exec_plpgsql db statement_name pre_sql fname

PostgreSQL only.

A helper procedure to execute a SQL statement, potentially binding depending on the value of the $bind variable in the calling environment (if set).

Low level replacement for db_exec which replaces inline code with a proc. db proc is dropped after execution. This is a temporary fix until we can port all of the db_exec_plsql calls to simple selects of the inline code wrapped in function calls.

emulation of plsql calls from oracle. This routine takes the plsql statements and wraps them in a function call, calls the function, and then drops the function. Future work might involve converting this to cache the function calls

This proc is private - use db_exec_plsql instead!

Parameters:
db (required)
statement_name (required)
pre_sql (required)
fname (required)
See Also:

Partial Call Graph (max 5 caller/called nodes):
%3 db_exec_plsql db_exec_plsql (public) db_exec_plpgsql db_exec_plpgsql db_exec_plsql->db_exec_plpgsql db_bind_var_substitution db_bind_var_substitution (public) db_exec_plpgsql->db_bind_var_substitution db_nextval db_nextval (public) db_exec_plpgsql->db_nextval db_qd_replace_sql db_qd_replace_sql (public) db_exec_plpgsql->db_qd_replace_sql ds_collect_db_call ds_collect_db_call (public) db_exec_plpgsql->ds_collect_db_call ns_dbquotevalue ns_dbquotevalue db_exec_plpgsql->ns_dbquotevalue

Testcases:
No testcase defined.

db_exec_plsql (public)

 db_exec_plsql [ -dbn dbn ] statement_name sql [ -bind bind ]

Oracle: Executes a PL/SQL statement, and returns the variable of bind variable :1.

PostgreSQL: Performs a pl/pgsql function or procedure call. The caller must perform a select query that returns the value of the function.

Examples:

    # Oracle:
    db_exec_plsql delete_note {
        begin  note.del(:note_id);  end;
    }

    # PostgreSQL:
    db_exec_plsql delete_note {
        select note__delete(:note_id);
    }
    

If you need the return value, then do something like this:

    # Oracle:
    set new_note_id [db_exec_plsql create_note {
        begin
        :1 := note.new(
                       owner_id => :user_id,
                       title    => :title,
                       body     => :body,
                       creation_user => :user_id,
                       creation_ip   => :peeraddr,
                       context_id    => :package_id
                       );
        end;
    }]

    # PostgreSQL:
    set new_note_id [db_exec_plsql create_note {
        select note__new(
                         null,
                         :user_id,
                         :title,
                         :body,
                         'note',
                         now(),
                         :user_id,
                         :peeraddr,
                         :package_id
                         );
    }]
    

You can call several pl/SQL statements at once, like this:

    # Oracle:
    db_exec_plsql delete_note {
        begin
        note.del(:note_id);
        note.del(:another_note_id);
        note.del(:yet_another_note_id);
        end;
    }

    # PostgreSQL:
    db_exec_plsql delete_note {
        select note__delete(:note_id);
        select note__delete(:another_note_id);
        select note__delete(:yet_another_note_id);
    }
    
If you are using xql files then put the body of the query in a yourfilename-oracle.xql or yourfilename-postgresql.xql file, as appropriate. E.g. the first example transformed to use xql files looks like this:

yourfilename.tcl:

    db_exec_plsql delete_note {}

yourfilename-oracle.xql:

    <fullquery name="delete_note">
    <querytext>
    begin
    note.del(:note_id);
    end;
    </querytext>
    </fullquery>

yourfilename-postgresql.xql:

    <fullquery name="delete_note">
    <querytext>
    select note__delete(:note_id);
    </querytext>
    </fullquery>

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-bind (optional)
Parameters:
statement_name (required)
sql (required)
See Also:

Partial Call Graph (max 5 caller/called nodes):
%3 test_db_bind_var_substitution db_bind_var_substitution (test acs-tcl) db_exec_plsql db_exec_plsql test_db_bind_var_substitution->db_exec_plsql db_driverkey db_driverkey (public) db_exec_plsql->db_driverkey db_exec db_exec (public) db_exec_plsql->db_exec db_exec_plpgsql db_exec_plpgsql (private) db_exec_plsql->db_exec_plpgsql db_qd_get_fullname db_qd_get_fullname (public) db_exec_plsql->db_qd_get_fullname db_qd_replace_sql db_qd_replace_sql (public) db_exec_plsql->db_qd_replace_sql acs_cr_scheduled_release_exec acs_cr_scheduled_release_exec (private) acs_cr_scheduled_release_exec->db_exec_plsql acs_sc::contract::new acs_sc::contract::new (public) acs_sc::contract::new->db_exec_plsql acs_sc::contract::operation::new acs_sc::contract::operation::new (public) acs_sc::contract::operation::new->db_exec_plsql acs_sc::impl::alias::new acs_sc::impl::alias::new (public) acs_sc::impl::alias::new->db_exec_plsql acs_sc::impl::binding::new acs_sc::impl::binding::new (public) acs_sc::impl::binding::new->db_exec_plsql

Testcases:
db_bind_var_substitution

db_flush_cache (public)

 db_flush_cache [ -cache_key_pattern cache_key_pattern ] \
    [ -cache_pool cache_pool ]

Flush the given cache of entries with keys that match the given pattern.

Switches:
-cache_key_pattern (optional, defaults to "*")
The "string match" pattern used to flush keys (default is to flush all entries)
-cache_pool (optional, defaults to "db_cache_pool")
The pool to flush (default is to flush db_cache_pool)
Author:
Don Baccus <dhogasa@pacifier.com>

Partial Call Graph (max 5 caller/called nodes):
%3 test_db__caching db__caching (test acs-tcl) db_flush_cache db_flush_cache test_db__caching->db_flush_cache acs::clusterwide acs::clusterwide db_flush_cache->acs::clusterwide acs_mail_lite::section_id_of acs_mail_lite::section_id_of (private) acs_mail_lite::section_id_of->db_flush_cache application_group::delete application_group::delete (public) application_group::delete->db_flush_cache packages/acs-lang/www/admin/locale-edit.tcl packages/acs-lang/ www/admin/locale-edit.tcl packages/acs-lang/www/admin/locale-edit.tcl->db_flush_cache

Testcases:
db__caching

db_foreach (public)

 db_foreach [ -dbn dbn ] [ -subst subst ] statement_name sql \
    [ args... ]

Usage:

db_foreach statement-name sql [ -bind bind_set_id | -bind bind_value_list ] [ -column_array array_name | -column_set set_name ] code_block [ if_no_rows if_no_rows_block ]

Performs the SQL query sql, executing code_block once for each row with variables set to column values (or a set or array populated if -column_array or column_set is specified). If the query returns no rows, executes if_no_rows_block (if provided). In place of 'if_no_rows' also the 'else' keyword can be used.

Example:

db_foreach greeble_query "select foo, bar from greeble" {
        ns_write "<li>foo=$foo; bar=$bar\n"
    } if_no_rows {
        # This block is optional.
        ns_write "<li>No greebles!\n"
    }

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-subst (optional, defaults to "all")
Parameters:
statement_name (required)
sql (required)

Partial Call Graph (max 5 caller/called nodes):
%3 test_db__db_foreach db__db_foreach (test acs-tcl) db_foreach db_foreach test_db__db_foreach->db_foreach test_db__transaction_bug_3440 db__transaction_bug_3440 (test acs-tcl) test_db__transaction_bug_3440->db_foreach ad_arg_parser ad_arg_parser (public) db_foreach->ad_arg_parser aa_test::get_test_doc aa_test::get_test_doc (private) aa_test::get_test_doc->db_foreach acs_mail_lite::send_immediately acs_mail_lite::send_immediately (private) acs_mail_lite::send_immediately->db_foreach acs_mail_lite::sweeper acs_mail_lite::sweeper (private) acs_mail_lite::sweeper->db_foreach acs_messaging_process_queue acs_messaging_process_queue (private) acs_messaging_process_queue->db_foreach acs_object_type_hierarchy acs_object_type_hierarchy (public) acs_object_type_hierarchy->db_foreach

Testcases:
db__db_foreach, db__transaction_bug_3440

db_get_database (public)

 db_get_database [ -dbn dbn ]

PostgreSQL and NSDB only. Return the database name from the first database pool. It assumes the datasource is properly formatted since we've already verified that we can connect to the pool. On the longer range, it might be better to use SQL queries, at least in cases, where database is already connected. PostgreSQL: SELECT current_database() Oracle: SELECT name from v$database; SELECT ora_database_name FROM dual

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Returns:
database name

Partial Call Graph (max 5 caller/called nodes):
%3 db_load_sql_data db_load_sql_data (public) db_get_database db_get_database db_load_sql_data->db_get_database db_source_sql_file db_source_sql_file (public) db_source_sql_file->db_get_database packages/acs-subsite/www/admin/system/db-locks.tcl packages/acs-subsite/ www/admin/system/db-locks.tcl packages/acs-subsite/www/admin/system/db-locks.tcl->db_get_database db_available_pools db_available_pools (public) db_get_database->db_available_pools

Testcases:
No testcase defined.

db_get_dbhost (public)

 db_get_dbhost [ -dbn dbn ]

PostgreSQL only.

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Returns:
the name of the database host from the first database pool. It assumes the datasource is properly formatted since we've already verified that we can connect to the pool.

Partial Call Graph (max 5 caller/called nodes):
%3 db_load_sql_data db_load_sql_data (public) db_get_dbhost db_get_dbhost db_load_sql_data->db_get_dbhost db_source_sql_file db_source_sql_file (public) db_source_sql_file->db_get_dbhost db_available_pools db_available_pools (public) db_get_dbhost->db_available_pools

Testcases:
No testcase defined.

db_get_password (public)

 db_get_password [ -dbn dbn ]
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Returns:
the password parameter from the driver section of the first database pool for the dbn.

Partial Call Graph (max 5 caller/called nodes):
%3 db_load_sql_data db_load_sql_data (public) db_get_password db_get_password db_load_sql_data->db_get_password db_source_sql_file db_source_sql_file (public) db_source_sql_file->db_get_password db_available_pools db_available_pools (public) db_get_password->db_available_pools

Testcases:
No testcase defined.

db_get_pgbin (public)

 db_get_pgbin [ -dbn dbn ]

PostgreSQL only.

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Returns:
the pgbin parameter from the driver section of the first database pool.

Partial Call Graph (max 5 caller/called nodes):
%3 test_acs_tcl_exec_optional_dependencies acs_tcl_exec_optional_dependencies (test acs-tcl) db_get_pgbin db_get_pgbin test_acs_tcl_exec_optional_dependencies->db_get_pgbin test_acs_tcl_exec_required_dependencies acs_tcl_exec_required_dependencies (test acs-tcl) test_acs_tcl_exec_required_dependencies->db_get_pgbin db_available_pools db_available_pools (public) db_get_pgbin->db_available_pools util::which util::which (public) db_get_pgbin->util::which _acs_tcl__acs_tcl_external_dependencies_helper _acs_tcl__acs_tcl_external_dependencies_helper (private) _acs_tcl__acs_tcl_external_dependencies_helper->db_get_pgbin db_load_sql_data db_load_sql_data (public) db_load_sql_data->db_get_pgbin db_source_sql_file db_source_sql_file (public) db_source_sql_file->db_get_pgbin

Testcases:
acs_tcl_exec_required_dependencies, acs_tcl_exec_optional_dependencies

db_get_port (public)

 db_get_port [ -dbn dbn ]

PostgreSQL only.

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Returns:
the port number from the first database pool. It assumes the datasource is properly formatted since we've already verified that we can connect to the pool. It returns an empty string for an empty port value.

Partial Call Graph (max 5 caller/called nodes):
%3 db_load_sql_data db_load_sql_data (public) db_get_port db_get_port db_load_sql_data->db_get_port db_source_sql_file db_source_sql_file (public) db_source_sql_file->db_get_port db_available_pools db_available_pools (public) db_get_port->db_available_pools

Testcases:
No testcase defined.

db_get_quote_indices (private)

 db_get_quote_indices sql

Given a piece of SQL, return the indices of single quotes. This is useful when we do bind var substitution because we should not attempt bind var substitution inside quotes. Examples:

    sql          return value
    {'a'}           {0 2}
    {'a''}           {}
    {'a'a'a'}       {0 2 4 6}
    {a'b'c'd'}      {1 3 5 7}
    

Parameters:
sql (required)
See Also:

Partial Call Graph (max 5 caller/called nodes):
%3 test_db_get_quote_indices db_get_quote_indices (test acs-tcl) db_get_quote_indices db_get_quote_indices test_db_get_quote_indices->db_get_quote_indices db_bind_var_quoted_p db_bind_var_quoted_p (private) db_bind_var_quoted_p->db_get_quote_indices

Testcases:
db_get_quote_indices

db_get_sql_user (public)

 db_get_sql_user [ -dbn dbn ]

Oracle only.

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Returns:
a valid Oracle user@database/password string to access a database through sqlplus.

This proc may well work for databases other than Oracle, but its return value won't really be of any use.

Partial Call Graph (max 5 caller/called nodes):
%3 db_load_sql_data db_load_sql_data (public) db_get_sql_user db_get_sql_user db_load_sql_data->db_get_sql_user db_source_sql_file db_source_sql_file (public) db_source_sql_file->db_get_sql_user db_source_sqlj_file db_source_sqlj_file (public) db_source_sqlj_file->db_get_sql_user db_available_pools db_available_pools (public) db_get_sql_user->db_available_pools

Testcases:
No testcase defined.

db_get_username (public)

 db_get_username [ -dbn dbn ]
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Returns:
the username parameter from the driver section of the first database pool for the dbn.

Partial Call Graph (max 5 caller/called nodes):
%3 db_load_sql_data db_load_sql_data (public) db_get_username db_get_username db_load_sql_data->db_get_username db_source_sql_file db_source_sql_file (public) db_source_sql_file->db_get_username db_available_pools db_available_pools (public) db_get_username->db_available_pools

Testcases:
No testcase defined.

db_getrow (private)

 db_getrow db selection

A helper procedure to perform an ns_db getrow, invoking developer support routines as necessary.

Parameters:
db (required)
selection (required)

Partial Call Graph (max 5 caller/called nodes):
%3 test_xowiki_test_cases xowiki_test_cases (test xowiki) db_getrow db_getrow test_xowiki_test_cases->db_getrow ds_collect_db_call ds_collect_db_call (public) db_getrow->ds_collect_db_call db_list db_list (public) db_list->db_getrow db_list_of_lists db_list_of_lists (public) db_list_of_lists->db_getrow db_list_of_ns_sets db_list_of_ns_sets (public) db_list_of_ns_sets->db_getrow xo::db::DB-postgresql instproc list xo::db::DB-postgresql instproc list xo::db::DB-postgresql instproc list->db_getrow xo::db::DB-postgresql instproc list_of_lists xo::db::DB-postgresql instproc list_of_lists xo::db::DB-postgresql instproc list_of_lists->db_getrow

Testcases:
xowiki_test_cases

db_known_database_types (public)

 db_known_database_types
Returns:
a list of three-element lists describing the database engines known to OpenACS. Each sublist contains the internal database name (used in file paths, etc), the driver name, and a "pretty name" to be used in selection forms displayed to the user. The nsv containing the list is initialized by the bootstrap script and should never be referenced directly by user code.

Partial Call Graph (max 5 caller/called nodes):
%3 apm_package_supported_databases apm_package_supported_databases (public) db_known_database_types db_known_database_types apm_package_supported_databases->db_known_database_types

Testcases:
No testcase defined.

db_last_used_handle (private)

 db_last_used_handle [ -dbn dbn ]

Get the last used inactive handle.

Switches:
-dbn (optional)
database name
Returns:
last active handle or empty string

Partial Call Graph (max 5 caller/called nodes):
%3 db_resultrows db_resultrows (public) db_last_used_handle db_last_used_handle db_resultrows->db_last_used_handle db_available_pools db_available_pools (public) db_last_used_handle->db_available_pools

Testcases:
No testcase defined.

db_legacy_package_p (private)

 db_legacy_package_p db_type_list
Parameters:
db_type_list (required)
Returns:
1 if the package is a legacy package. We can only tell for certain if it explicitly supports Oracle 8.1.6 rather than the OpenACS more general oracle.

Partial Call Graph (max 5 caller/called nodes):
%3

Testcases:
No testcase defined.

db_list (public)

 db_list [ -dbn dbn ] [ -cache_key cache_key ] \
    [ -cache_pool cache_pool ] [ -subst subst ] statement_name sql \
    [ -bind bind ]
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-cache_key (optional)
Cache the result using given value as the key. Default is to not cache.
-cache_pool (optional, defaults to "db_cache_pool")
Override the default db_cache_pool
-subst (optional, defaults to "all")
Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands
-bind (optional)
bind variables, passed either as an ns_set id, or via bind value list
Parameters:
statement_name (required)
name of the SQL query.
sql (required)
SQL query to be executed.
Returns:
a Tcl list of the values in the first column of the result of SQL query sql. If the SQL query doesn't return any rows, returns an empty list.

Partial Call Graph (max 5 caller/called nodes):
%3 test_db__caching db__caching (test acs-tcl) db_list db_list test_db__caching->db_list test_db__list_variants db__list_variants (test acs-tcl) test_db__list_variants->db_list db_exec db_exec (public) db_list->db_exec db_getrow db_getrow (private) db_list->db_getrow db_qd_get_fullname db_qd_get_fullname (public) db_list->db_qd_get_fullname db_with_handle db_with_handle (public) db_list->db_with_handle acs_mail_lite::inbound_queue_pull acs_mail_lite::inbound_queue_pull (private) acs_mail_lite::inbound_queue_pull->db_list acs_mail_lite::inbound_queue_release acs_mail_lite::inbound_queue_release (private) acs_mail_lite::inbound_queue_release->db_list acs_object_type::supertypes acs_object_type::supertypes (private) acs_object_type::supertypes->db_list acs_sc::contract::get_operations acs_sc::contract::get_operations (public) acs_sc::contract::get_operations->db_list acs_user_extension::list_extensions acs_user_extension::list_extensions (public) acs_user_extension::list_extensions->db_list

Testcases:
db__caching, db__list_variants

db_list_of_lists (public)

 db_list_of_lists [ -dbn dbn ] [ -cache_key cache_key ] \
    [ -cache_pool cache_pool ] [ -with_headers ] [ -subst subst ] \
    [ -columns_var columns_var ] statement_name sql [ -bind bind ]
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-cache_key (optional)
Cache the result using given value as the key. Default is to not cache.
-cache_pool (optional, defaults to "db_cache_pool")
Override the default db_cache_pool
-with_headers (optional, boolean)
when specified, first line of returned list of lists will always be the list of column names as reported by the database. Useful when you want to dynamically assign variables to values returned in the list of lists.
-subst (optional, defaults to "all")
Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands
-columns_var (optional)
-bind (optional)
bind variables, passed either as an ns_set id, or via bind value list
Parameters:
statement_name (required)
name of the SQL query.
sql (required)
SQL query to be executed.
Returns:
a Tcl list, each element of which is a list of all column values in a row of the result of the SQL querysql. If sql doesn't return any rows, returns an empty list, unless with_headers flag was specified and in this case the only element in the list will be the list of headers. It checks if the element is I18N and replaces it, thereby reducing the need to do this with every single package

Partial Call Graph (max 5 caller/called nodes):
%3 test_auth_authority_api auth_authority_api (test acs-authentication) db_list_of_lists db_list_of_lists test_auth_authority_api->db_list_of_lists test_db__caching db__caching (test acs-tcl) test_db__caching->db_list_of_lists test_db__list_variants db__list_variants (test acs-tcl) test_db__list_variants->db_list_of_lists test_parameter__check_procs parameter__check_procs (test acs-tcl) test_parameter__check_procs->db_list_of_lists acs::icanuse acs::icanuse (public) db_list_of_lists->acs::icanuse db_exec db_exec (public) db_list_of_lists->db_exec db_getrow db_getrow (private) db_list_of_lists->db_getrow db_qd_get_fullname db_qd_get_fullname (public) db_list_of_lists->db_qd_get_fullname db_with_handle db_with_handle (public) db_list_of_lists->db_with_handle acs_mail_lite::inbound_queue_pull_one acs_mail_lite::inbound_queue_pull_one (private) acs_mail_lite::inbound_queue_pull_one->db_list_of_lists acs_mail_lite::unique_id_parse acs_mail_lite::unique_id_parse (private) acs_mail_lite::unique_id_parse->db_list_of_lists acs_sc::impl::get_options acs_sc::impl::get_options (public) acs_sc::impl::get_options->db_list_of_lists ad_get_node_id_from_host_node_map ad_get_node_id_from_host_node_map (private) ad_get_node_id_from_host_node_map->db_list_of_lists apm::get_package_descendent_options apm::get_package_descendent_options (public) apm::get_package_descendent_options->db_list_of_lists

Testcases:
auth_authority_api, parameter__check_procs, db__caching, db__list_variants

db_list_of_ns_sets (public)

 db_list_of_ns_sets [ -dbn dbn ] [ -subst subst ] \
    [ -columns_var columns_var ] statement_name sql [ -bind bind ]
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-subst (optional, defaults to "all")
-columns_var (optional)
-bind (optional)
bind variables, passed either as an ns_set id, or via bind value list
Parameters:
statement_name (required)
name of the SQL query.
sql (required)
SQL query to be executed.
Returns:
a list of ns_sets with the values of each column of each row returned by the SQL query specified.
list of ns_sets, one per each row return by the SQL query

Partial Call Graph (max 5 caller/called nodes):
%3 test_db__caching db__caching (test acs-tcl) db_list_of_ns_sets db_list_of_ns_sets test_db__caching->db_list_of_ns_sets test_db__db_foreach db__db_foreach (test acs-tcl) test_db__db_foreach->db_list_of_ns_sets test_db__list_variants db__list_variants (test acs-tcl) test_db__list_variants->db_list_of_ns_sets test_db__transaction_bug_3440 db__transaction_bug_3440 (test acs-tcl) test_db__transaction_bug_3440->db_list_of_ns_sets acs::icanuse acs::icanuse (public) db_list_of_ns_sets->acs::icanuse db_exec db_exec (public) db_list_of_ns_sets->db_exec db_getrow db_getrow (private) db_list_of_ns_sets->db_getrow db_qd_get_fullname db_qd_get_fullname (public) db_list_of_ns_sets->db_qd_get_fullname db_with_handle db_with_handle (public) db_list_of_ns_sets->db_with_handle acs_mail_lite::check_bounces acs_mail_lite::check_bounces (private) acs_mail_lite::check_bounces->db_list_of_ns_sets forum::list_forums forum::list_forums (public) forum::list_forums->db_list_of_ns_sets fs::get_folder_contents fs::get_folder_contents (public, deprecated) fs::get_folder_contents->db_list_of_ns_sets notification::sweep::sweep_notifications notification::sweep::sweep_notifications (private) notification::sweep::sweep_notifications->db_list_of_ns_sets

Testcases:
db__db_foreach, db__caching, db__transaction_bug_3440, db__list_variants

db_load_sql_data (public)

 db_load_sql_data [ -dbn dbn ] [ -callback callback ] file

Loads a CSV formatted file into a table using PostgreSQL's COPY command or Oracle's SQL*Loader utility. The filename format consists of a sequence number used to control the order in which tables are loaded, and the table name with "-" replacing "_". This is a bit of a kludge but greatly speeds the loading of large amounts of data, such as is done when various "ref-*" packages are installed.

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-callback (optional, defaults to "apm_ns_write_callback")
Parameters:
file (required)
Filename in the format dd-table-name.ctl where 'dd' is a sequence number used to control the order in which data is loaded. This file is an RDBMS-specific data loader control file.

Partial Call Graph (max 5 caller/called nodes):
%3 apm_package_install_data_model apm_package_install_data_model (private) db_load_sql_data db_load_sql_data apm_package_install_data_model->db_load_sql_data ref_timezones::apm::after_upgrade ref_timezones::apm::after_upgrade (private) ref_timezones::apm::after_upgrade->db_load_sql_data ad_file ad_file (public) db_load_sql_data->ad_file ad_tmpdir ad_tmpdir (public) db_load_sql_data->ad_tmpdir apm_callback_and_log apm_callback_and_log (public) db_load_sql_data->apm_callback_and_log db_driverkey db_driverkey (public) db_load_sql_data->db_driverkey db_get_database db_get_database (public) db_load_sql_data->db_get_database

Testcases:
No testcase defined.

db_multirow (public)

 db_multirow [ -local ] [ -append ] [ -upvar_level upvar_level ] \
    [ -unclobber ] [ -extend extend ] [ -dbn dbn ] \
    [ -cache_key cache_key ] [ -cache_pool cache_pool ] \
    [ -subst subst ] var_name statement_name sql [ args... ]

Performs the SQL query sql, saving results in variables of the form var_name:1, var_name:2, etc, setting var_name:rowcount to the total number of rows, and setting var_name:columns to a list of column names. Usage:

db_multirow [ -local ] [ -upvar_level n_levels_up ] [ -append ] [ -extend column_list ] var-name statement-name sql [ -bind bind_set_id | -bind bind_value_list ] code_block [ if_no_rows if_no_rows_block ]

If "cache_key" is set, cache the array that results from the query *and* any code block for future use. When this result is returned from cache, THE CODE BLOCK IS NOT EXECUTED. Therefore, any values calculated by the code block that aren't listed as arguments to "extend" will not be created. In practice this impacts relatively few queries, but do take care.

You can not simultaneously append to and cache a nonempty multirow.

Each row also has a column, rownum, automatically added and set to the row number, starting with 1. Note that this will override any column in the SQL statement named 'rownum', also if you're using the Oracle rownum pseudo-column.

If the -local is passed, the variables defined by db_multirow will be set locally (useful if you're compiling dynamic templates in a function or similar situations). Use the -upvar_level switch to specify how many levels up the variable should be set. The default behavior (i.e., when no "-local" is specified) depends on the calling environment: when "db_multirow" is called from an ADP file the variables are set in the ADP environment. Otherwise, the default behavior is "-local".

You may supply a code block, which will be executed for each row in the loop. This is very useful if you need to make computations that are better done in Tcl than in SQL, for example using ns_urlencode or ns_quotehtml, etc. When the Tcl code is executed, all the columns from the SQL query will be set as local variables in that code. Any changes made to these local variables will be copied back into the multirow.

You may also add additional, computed columns to the multirow, using the -extend { col_1 col_2 ... } switch. This is useful for things like constructing a URL for the object retrieved by the query.

If you're constructing your multirow through multiple queries with the same set of columns, but with different rows, you can use the -append switch. This causes the rows returned by this query to be appended to the rows already in the multirow, instead of starting a clean multirow, as is the normal behavior. The columns must match the columns in the original multirow, or an error will be thrown.

Your code block may call continue in order to skip a row and not include it in the multirow. Or you can call break to skip this row and quit looping.

Notice the nonstandard numbering (everything else in Tcl starts at 0); the reason is that the graphics designer, a non-programmer, may wish to work with row numbers.

Example:

db_multirow -extend { user_url } users users_query {
        select user_id first_names, last_name, email from cc_users
    } {
        set user_url [acs_community_member_url -user_id $user_id]
    }

Switches:
-local (optional, boolean)
-append (optional, boolean)
-upvar_level (optional, defaults to "1")
-unclobber (optional, boolean)
If set, will cause the proc to not overwrite local variables. Actually, what happens is that the local variables will be overwritten, so you can access them within the code block. However, if you specify -unclobber, we will revert them to their original state after execution of this proc.
-extend (optional)
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-cache_key (optional)
Cache the result using given value as the key. Default is to not cache.
-cache_pool (optional, defaults to "db_cache_pool")
Override the default db_cache_pool
-subst (optional, defaults to "all")
Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands
Parameters:
var_name (required)
name of the Tcl multirow array
statement_name (required)
name of the SQL query
sql (required)
SQL query to be executed
See Also:

Partial Call Graph (max 5 caller/called nodes):
%3 test_db__caching db__caching (test acs-tcl) db_multirow db_multirow test_db__caching->db_multirow test_db__transaction_bug_3440 db__transaction_bug_3440 (test acs-tcl) test_db__transaction_bug_3440->db_multirow ad_arg_parser ad_arg_parser (public) db_multirow->ad_arg_parser db_multirow_helper db_multirow_helper (private) db_multirow->db_multirow_helper db_qd_get_fullname db_qd_get_fullname (public) db_multirow->db_qd_get_fullname template::adp_level template::adp_level (public) db_multirow->template::adp_level acs_user::demote_user acs_user::demote_user (public) acs_user::demote_user->db_multirow bm_folder_selection bm_folder_selection (public) bm_folder_selection->db_multirow bug_tracker::bug::get_multirow bug_tracker::bug::get_multirow (public) bug_tracker::bug::get_multirow->db_multirow category::get_mapped_categories_multirow category::get_mapped_categories_multirow (public) category::get_mapped_categories_multirow->db_multirow doc::func_multirow doc::func_multirow (private, deprecated) doc::func_multirow->db_multirow

Testcases:
db__caching, db__transaction_bug_3440

db_multirow_group_last_row_p (public)

 db_multirow_group_last_row_p -column column

Used inside the code_block to db_multirow to ask whether this row is the last row before the value of 'column' changes, or the last row of the result set.

This is useful when you want to build up a multirow for a master/slave table pair, where you only want one row per row in the master table, but you want to include data from the slave table in a column of the multirow.

Here's an example:

    # Initialize the lines variable to hold a list of order line summaries
    set lines [list]

    # Start building the multirow. We add the dynamic column 'lines_pretty', which will
    # contain the pretty summary of the order lines.
    db_multirow -extend { lines_pretty } orders select_orders_and_lines {
        select o.order_id,
        o.customer_name,
        l.item_name,
        l.quantity
        from   orders o,
        order_lines l
        where  l.order_id = o.order_id
        order  by o.order_id, l.item_name
    } {
        lappend lines "$quantity $item_name"
        if { [db_multirow_group_last_row_p -column order_id] } {
            # Last row of this order, prepare the pretty version of the order lines
            set lines_pretty [join $lines ", "]

            # Reset the lines list, so we start from a fresh with the next row
            set lines [list]
        } else {
            # There are yet more order lines to come for this order,
            # continue until we've collected all the order lines
            # The 'continue' keyword means this line will not be added to the resulting multirow
            continue
        }
    }
    

Switches:
-column (required)
The name of the column defining the groups.
Returns:
1 if this is the last row before the column value changes, 0 otherwise.
Author:
Lars Pind <lars@collaboraid.biz>

Partial Call Graph (max 5 caller/called nodes):
%3 bug_tracker::bug::get_multirow bug_tracker::bug::get_multirow (public) db_multirow_group_last_row_p db_multirow_group_last_row_p bug_tracker::bug::get_multirow->db_multirow_group_last_row_p

Testcases:
No testcase defined.

db_multirow_helper (private)

 db_multirow_helper

Helper function for db_multirow, performing the actual DB queries.

Partial Call Graph (max 5 caller/called nodes):
%3 db_multirow db_multirow (public) db_multirow_helper db_multirow_helper db_multirow->db_multirow_helper

Testcases:
No testcase defined.

db_name (public)

 db_name [ -dbn dbn ]
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Returns:
the name of the database as reported by the driver.

Partial Call Graph (max 5 caller/called nodes):
%3 test_acs_object_procs_test acs_object_procs_test (test acs-tcl) db_name db_name test_acs_object_procs_test->db_name test_acs_tcl_exec_optional_dependencies acs_tcl_exec_optional_dependencies (test acs-tcl) test_acs_tcl_exec_optional_dependencies->db_name test_acs_tcl_exec_required_dependencies acs_tcl_exec_required_dependencies (test acs-tcl) test_acs_tcl_exec_required_dependencies->db_name test_cr_item_search_triggers cr_item_search_triggers (test acs-content-repository) test_cr_item_search_triggers->db_name test_datamodel__named_constraints datamodel__named_constraints (test acs-tcl) test_datamodel__named_constraints->db_name db_with_handle db_with_handle (public) db_name->db_with_handle _acs_tcl__acs_tcl_external_dependencies_helper _acs_tcl__acs_tcl_external_dependencies_helper (private) _acs_tcl__acs_tcl_external_dependencies_helper->db_name

Testcases:
cr_item_search_triggers, acs_tcl_exec_required_dependencies, acs_tcl_exec_optional_dependencies, datamodel__named_constraints, acs_object_procs_test

db_nextval (public)

 db_nextval [ -dbn dbn ] sequence

Example:

    set new_object_id [db_nextval acs_object_id_seq]
    

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Parameters:
sequence (required)
the name of an SQL sequence
Returns:
the next value for a sequence. This can utilize a pool of sequence values.
See Also:

Partial Call Graph (max 5 caller/called nodes):
%3 test_acs_object_procs_test acs_object_procs_test (test acs-tcl) db_nextval db_nextval test_acs_object_procs_test->db_nextval test_ad_proc_permission_grant_and_revoke ad_proc_permission_grant_and_revoke (test acs-tcl) test_ad_proc_permission_grant_and_revoke->db_nextval test_ad_proc_permission_permission_p ad_proc_permission_permission_p (test acs-tcl) test_ad_proc_permission_permission_p->db_nextval ad_log ad_log (public) db_nextval->ad_log db_0or1row db_0or1row (public) db_nextval->db_0or1row db_driverkey db_driverkey (public) db_nextval->db_driverkey db_list db_list (public) db_nextval->db_list db_string db_string (public) db_nextval->db_string acs_mail_lite::inbound_cache_hit_p acs_mail_lite::inbound_cache_hit_p (private) acs_mail_lite::inbound_cache_hit_p->db_nextval acs_mail_lite::inbound_queue_insert acs_mail_lite::inbound_queue_insert (private) acs_mail_lite::inbound_queue_insert->db_nextval acs_mail_lite::section_id_of acs_mail_lite::section_id_of (private) acs_mail_lite::section_id_of->db_nextval ad_form ad_form (public) ad_form->db_nextval application_data_link::new_from application_data_link::new_from (public) application_data_link::new_from->db_nextval

Testcases:
acs_object_procs_test, ad_proc_permission_grant_and_revoke, ad_proc_permission_permission_p

db_nth_pool_name (public)

 db_nth_pool_name [ -dbn dbn ] n
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Parameters:
n (required)
Returns:
the name of the pool used for the nth-nested selection (0-relative).

Partial Call Graph (max 5 caller/called nodes):
%3 db_available_pools db_available_pools (public) db_nth_pool_name db_nth_pool_name db_nth_pool_name->db_available_pools

Testcases:
No testcase defined.

db_null (public, deprecated)

 db_null
Deprecated. Invoking this procedure generates a warning.

Returns:
an empty string, which Oracle thinks is null. Deprecated: This routine was invented to provide an RDBMS-specific null value but doesn't actually work. I (DRB) left it in to speed porting - we should really clean up the code and pull out the calls instead, though.
See Also:
  • ""

Partial Call Graph (max 5 caller/called nodes):
%3 ad_log_deprecated ad_log_deprecated (public) db_null db_null db_null->ad_log_deprecated

Testcases:
No testcase defined.

db_nullify_empty_string (public, deprecated)

 db_nullify_empty_string string
Deprecated. Invoking this procedure generates a warning.

A convenience function that returns [db_null] if $string is the empty string. Deprecated: essentially just returns the passed string.

Parameters:
string (required)
See Also:
  • db_null

Partial Call Graph (max 5 caller/called nodes):
%3 ad_log_deprecated ad_log_deprecated (public) db_nullify_empty_string db_nullify_empty_string db_nullify_empty_string->ad_log_deprecated

Testcases:
No testcase defined.

db_quote (public, deprecated)

 db_quote string
Deprecated. Invoking this procedure generates a warning.

Quotes a string value to be placed in a SQL statement. Use the built-in ns_dbquotevalue instead, which cares also about the surrounding quotes.

Parameters:
string (required)
See Also:

Partial Call Graph (max 5 caller/called nodes):
%3 ad_log_deprecated ad_log_deprecated (public) db_quote db_quote db_quote->ad_log_deprecated

Testcases:
No testcase defined.

db_release_unused_handles (public)

 db_release_unused_handles [ -dbn dbn ]

Releases any database handles that are presently unused.

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.

Partial Call Graph (max 5 caller/called nodes):
%3 test_db__caching db__caching (test acs-tcl) db_release_unused_handles db_release_unused_handles test_db__caching->db_release_unused_handles test_db__db_foreach db__db_foreach (test acs-tcl) test_db__db_foreach->db_release_unused_handles test_db__transaction_bug_3440 db__transaction_bug_3440 (test acs-tcl) test_db__transaction_bug_3440->db_release_unused_handles db_available_pools db_available_pools (public) db_release_unused_handles->db_available_pools db_state_array_name_is db_state_array_name_is (private) db_release_unused_handles->db_state_array_name_is ds_collect_db_call ds_collect_db_call (public) db_release_unused_handles->ds_collect_db_call ad_progress_bar_begin ad_progress_bar_begin (public) ad_progress_bar_begin->db_release_unused_handles adp_parse_ad_conn_file adp_parse_ad_conn_file (private) adp_parse_ad_conn_file->db_release_unused_handles apm_bootstrap_load_libraries apm_bootstrap_load_libraries (private) apm_bootstrap_load_libraries->db_release_unused_handles apm_files_load apm_files_load (private) apm_files_load->db_release_unused_handles apm_load_libraries apm_load_libraries (private) apm_load_libraries->db_release_unused_handles

Testcases:
db__db_foreach, db__caching, db__transaction_bug_3440

db_resultrows (public)

 db_resultrows [ -dbn dbn ]
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Returns:
the number of rows affected by the last DML command.

Partial Call Graph (max 5 caller/called nodes):
%3 test_xowiki_test_cases xowiki_test_cases (test xowiki) db_resultrows db_resultrows test_xowiki_test_cases->db_resultrows db_driverkey db_driverkey (public) db_resultrows->db_driverkey db_last_used_handle db_last_used_handle (private) db_resultrows->db_last_used_handle acs_mail_lite::bounce_ministry acs_mail_lite::bounce_ministry (private) acs_mail_lite::bounce_ministry->db_resultrows acs_mail_lite::log_mail_sending acs_mail_lite::log_mail_sending (private) acs_mail_lite::log_mail_sending->db_resultrows acs_mail_lite::record_bounce acs_mail_lite::record_bounce (public, deprecated) acs_mail_lite::record_bounce->db_resultrows acs_mail_lite::sweeper acs_mail_lite::sweeper (private) acs_mail_lite::sweeper->db_resultrows attribute::value_delete attribute::value_delete (public) attribute::value_delete->db_resultrows

Testcases:
xowiki_test_cases

db_source_sql_file (public)

 db_source_sql_file [ -dbn dbn ] [ -callback callback ] file

Sources a SQL file into Oracle (SQL*Plus format file) or PostgreSQL (psql format file).

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-callback (optional, defaults to "apm_ns_write_callback")
Parameters:
file (required)

Partial Call Graph (max 5 caller/called nodes):
%3 ad_acs_require_basic_schemata ad_acs_require_basic_schemata (private) db_source_sql_file db_source_sql_file ad_acs_require_basic_schemata->db_source_sql_file apm_package_delete apm_package_delete (public) apm_package_delete->db_source_sql_file apm_package_install_data_model apm_package_install_data_model (private) apm_package_install_data_model->db_source_sql_file install::xml::action::source install::xml::action::source (private) install::xml::action::source->db_source_sql_file tsearch2_driver::install::preinstall_checks tsearch2_driver::install::preinstall_checks (private) tsearch2_driver::install::preinstall_checks->db_source_sql_file ad_file ad_file (public) db_source_sql_file->ad_file apm_callback_and_log apm_callback_and_log (public) db_source_sql_file->apm_callback_and_log db_driverkey db_driverkey (public) db_source_sql_file->db_driverkey db_get_database db_get_database (public) db_source_sql_file->db_get_database db_get_dbhost db_get_dbhost (public) db_source_sql_file->db_get_dbhost

Testcases:
No testcase defined.

db_source_sqlj_file (public)

 db_source_sqlj_file [ -dbn dbn ] [ -callback callback ] file

Oracle only.

Sources a SQLJ file using loadjava.

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-callback (optional, defaults to "apm_ns_write_callback")
Parameters:
file (required)

Partial Call Graph (max 5 caller/called nodes):
%3 apm_package_install_data_model apm_package_install_data_model (private) db_source_sqlj_file db_source_sqlj_file apm_package_install_data_model->db_source_sqlj_file ad_file ad_file (public) db_source_sqlj_file->ad_file apm_callback_and_log apm_callback_and_log (public) db_source_sqlj_file->apm_callback_and_log db_get_sql_user db_get_sql_user (public) db_source_sqlj_file->db_get_sql_user

Testcases:
No testcase defined.

db_state_array_name_is (private)

 db_state_array_name_is [ -dbn dbn ]
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Returns:
the name of the global db_state array for the given database name.
Author:
Andrew Piskorski <atp@piskorski.com>
Created:
2003/03/16

Partial Call Graph (max 5 caller/called nodes):
%3 db_abort_transaction db_abort_transaction (public) db_state_array_name_is db_state_array_name_is db_abort_transaction->db_state_array_name_is db_abort_transaction_p db_abort_transaction_p (private) db_abort_transaction_p->db_state_array_name_is db_release_unused_handles db_release_unused_handles (public) db_release_unused_handles->db_state_array_name_is db_transaction db_transaction (public) db_transaction->db_state_array_name_is

Testcases:
No testcase defined.

db_string (public)

 db_string [ -dbn dbn ] [ -cache_key cache_key ] \
    [ -cache_pool cache_pool ] [ -subst subst ] statement_name sql \
    [ -default default ] [ -bind bind ]
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
-cache_key (optional)
Cache the result using given value as the key. Default is to not cache.
-cache_pool (optional, defaults to "db_cache_pool")
Override the default db_cache_pool
-subst (optional, defaults to "all")
Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands
-default (optional)
Return value in case the SQL query returns no value
-bind (optional)
bind variables, passed either as an ns_set id, or via bind value list
Parameters:
statement_name (required)
name of the SQL query
sql (required)
SQL query to be executed
Returns:
the first column of the result of the SQL query sql. If the query doesn't return a row, returns default or raises an error if no default is provided.

Partial Call Graph (max 5 caller/called nodes):
%3 test_db__caching db__caching (test acs-tcl) db_string db_string test_db__caching->db_string test_db__string db__string (test acs-tcl) test_db__string->db_string test_db__transaction db__transaction (test acs-tcl) test_db__transaction->db_string test_db__transaction_bug_3440 db__transaction_bug_3440 (test acs-tcl) test_db__transaction_bug_3440->db_string test_nullchar nullchar (test acs-tcl) test_nullchar->db_string db_exec db_exec (public) db_string->db_exec db_qd_get_fullname db_qd_get_fullname (public) db_string->db_qd_get_fullname db_with_handle db_with_handle (public) db_string->db_with_handle acs_lookup_magic_object_no_cache acs_lookup_magic_object_no_cache (private) acs_lookup_magic_object_no_cache->db_string acs_object::object_p acs_object::object_p (public) acs_object::object_p->db_string acs_object::package_id_not_cached acs_object::package_id_not_cached (private) acs_object::package_id_not_cached->db_string acs_object_name acs_object_name (public) acs_object_name->db_string acs_object_type acs_object_type (public) acs_object_type->db_string

Testcases:
db__caching, db__transaction, db__transaction_bug_3440, nullchar, db__string

db_table_exists (public)

 db_table_exists [ -dbn dbn ] table_name
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Parameters:
table_name (required)
Returns:
1 if a table with the specified name exists in the database, otherwise 0.
Authors:
Don Baccus <dhogaza@pacifier.com>
Lars Pind <lars@pinds.com>

Partial Call Graph (max 5 caller/called nodes):
%3 test_datamodel__acs_object_type_check datamodel__acs_object_type_check (test acs-tcl) db_table_exists db_table_exists test_datamodel__acs_object_type_check->db_table_exists test_object_type_table_name object_type_table_name (test acs-tcl) test_object_type_table_name->db_table_exists db_driverkey db_driverkey (public) db_table_exists->db_driverkey db_string db_string (public) db_table_exists->db_string acs::db::postgresql instproc get_all_package_functions acs::db::postgresql instproc get_all_package_functions (public) acs::db::postgresql instproc get_all_package_functions->db_table_exists ad_verify_install ad_verify_install (public) ad_verify_install->db_table_exists fs::dav::require fs::dav::require (private) fs::dav::require->db_table_exists group_type::delete group_type::delete (public) group_type::delete->db_table_exists group_type::new group_type::new (public) group_type::new->db_table_exists

Testcases:
datamodel__acs_object_type_check, object_type_table_name

db_tables (public)

 db_tables [ -pattern pattern ] [ -dbn dbn ]
Switches:
-pattern (optional)
Will be used as LIKE 'pattern%' to limit the number of tables returned.
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Returns:
a Tcl list of all the tables owned by the connected user.
Authors:
Don Baccus <dhogaza@pacifier.com>
Lars Pind <lars@pinds.com>
Changelog:
yon@arsdigita.com 20000711 changed to return lowercase table names

Partial Call Graph (max 5 caller/called nodes):
%3 db_driverkey db_driverkey (public) db_foreach db_foreach (public) db_tables db_tables db_tables->db_driverkey db_tables->db_foreach

Testcases:
No testcase defined.

db_transaction (public)

 db_transaction [ -dbn dbn ] transaction_code [ args... ]

Usage: db_transaction transaction_code [ on_error { error_code_block } ] Executes transaction_code with transactional semantics. This means that either all of the database commands within transaction_code are committed to the database or none of them are. Multiple db_transactions may be nested (end transaction is transparently ns_db dml'ed when the outermost transaction completes).

To handle errors, use db_transaction {transaction_code} on_error {error_code_block}. Any error generated in transaction_code will be caught automatically and process control will transfer to error_code_block with a variable errmsg set. The error_code block can then clean up after the error, such as presenting a usable error message to the user. Following the execution of error_code_block the transaction will be aborted. If you want to explicitly abort the transaction, call db_abort_transaction from within the transaction_code block or the error_code block.

Example 1:
In this example, db_dml triggers an error, so control passes to the on_error block which prints a readable error.

    db_transaction {
        db_dml test "nonsense"
    } on_error {
        ad_return_error "Error in blah/foo/bar" "The error was: $errmsg"
    }
    
Example 2:
In this example, the second command, "nonsense" triggers an error. There is no on_error block, so the transaction is immediately halted and aborted.
    db_transaction {
        db_dml test {insert into footest values(1)}
        nonsense
        db_dml test {insert into footest values(2)}
    }
    

Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Parameters:
transaction_code (required)

Partial Call Graph (max 5 caller/called nodes):
%3 test_db__transaction db__transaction (test acs-tcl) db_transaction db_transaction test_db__transaction->db_transaction test_db__transaction_bug_3440 db__transaction_bug_3440 (test acs-tcl) test_db__transaction_bug_3440->db_transaction db_abort_transaction db_abort_transaction (public) db_transaction->db_abort_transaction db_abort_transaction_p db_abort_transaction_p (private) db_transaction->db_abort_transaction_p db_release_unused_handles db_release_unused_handles (public) db_transaction->db_release_unused_handles db_state_array_name_is db_state_array_name_is (private) db_transaction->db_state_array_name_is db_type db_type (public) db_transaction->db_type Class ::xo::Authorize Class ::xo::Authorize (public) Class ::xo::Authorize->db_transaction aa_run_with_teardown aa_run_with_teardown (public) aa_run_with_teardown->db_transaction acs::test::auth::install acs::test::auth::install (private) acs::test::auth::install->db_transaction acs_mail_lite::after_upgrade acs_mail_lite::after_upgrade (private) acs_mail_lite::after_upgrade->db_transaction acs_mail_lite::imap_conn_set acs_mail_lite::imap_conn_set (private) acs_mail_lite::imap_conn_set->db_transaction

Testcases:
db__transaction, db__transaction_bug_3440

db_type (public)

 db_type
Returns:
the RDBMS type (i.e. oracle, postgresql) this OpenACS installation is using. The nsv ad_database_type is set up during the bootstrap process.

Partial Call Graph (max 5 caller/called nodes):
%3 test_db_bind_var_substitution db_bind_var_substitution (test acs-tcl) db_type db_type test_db_bind_var_substitution->db_type test_nullchar nullchar (test acs-tcl) test_nullchar->db_type test_sql_date sql_date (test acs-templating) test_sql_date->db_type apm_file_watchable_p apm_file_watchable_p (public) apm_file_watchable_p->db_type apm_load_queries apm_load_queries (private) apm_load_queries->db_type apm_package_supports_rdbms_p apm_package_supports_rdbms_p (public) apm_package_supports_rdbms_p->db_type apm_query_files_find apm_query_files_find (private) apm_query_files_find->db_type content::type::attribute::new content::type::attribute::new (public) content::type::attribute::new->db_type

Testcases:
db_bind_var_substitution, nullchar, sql_date

db_version (public)

 db_version
Returns:
the RDBMS version (i.e. 8.1.6 is a recent Oracle version; 7.1 a recent PostgreSQL version)

Partial Call Graph (max 5 caller/called nodes):
%3 test_sql_date sql_date (test acs-templating) db_version db_version test_sql_date->db_version db_current_rdbms db_current_rdbms (public) db_current_rdbms->db_version packages/acs-api-browser/lib/search.tcl packages/acs-api-browser/ lib/search.tcl packages/acs-api-browser/lib/search.tcl->db_version template::util::date::get_property template::util::date::get_property (public) template::util::date::get_property->db_version tsearch2::build_query tsearch2::build_query (private) tsearch2::build_query->db_version

Testcases:
sql_date

db_with_handle (public)

 db_with_handle [ -dbn dbn ] db code_block

Place a usable database handle in db and executes code_block.

Switches:
-dbn (optional)
Database name to use. If empty_string, use the default database
Parameters:
db (required)
Name of the handle variable used in the code block
code_block (required)
code block to be executed with handle

Partial Call Graph (max 5 caller/called nodes):
%3 test_db__0or1row db__0or1row (test acs-tcl) db_with_handle db_with_handle test_db__0or1row->db_with_handle test_db__1row db__1row (test acs-tcl) test_db__1row->db_with_handle test_xowiki_test_cases xowiki_test_cases (test xowiki) test_xowiki_test_cases->db_with_handle ad_log ad_log (public) db_with_handle->ad_log db_available_pools db_available_pools (public) db_with_handle->db_available_pools ds_collect_db_call ds_collect_db_call (public) db_with_handle->ds_collect_db_call acs::db::nsdb-postgresql instproc {call acs add_user} acs::db::nsdb-postgresql instproc {call acs add_user} (public) acs::db::nsdb-postgresql instproc {call acs add_user}->db_with_handle acs::db::nsdb-postgresql instproc {call acs magic_object_id} acs::db::nsdb-postgresql instproc {call acs magic_object_id} (public) acs::db::nsdb-postgresql instproc {call acs magic_object_id}->db_with_handle acs::db::nsdb-postgresql instproc {call acs remove_user} acs::db::nsdb-postgresql instproc {call acs remove_user} (public) acs::db::nsdb-postgresql instproc {call acs remove_user}->db_with_handle acs::db::nsdb-postgresql instproc {call acs_activity delete} acs::db::nsdb-postgresql instproc {call acs_activity delete} (public) acs::db::nsdb-postgresql instproc {call acs_activity delete}->db_with_handle acs::db::nsdb-postgresql instproc {call acs_activity edit} acs::db::nsdb-postgresql instproc {call acs_activity edit} (public) acs::db::nsdb-postgresql instproc {call acs_activity edit}->db_with_handle

Testcases:
db__0or1row, db__1row, xowiki_test_cases

db_write_blob (public)

 db_write_blob [ -dbn dbn ] statement_name sql [ args... ]
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Parameters:
statement_name (required)
sql (required)

Partial Call Graph (max 5 caller/called nodes):
%3 cr_write_content-lob cr_write_content-lob (private) db_write_blob db_write_blob cr_write_content-lob->db_write_blob ad_arg_parser ad_arg_parser (public) db_write_blob->ad_arg_parser db_exec_lob db_exec_lob (private) db_write_blob->db_exec_lob db_qd_get_fullname db_qd_get_fullname (public) db_write_blob->db_qd_get_fullname db_with_handle db_with_handle (public) db_write_blob->db_with_handle

Testcases:
No testcase defined.

db_write_clob (public)

 db_write_clob [ -dbn dbn ] statement_name sql [ args... ]
Switches:
-dbn (optional)
The database name to use. If empty_string, uses the default database.
Parameters:
statement_name (required)
sql (required)

Partial Call Graph (max 5 caller/called nodes):
%3 ad_arg_parser ad_arg_parser (public) db_driverkey db_driverkey (public) db_exec db_exec (public) db_qd_get_fullname db_qd_get_fullname (public) db_with_handle db_with_handle (public) db_write_clob db_write_clob db_write_clob->ad_arg_parser db_write_clob->db_driverkey db_write_clob->db_exec db_write_clob->db_qd_get_fullname db_write_clob->db_with_handle

Testcases:
No testcase defined.
[ hide source ] | [ make this the default ]

Content File Source

ad_library {

    An API for managing database queries.

    @creation-date 15 Apr 2000
    @author Jon Salz (jsalz@arsdigita.com)
    @cvs-id $Id: 01-database-procs.tcl,v 1.5 2024/10/21 15:49:22 gustafn Exp $
}

# Database caching.
#
# Values returned by a query are cached if you pass the "-cache_key" switch
# to the database procedure.  The switch value will be used as the key in the
# ns_cache eval call used to execute the query and processing code.  The
# db_flush proc should be called to flush the cache when appropriate.  The
# "-cache_pool" parameter can be used to specify the cache pool to be used,
# and defaults to db_cache_pool.  The # size of the default cache is governed
# by the kernel parameter "DBCacheSize" in the "caching" section.
#
# Currently db_string, db_list, db_list_of_lists, db_0or1row, and db_multirow support
# caching.
#
# Don Baccus 2/25/2006 - my 52nd birthday!

# As originally released in (at least) ACS 4.2 through OpenACS 4.6,
# this DB API supported only a single, default database.  You could
# define any number of different database drivers and pools in
# AOLserver, but could only use ONE database here.
#
# I have eliminated this restriction.  Now, in OpenACS 5.0 and later,
# to access a nondefault database, simply pass the optional -dbn
# (Database Name) switch to any of the DB API procs which support it.
#
# Supported AOLserver database drivers:
#
# - Oracle (nsoracle): Everything should work.
#
# - PostgreSQL (nspostgres): Everything should work.
#
# - ODBC (nsodbc):
#   - Anything using bind variables will only work if you're using a
#     version of the driver with bind variable emulation hacked in
#     (copied from the PostgreSQL driver).
#   - Some features, like LOBs, simply won't work at all.
#   - The basic functionality worked fine back in Sept. 2001, but I
#     have NOT tested it since then at all, so maybe there are bugs.
#
# - Any others: Basic stuff using only the standard ns_db API will
#   likely work, but any special features of the driver (e.g., LOBs)
#   definitely won't.  Feel free to add support!
#
# --atp@piskorski.com, 2003/04/09 19:18 EDT

# Note that "-dbn" specifies a "Database Name", NOT a database pool!
#
# I could have provided access to secondary databases via a -pool
# rather than a -dbn switch, but chose not to, as the existing DB API
# already had the nicely general feature that if you try to do nested
# queries, the DB API will transparently grab a second database handle
# from another pool to make it work.  You can nest your queries as
# many levels deep as you have database pools defined for that
# database.  So, the existing API essentially already supported the
# notion of "binning" database pools into logical "databases", it just
# didn't provide any way to define more than the single, default
# database!  Thus I chose to preserve this "binning" by specifying
# databases via the -dbn switch rather than database pools via a -pool
# switch.

# (JoelA, 27 Dec 2004 - replaced example config.tcl with link)
#
# see http://openacs.org/doc/openacs-5-1/tutorial-second-database
# for config and usage examples

# TODO: The "driverkey_" overrides in the config file are NOT
# implemented yet!
#
# --atp@piskorski.com, 2003/03/16 21:30 EST

# NOTE: don't forget to add your new pools into the
# ns_section ns/db/pools


# The "driverkey" indirection layer:
#
# Note that in the AOLserver config file, you may optionally add one
# entry for each database defining its "driver key".  If you do NOT
# specify a driver key in the AOLserver config file, the appropriate
# key will be determined for you by calling "ns_db driver" once on
# startup for the first pool defined in each database.  Therefore,
# most people should NOT bother to give a driverkey in the config
# file.
#
# So, just what is this "driverkey" thing used for anyway?  AOLserver
# defines the ns_db API, and the OpenACS db_* API depends utterly on
# it.  However, there are a few holes in the functionality of the
# ns_db API, and each AOLserver database driver tends to fill in those
# holes by adding extra functionality with its own, drive specific
# functions.  Therefore, in order to make the db_* API work with
# multiple db drivers, we need to introduce some switches or if
# statements in our code.
#
# Currently (2003/04/08), at least for the Oracle, PostgreSQL, and
# ODBC drivers, the database driver name returned by "ns_db driver" is
# completely sufficient for these switch statements.  But, rather than
# using ns_db driver directly in the switches, we add the simple
# "driver key" layer of indirection between the two, to make the
# default behavior easier to override if that should ever be
# necessary.
#
# --atp@piskorski.com, 2003/04/08 03:39 EDT


# We now use the following global variables:
#
# Server-Wide NSV arrays, keys:
#     db_driverkey         $dbn
#     db_pool_to_dbn       $pool
#
# Global Variables
#    ::acs::default_database
#    ::acs::db_pools($dbn)        (used in db_available_pools)
#    ::acs::db_pool_to_dbn($pool) (used for caching access to nsv db_pool_to_dbn)
#
# Per-thread Tcl global variables:
#   One Tcl Array per Database Name:
#     db_state_${dbn}
#
# The db_available_pools and db_state arrays are used in exactly the
# same manner as they were originally (in ACS 4.0 to OpenACS 4.6
# code), except that in the original DB API we had only one of each
# array total, while now we have one of each array per database.
#
# The db_pool_to_dbn nsv is simply a map to quickly tell use which dbn
# each AOLserver database pool belongs to.  (Any pools which do not
# belong to any dbn have no entry here.)
#
# We use the procs db_state_array_name_is, db_available_pools, and
# db_driverkey to help keep track of these different arrays.
# Note that most code should now NEVER read from any of the
# db_available_pools nsvs listed above, but should instead use the
# proc db_available_pools provided for that purpose.
#
# The original implementation comments on the use of these global
# variables are below:
#
# --atp@piskorski.com, 2003/03/16 21:30 EST


d_proc -private db_state_array_name_is {
    {-dbn ""}
} {
    @return the name of the global db_state array for the given
    database name.

    @param dbn The database name to use.  If empty_string, uses the
    default database.

    @author Andrew Piskorski (atp@piskorski.com)
    @creation-date 2003/03/16
} {
    if { $dbn eq "" } {
        set dbn $::acs::default_database
    }
    #if {[llength [trace info variable ::db_state_${dbn}]] == 0} {
    #    trace add variable ::db_state_${dbn} {array read write unset} [list ::db_tracer ::db_state_${dbn}]
    #}
    return "::db_state_${dbn}"
}

# proc db_tracer {varname name1 name2 op} {
#     if {$name2 eq "handles"} {
#         #ns_log notice "### variable $varname: $name1 ($name2) $op"
#         if {$op eq "write"} {
#             ns_log notice "###### handles updated to <[set ::${varname}($name2)]>"
#         }
#     }
# }

#
# Make sure, we have an nsv_array "db_driverkey", otherwise "nsv_get
# db_driverkey ..." will cause an exception.
#
if {![nsv_array exists db_driverkey]} {
    nsv_set db_driverkey . .
}

d_proc -public db_driverkey {
    {-handle_p 0}
    dbn
} {
    Normally, a dbn is passed to this proc.  Unfortunately, there are
    one or two cases where a proc that needs to call this one has only
    a db handle, not the dbn that handle came from.  Therefore, they
    instead use <code>-handle_p 1</code> and pass the db handle.

    Hmm, as of 2018, it seems that in most cases, db_driverkey is
    called with a handle.

    @return The driverkey for use in db_* API switch statements.

    @author Andrew Piskorski (atp@piskorski.com)
    @creation-date 2003/04/08
} {
    if { $handle_p } {
        #
        # Input is a handle, not a database name. Determine from the
        # handle the db-name (via the pool information)
        #
        set handle $dbn
        set pool [ns_db poolname $handle]
        set dbn $::acs::db_pool_to_dbn($pool)
    }

    if { ![nsv_get db_driverkey $dbn driverkey] } {
        #
        # This ASSUMES that any overriding of this default value via
        # "ns_param driverkey_dbn" has already been done:
        #
        if { $handle_p } {
            set driver [ns_db driver $handle]
        } else {
            db_with_handle -dbn $dbn handle {
                set driver [ns_db driver $handle]
            }
        }

        # These are the default driverkey values, if they are not set
        # in the config file:

        if { [string match "Oracle*" $driver] } {
            set driverkey {oracle}
        } elseif$driver eq "PostgreSQL" } {
            set driverkey "postgresql"
        } elseif$driver eq "ODBC" } {
            set driverkey "nsodbc"
        } else {
            set driverkey {}
            ns_log Error "db_driverkey: Unknown driver '$driver'."
        }

        nsv_set db_driverkey $dbn $driverkey
    }
    return $driverkey
}


ad_proc -public db_type {} {
    @return the RDBMS type (i.e. oracle, postgresql) this OpenACS installation is using.
    The nsv ad_database_type is set up during the bootstrap process.
} {
    #
    # Currently this should always be either "oracle" or "postgresql":
    # --atp@piskorski.com, 2003/03/16 22:01 EST
    #
    # First check, if the database type exists in the namespaced
    # variable. This should be always the case. If this fail, fall
    # back to the old-style nsv (which can be costly in tight db loops)
    #
    if {[info exists ::acs::database_type]} {
        set result $::acs::database_type
    } else {
        set result [nsv_get ad_database_type .]
        ns_log Warning "db_type '$result' had to be obtained from the nsv 'ad_database_type'"
        set ::acs::database_type $result
    }
    return $result
}

ad_proc -public db_compatible_rdbms_p { db_type } {
    @return 1 if the given db_type is compatible with the current RDBMS.
} {
    return [expr { $db_type eq "" || [db_type] eq $db_type }]
}



ad_proc -private db_legacy_package_p { db_type_list } {
    @return 1 if the package is a legacy package.  We can only tell for certain if it explicitly supports Oracle 8.1.6 rather than the OpenACS more general oracle.
} {
    if {"oracle-8.1.6" in $db_type_list} {
        return 1
    }
    return 0
}

ad_proc -public db_version {} {
    @return the RDBMS version (i.e. 8.1.6 is a recent Oracle version; 7.1 a
                               recent PostgreSQL version)
} {
    return [nsv_get ad_database_version .]
}


ad_proc -public db_known_database_types {} {
    @return a list of three-element lists describing the database engines known
    to OpenACS.  Each sublist contains the internal database name
    (used in file paths, etc), the driver name, and a "pretty name"
    to be used in selection forms displayed to the user.

    The nsv containing the list is initialized by the bootstrap script and should
    never be referenced directly by user code.
} {
    return $::acs::known_database_types
}


# db_null, db_quote, db_nullify_empty_string - were all previously
# defined Oracle only, no Postgres equivalent existed at all.  So, it
# can't hurt anything to have them defined in when OpenACS is using
# Postgres too.  --atp@piskorski.com, 2003/04/08 05:34 EDT

ad_proc -deprecated db_null {} {

    @return an empty string, which Oracle thinks is null.

    Deprecated: This routine was invented to provide an RDBMS-specific null
    value but doesn't actually work.  I (DRB) left it in to speed porting - we
    should really clean up the code and pull out the calls instead, though.

    @see ""
} {
    return ""
}

ad_proc -deprecated db_quote { string } {

    Quotes a string value to be placed in a SQL statement.  Use the
    built-in ns_dbquotevalue instead, which cares also about the
    surrounding quotes.

    @see ns_dbquotevalue
} {
    regsub -all -- {'} "$string" {''} result
    return $result
}

ad_proc -public -deprecated db_nullify_empty_string { string } {
    A convenience function that returns [db_null] if $string is the empty string.

    Deprecated: essentially just returns the passed string.

    @see: db_null
} {
    return $string
}

ad_proc -public db_boolean { bool } {
    Converts a Tcl boolean (1/0) into a SQL boolean (t/f)
    @return t or f
} {
    if { $bool } {
        return "t"
    } else {
        return "f"
    }
}

d_proc -public db_nextval {
    { -dbn "" }
    sequence
} {

    Example:

    <pre>
    set new_object_id [db_nextval acs_object_id_seq]
    </pre>

    @return the next value for a sequence. This can utilize a pool of
    sequence values.

    @param sequence the name of an SQL sequence

    @param dbn The database name to use.  If empty_string, uses the default database.

    @see <a href="/doc/db-api-detailed">/doc/db-api-detailed</a>
} {
    set driverkey [db_driverkey $dbn]

    # PostgreSQL has a special implementation here, any other db will
    # probably work with the default:

    switch -- $driverkey {

        postgresql {
            #             # the following query will return a nextval if the sequence
            #             # is of relkind = 'S' (a sequence).  if it is not of relkind = 'S'
            #             # we will try querying it as a view:

            #             if { [db_0or1row -dbn $dbn nextval_sequence "
            #                 select nextval('${sequence}') as nextval
            #                 where (select relkind
            #                        from pg_class
            #                        where relname = '${sequence}') = 'S'
            #             "]} {
            #                 return $nextval
            #             } else {
            #                 ns_log debug "db_nextval: sequence($sequence) is not a real sequence.  perhaps it uses the view hack."
            #                 db_0or1row -dbn $dbn nextval_view "select nextval from ${sequence}"
            #                 return $nextval
            #             }
            #
            # The code above is just for documentation, how it worked
            # before the change below. We keep now a per-thread table of
            # the "known" sequences to avoid at run time the query,
            # whether the specified sequence is a real sequence or a
            # view. This change makes this function more than a factor
            # of 2 faster than before.
            #
            # Note that solely the per-thread information won't work for
            # freshly created sequences. Therefore, we keep the old
            # code for checking at run time in the database for such
            # occurrences.
            #
            # Note that the sequence handling in OpenACS is quite a
            # mess.  Some sequences are named t_SEQUENCE (10 in
            # dotlrn), others are called just SEQUENCE (18 in dotlrn),
            # for some sequences, additional views are defined with an
            # attribute 'nextval', and on top of this, db_nextval is
            # called sometimes with the view name and sometimes with
            # the sequence name. Checking this at run time is
            # unnecessary complex and costly.
            #
            # The best solution would certainly be to call "db_nextval"
            # only with real sequence names (as defined in SQL). In that
            # case, the whole function would for postgres would collapse
            # to a single line, without any need for sequence name
            # caching. But in that case, one should rename the sequences
            # from t_SEQUENCE to SEQUENCE for postgres.
            #
            # However, since Oracle uses the pseudo column ".nextval",
            # which is emulated via the view, it is not clear, how
            # feasible this is to remove all such views without breaking
            # installed applications.  We keep for such cases the view,
            # but nevertheless, the function "db_nextval" should always
            # be called with names without the "t_" prefix to achieve
            # Oracle compatibility.

            if {![info exists ::db::sequences]} {
                ns_log notice "-- creating per thread sequence table"
                namespace eval ::db {}
                foreach s [db_list -dbn $dbn relnames "select relname, relkind  from pg_class where relkind = 'S'"] {
                    set ::db::sequences($s) 1
                }
            }
            if {[info exists ::db::sequences(t_$sequence)]} {
                #ns_log notice "-- found t_$sequence
                #ad_log Warning "Deprecated sequence name 't_$sequence' is used. Use instead 't_$sequence'"
                set nextval [db_string -dbn $dbn nextval "select nextval('t_$sequence')"]
            } elseif {[info exists ::db::sequences($sequence)]} {
                #ns_log notice "-- found $sequence"
                set nextval [db_string -dbn $dbn nextval "select nextval('$sequence')"]
                if {[string match t_* $sequence]} {
                    ad_log Warning "For portability, db_nextval should be called without the leading 't_' prefix: 't_$sequence'"
                }
            } elseif { [db_0or1row -dbn $dbn nextval_sequence "
                 select nextval('${sequence}') as nextval
                 where (select relkind
                        from pg_class
                        where relname = '${sequence}') = 'S'
             "]} {
                #
                # We do not have an according sequence-table. Use the system catalog to check
                # for the sequence
                #
                # ... the query sets nextval if it succeeds
                #
                ad_log Warning "Probably deprecated sequence name '$sequence' is used (no sequence table found)"
            } else {
                #
                # Finally, there might be a view with a nextval
                #
                ns_log debug "db_nextval: sequence($sequence) is not a real sequence.  perhaps it uses the view hack."
                set nextval [db_string -dbn $dbn nextval "select nextval from $sequence"]
                ad_log Warning "Using deprecated sequence view hack for '$sequence'. Is there not real sequence?"
            }

            return $nextval
        }

        oracle -
        nsodbc -
        default {
            return [db_string -dbn $dbn nextval "select $sequence.nextval from dual"]
        }
    }
}

d_proc -public db_nth_pool_name {
    { -dbn "" }
    n
} {
    @return the name of the pool used for the nth-nested selection (0-relative).
    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    set available_pools [db_available_pools $dbn]

    if { $n < [llength $available_pools] } {
        set pool [lindex $available_pools $n]
    } else {
        return -code error "Ran out of database pools ($available_pools)"
    }
    return $pool
}

if {[acs::icanuse "ns_db currenthandles"]} {

    ns_log notice "... I can use 'ns_db currenthandles'"

    #
    # This branch uses "ns_db currenthandles" to implement
    # "db_with_handle" instead of the old approach based on the global
    # db_state variables. The new approach has the advantage that it
    # is:
    #
    # - more robust (deletion and creation of the per-request variables,
    #   no coherency problem),
    # - simpler, and
    # - faster (less overhead per db_with_handle call)
    #
    #     time {db_string . {select object_id from acs_objects limit 1}} 1000
    #     old: 160-190 microseconds per iteration
    #     new: 150-180 microseconds per iteration
    #
    #     time {xo::dc get_value . {select object_id from acs_objects limit 1}} 1000
    #     old: 110-120
    #     new: 105-110
    #
    #     set id -1
    #     time {xo::dc get_value -prepare {int} . {select object_id from acs_objects where object_id=:id}} 1000
    #     old: 80-100
    #     new: 76-90
    #
    # Still, more improvement can be done (GN).
    #
    d_proc -public db_with_handle {
        { -dbn "" }
        db code_block
    } {
        Place a usable database handle in <i>db</i> and executes
        <i>code_block</i>.

        @param dbn         Database name to use. If empty_string, use the default database
        @param db          Name of the handle variable used in the code block
        @param code_block  code block to be executed with handle
    } {
        #
        # Let the caller decide, how the handle variable is called in
        # the code block.
        #
        upvar 1 $db dbh

        #
        # Get the pools and the current allocated handles for this thread.
        #
        set pools [db_available_pools $dbn]
        set currentHandles [ns_db currenthandles]
        #ns_log notice "### pools <$pools> currentHandles <$currentHandles>"

        set db ""
        set n 0
        foreach pool $pools {
            #
            # Do we have already handles allocated from this pool?
            #
            if {[dict exists $currentHandles $pool]} {
                #
                # Are there handles, which are not active (i.e. not in
                # a currently open "ns_db select" and "ns_db getrow"
                # context.
                #
                foreach {handle active} [dict get $currentHandles $pool] {
                    #ns_log notice "### FOUND pool $pool handle $handle active $active"
                    if {$active eq "0"} {
                        #
                        # We can use this handle
                        #
                        set db $handle
                        break
                    }
                }
            } else {
                break
            }
            incr n
        }
        #
        # In case, we got no handle above, we have to allocate a
        # handle from the next pool, from which we have not got a
        # handle before.
        #
        if {$db eq ""} {
            #
            # We were not successful above
            #
            set pool [lindex $pools $n]
            if {$pool eq ""} {
                ad_log error "handles from all pools <$pools> are exhausted"
                error "could not obtain handle, all pools are exhausted"
            }
            set start_time [expr {[clock clicks -microseconds]/1000.0}]
            #ns_log notice "### BEFORE gethandle $pool ($n)"
            set errno [catch {
                set db [ns_db gethandle $pool]
            } error]
            #ad_log notice "### AFTER gethandle $pool errno $errno handle <$db> currentHandles [ns_db currenthandles]"
            ds_collect_db_call $db gethandle "" $pool $start_time $errno $error
            if { $errno } {
                ns_log notice "### RETURNING error $error"
                return -code $errno -errorcode $::errorCode -errorinfo $::errorInfo $error
            }
        }
        #ns_log notice "### db_with_handle has handle <$db>"

        set dbh $db
        set errno [catch { uplevel 1 $code_block } error]

        # Unset dbh, so any subsequence use of this variable will bomb.
        unset -nocomplain dbh

        # If errno is 1, it's an error, so return errorCode and errorInfo;
        # if errno = 2, it's a return, so don't try to return errorCode/errorInfo
        # errno = 3 or 4 give undefined results

        if { $errno == 1 } {
            # A real error occurred
            ns_log notice "### db_with_handle returned error <$error> for statement $code_block"
            return -code $errno -errorcode $::errorCode -errorinfo $::errorInfo $error
        }

        if { $errno == 2 } {

            # The code block called a "return", so pass the message through but don't try
            # to return errorCode or errorInfo since they may not exist

            return -code $errno $error
        }
    }

    #
    # db_last_used_handle
    #
    ad_proc -private db_last_used_handle {{-dbn ""}} {
        Get the last used inactive handle.

        @param dbn database name
        @return last active handle or empty string
    } {
        set pools [db_available_pools $dbn]
        set currentHandles [ns_db currenthandles]

        set last_used_handle ""
        foreach pool $pools {
            if {[dict exists $currentHandles $pool]} {
                foreach {handle active} [dict get $currentHandles $pool] {
                    #ns_log notice "### FOUND pool $pool handle $handle active $active"
                    if {$active eq 0} {
                        set last_used_handle $handle
                    }
                }
            }
        }
        #ns_log notice "###### db_last_used_handle: <$currentHandles> last used $last_used_handle"
        return $last_used_handle
    }

    #
    # db_release_unused_handles
    #
    ad_proc -public db_release_unused_handles {{-dbn ""}} {
        Releases any database handles that are presently unused.

        @param dbn The database name to use.  If empty_string, uses the default database.
    } {
        # we need the state array still for transaction handling
        upvar "#0" [db_state_array_name_is -dbn $dbn] db_state

        set pools [db_available_pools $dbn]
        set currentHandles [ns_db currenthandles]

        foreach pool $pools {
            if {[dict exists $currentHandles $pool]} {
                foreach {handle active} [dict get $currentHandles $pool] {
                    #ns_log notice "### FOUND pool $pool handle $handle active $active"
                    if {$active eq 0} {
                        # Don't release handles which are part of a transaction.
                        if { [info exists db_state(transaction_level,$handle)]
                             && $db_state(transaction_level,$handle) > 0
                         } {
                            continue
                        }
                        set start_time [expr {[clock clicks -microseconds]/1000.0}]
                        ns_db releasehandle $handle
                        #ns_log notice "### AFTER releasehandle [ns_db currenthandles $pool]"
                        ds_collect_db_call $handle releasehandle "" "" $start_time 0 ""
                    }
                }
            }
        }
    }


} else {

    #
    # This is the legacy branch without [ns_db currenthandles], using
    # the global state variables.
    #
    ns_log notice "... cannot use 'ns_db currenthandles'"

    d_proc -public db_with_handle {
        { -dbn "" }
        db code_block
    } {

        Places a usable database handle in <i>db</i> and executes <i>code_block</i>.

        @param dbn The database name to use.  If empty_string, uses the default database.
    } {
        upvar 1 $db dbh
        upvar "#0" [db_state_array_name_is -dbn $dbn] db_state

        # Initialize bookkeeping variables.
        if { ![info exists db_state(handles)] } {
            set db_state(handles) [list]
        }
        if { ![info exists db_state(n_handles_used)] } {
            set db_state(n_handles_used) 0
        }
        if { $db_state(n_handles_used) >= [llength $db_state(handles)] } {
            set pool [db_nth_pool_name -dbn $dbn $db_state(n_handles_used)]
            set start_time [expr {[clock clicks -microseconds]/1000.0}]
            set errno [catch {
                set db [ns_db gethandle $pool]
            } error]
            ds_collect_db_call $db gethandle "" $pool $start_time $errno $error
            lappend db_state(handles) $db
            if { $errno } {
                return -code $errno -errorcode $::errorCode -errorinfo $::errorInfo $error
            }
        }
        set my_dbh [lindex $db_state(handles) $db_state(n_handles_used)]
        set dbh $my_dbh
        set db_state(last_used) $my_dbh

        incr db_state(n_handles_used)
        set errno [catch { uplevel 1 $code_block } error]
        incr db_state(n_handles_used) -1

        # This may have changed while the code_block was being evaluated.
        set db_state(last_used) $my_dbh

        # Unset dbh, so any subsequence use of this variable will bomb.
        unset -nocomplain dbh

        # If errno is 1, it's an error, so return errorCode and errorInfo;
        # if errno = 2, it's a return, so don't try to return errorCode/errorInfo
        # errno = 3 or 4 give undefined results

        if { $errno == 1 } {
            # A real error occurred
            return -code $errno -errorcode $::errorCode -errorinfo $::errorInfo $error
        }

        if { $errno == 2 } {

            # The code block called a "return", so pass the message through but don't try
            # to return errorCode or errorInfo since they may not exist

            return -code $errno $error
        }
    }

    ad_proc -private db_last_used_handle {{-dbn ""}} {
        Get the last used handle

        @param dbn database name
        @return last active handle or empty string
    } {
        upvar "#0" [db_state_array_name_is -dbn $dbn] db_state

        return $db_state(last_used)
    }

    ad_proc -public db_release_unused_handles {{-dbn ""}} {

        Releases any database handles that are presently unused.

        @param dbn The database name to use.  If empty_string, uses the default database.
    } {
        upvar "#0" [db_state_array_name_is -dbn $dbn] db_state

        if { [info exists db_state(n_handles_used)] } {
            # Examine the elements at the end of db_state(handles), killing off
            # handles that are unused and not engaged in a transaction.

            set index_to_examine [expr { [llength $db_state(handles)] - 1 }]
            while { $index_to_examine >= $db_state(n_handles_used) } {
                set db [lindex $db_state(handles) $index_to_examine]

                # Stop now if the handle is part of a transaction.
                if { [info exists db_state(transaction_level,$db)]
                     && $db_state(transaction_level,$db) > 0
                 } {
                    break
                }

                set pool [db_nth_pool_name -dbn $dbn $db_state(n_handles_used)]
                set start_time [expr {[clock clicks -microseconds]/1000.0}]
                ns_db releasehandle $db
                ds_collect_db_call $db releasehandle "" "" $start_time 0 ""
                incr index_to_examine -1
            }
            set db_state(handles) [lrange $db_state(handles) 0 $index_to_examine]
        }
    }


}

ad_proc -public db_resultrows {{-dbn ""}} {
    @return the number of rows affected by the last DML command.

    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    set driverkey [db_driverkey $dbn]

    switch -- $driverkey {
        oracle {
            return [ns_ora resultrows [db_last_used_handle -dbn $dbn]]
        }
        postgresql {
            return [ns_pg ntuples [db_last_used_handle -dbn $dbn]]
        }
        nsodbc {
            error "db_resultrows is not supported for this database."
        }
        default {
            error "Unknown database driver.  db_resultrows is not supported for this database."
        }
    }
}



d_proc -public db_exec_plsql {
    {-dbn ""}
    statement_name
    sql
    -bind
} {

    <strong>Oracle:</strong>
    Executes a PL/SQL statement, and returns the variable of bind
    variable <code>:1</code>.

    <p>
    <strong>PostgreSQL:</strong>
    Performs a pl/pgsql function or procedure call.  The caller must
    perform a select query that returns the value of the function.

    <p>
    Examples:

    <p>
    <pre>
    # Oracle:
    db_exec_plsql delete_note {
        begin  note.del(:note_id);  end;
    }

    # PostgreSQL:
    db_exec_plsql delete_note {
        select note__delete(:note_id);
    }
    </pre>

    <p>
    If you need the return value, then do something like this:

    <p>
    <pre>
    # Oracle:
    set new_note_id [db_exec_plsql create_note {
        begin
        :1 := note.new(
                       owner_id => :user_id,
                       title    => :title,
                       body     => :body,
                       creation_user => :user_id,
                       creation_ip   => :peeraddr,
                       context_id    => :package_id
                       );
        end;
    }]

    # PostgreSQL:
    set new_note_id [db_exec_plsql create_note {
        select note__new(
                         null,
                         :user_id,
                         :title,
                         :body,
                         'note',
                         now(),
                         :user_id,
                         :peeraddr,
                         :package_id
                         );
    }]
    </pre>

    <p>
    You can call several pl/SQL statements at once, like this:

    <p>
    <pre>
    # Oracle:
    db_exec_plsql delete_note {
        begin
        note.del(:note_id);
        note.del(:another_note_id);
        note.del(:yet_another_note_id);
        end;
    }

    # PostgreSQL:
    db_exec_plsql delete_note {
        select note__delete(:note_id);
        select note__delete(:another_note_id);
        select note__delete(:yet_another_note_id);
    }
    </pre>

    If you are using xql files then put the body of the query in a
    <code>yourfilename-oracle.xql</code> or <code>yourfilename-postgresql.xql</code> file, as appropriate. E.g. the first example
    transformed to use xql files looks like this:


    <p>
    <code>yourfilename.tcl</code>:<br>
    <p>
    <pre>
    db_exec_plsql delete_note {}</pre>

    <p>
    <code>yourfilename-oracle.xql</code>:<br>
    <p>
    <pre>
    &lt;fullquery name="delete_note">
    &lt;querytext>
    begin
    note.del(:note_id);
    end;
    &lt;/querytext>
    &lt;/fullquery></pre>

    <p>
    <code>yourfilename-postgresql.xql</code>:<br>
    <p>
    <pre>
    &lt;fullquery name="delete_note">
    &lt;querytext>
    select note__delete(:note_id);
    &lt;/querytext>
    &lt;/fullquery></pre>


    @param dbn The database name to use.  If empty_string, uses the default database.

    @see <a href="/doc/db-api-detailed">/doc/db-api-detailed</a>
} {

    # Query Dispatcher (OpenACS - ben)
    set full_statement_name [db_qd_get_fullname $statement_name]

    set driverkey [db_driverkey $dbn]
    switch -- $driverkey {
        postgresql {
            set postgres_p 1
        }

        oracle -
        nsodbc -
        default {
            set postgres_p 0
        }
    }

    if { ! $postgres_p } {
        db_with_handle -dbn $dbn db {
            # Right now, use :1 as the output value if it occurs in the statement,
            # or not otherwise.
            set test_sql [db_qd_replace_sql $full_statement_name $sql]
            if { [regexp {:1} $test_sql] } {
                return [db_exec exec_plsql_bind $db $full_statement_name $sql 2 1 ""]
            } else {
                return [db_exec dml $db $full_statement_name $sql]
            }
        }
    } else {
        # Postgres doesn't have PL/SQL, of course, but it does have
        # PL/pgSQL and other procedural languages.  Rather than assign the
        # result to a bind variable which is then returned to the caller,
        # the Postgres version of OpenACS requires the caller to perform a
        # select query that returns the value of the function.

        # I'm not happy about having to get the fullname here, but right now
        # I can't figure out a cleaner way to do it. I will have to
        # revisit this ASAP. (ben)
        set full_statement_name [db_qd_get_fullname $statement_name]

        db_with_handle -dbn $dbn db {
            # plsql calls that are simple selects bypass the plpgsql
            # mechanism for creating anonymous functions (OpenACS - Dan).
            # if a table is being created, we need to bypass things, too (OpenACS - Ben).
            set test_sql [db_qd_replace_sql $full_statement_name $sql]
            if {[regexp -nocase -- {^\s*select} $test_sql match]} {
                # ns_log Debug "PLPGSQL: bypassed anon function"
                set selection [db_exec 0or1row $db $full_statement_name $sql]
            } elseif {[regexp -nocase -- {^\s*(create|drop) table} $test_sql match]} {
                ns_log Debug "PLPGSQL: bypassed anon function for create/drop table"
                set selection [db_exec dml $db $full_statement_name $sql]
                return ""
            } else {
                # ns_log Debug "PLPGSQL: using anonymous function"
                set selection [db_exec_plpgsql $db $full_statement_name $sql $statement_name]
            }
            return [ns_set value $selection 0]
        }
    }
}


ad_proc -private db_exec_plpgsql { db statement_name pre_sql fname } {

    <strong>PostgreSQL only.</strong>
    <p>

    A helper procedure to execute a SQL statement, potentially binding
    depending on the value of the $bind variable in the calling environment
    (if set).

    <p>
    Low level replacement for db_exec which replaces inline code with a proc.
    db proc is dropped after execution.  This is a temporary fix until we can
    port all of the db_exec_plsql calls to simple selects of the inline code
    wrapped in function calls.

    <p>
    emulation of plsql calls from oracle.  This routine takes the plsql
    statements and wraps them in a function call, calls the function, and then
    drops the function. Future work might involve converting this to cache the
    function calls

    <p>
    This proc is <b>private</b> - use db_exec_plsql instead!

    @see db_exec_plsql

} {
    set start_time [expr {[clock clicks -microseconds]/1000.0}]

    set sql [db_qd_replace_sql -ulevel 3 $statement_name $pre_sql]

    set unique_id [db_nextval "anon_func_seq"]

    set function_name "__exec_${unique_id}_${fname}"

    # insert Tcl variable values (OpenACS - Dan)
    if {$sql ne $pre_sql } {
        set sql [uplevel 2 [list subst -nobackslashes $sql]]
    }
    ns_log Debug "PLPGSQL: converted: $sql to: select $function_name ()"

    # create a function definition statement for the inline code
    # binding is emulated in tcl. (OpenACS - Dan)

    set errno [catch {
        upvar bind bind
        if { [info exists bind] && [llength $bind] != 0 } {
            if { [llength $bind] == 1 } {
                set proc_sql [db_bind_var_substitution $sql [ns_set array $bind]]
            } else {
                set proc_sql [db_bind_var_substitution $sql $bind]
            }
        } else {
            set proc_sql [uplevel 2 [list db_bind_var_substitution $sql]]
        }

        ns_db dml $db "create function $function_name () returns varchar as [::ns_dbquotevalue $proc_sql] language 'plpgsql'"

        set ret_val [ns_db 0or1row $db "select $function_name ()"]

        # drop the anonymous function (OpenACS - Dan)
        # JCD: ignore return code -- maybe we should be smarter about this though.
        catch {ns_db dml $db "drop function $function_name ()"}

        return $ret_val

    } error]

    set errinfo $::errorInfo
    set errcode $::errorCode

    ds_collect_db_call $db 0or1row $statement_name $sql $start_time $errno $error

    if { $errno == 2 } {
        return $error
    } else {
        catch {ns_db dml $db "drop function $function_name ()"}
    }

    return -code $errno -errorinfo $errinfo -errorcode $errcode $error
}

ad_proc -private db_get_quote_indices { sql } {
    Given a piece of SQL, return the indices of single quotes.
    This is useful when we do bind var substitution because we should
    not attempt bind var substitution inside quotes. Examples:

    <pre>
    sql          return value
    {'a'}           {0 2}
    {'a''}           {}
    {'a'a'a'}       {0 2 4 6}
    {a'b'c'd'}      {1 3 5 7}
    </pre>

    @see db_bind_var_substitution
} {
    set quote_indices [list]

    # Returns a list on the format
    # Example - for sql={'a'a'a'} returns
    # {0 2} {0 0} {2 2} {3 6} {4 4} {6 6}
    set all_indices [regexp -inline -indices -all -- {(?:^|[^'])(')(?:[^']|'')+(')(?=$|[^'])} $sql]

    for {set i 0} { $i < [llength $all_indices] } { incr i 3 } {
        lappend quote_indices [lindex $all_indices $i+1 0] [lindex $all_indices $i+2 0]
    }

    return $quote_indices
}

ad_proc -private db_bind_var_quoted_p { sql bind_start_idx bind_end_idx} {

} {
    foreach {quote_start_idx quote_end_idx} [db_get_quote_indices $sql] {
        if { $bind_start_idx > $quote_start_idx && $bind_end_idx < $quote_end_idx } {
            return 1
        }
    }

    return 0
}

ad_proc -public db_bind_var_substitution { sql { bind "" } } {

    This proc emulates the bind variable substitution in the PostgreSQL driver.
    Since this is a temporary hack, we do it in Tcl instead of hacking up the
    driver to support plsql calls.  This is only used for the db_exec_plpgsql
    function.

} {
    if {$bind eq ""} {
        upvar __db_sql lsql
        set lsql $sql
        uplevel {
            set __db_lst [regexp -inline -indices -all -- {:?:\w+} $__db_sql]
            for {set __db_i [expr {[llength $__db_lst] - 1}]} {$__db_i >= 0} {incr __db_i -1} {
                set __db_ws [lindex $__db_lst $__db_i 0]
                set __db_we [lindex $__db_lst $__db_i 1]
                set __db_bind_var [string range $__db_sql $__db_ws $__db_we]
                if {![string match "::*" $__db_bind_var] && ![db_bind_var_quoted_p $__db_sql $__db_ws $__db_we]} {
                    set __db_tcl_var [string range $__db_bind_var 1 end]
                    set __db_tcl_var [set $__db_tcl_var]
                    if {$__db_tcl_var eq ""} {
                        set __db_tcl_var null
                    } else {
                        set __db_tcl_var "[::ns_dbquotevalue $__db_tcl_var]"
                    }
                    set __db_sql [string replace $__db_sql $__db_ws $__db_we $__db_tcl_var]
                }
            }
        }
    } else {

        array set bind_vars $bind

        set lsql $sql
        set lst [regexp -inline -indices -all -- {:?:\w+} $sql]
        for {set i [expr {[llength $lst] - 1}]} {$i >= 0} {incr i -1} {
            set ws [lindex $lst $i 0]
            set we [lindex $lst $i 1]
            set bind_var [string range $sql $ws $we]
            if {![string match "::*" $bind_var] && ![db_bind_var_quoted_p $lsql $ws $we]} {
                set tcl_var [string range $bind_var 1 end]
                set val $bind_vars($tcl_var)
                if {$val eq ""} {
                    set val null
                } else {
                    set val "[::ns_dbquotevalue $val]"
                }
                set lsql [string replace $lsql $ws $we $val]
            }
        }
    }

    return $lsql
}


ad_proc -private db_getrow { db selection } {

    A helper procedure to perform an ns_db getrow, invoking developer support
    routines as necessary.

} {
    set start_time [expr {[clock clicks -microseconds]/1000.0}]
    set errno [catch { return [ns_db getrow $db $selection] } error]
    ds_collect_db_call $db getrow "" "" $start_time $errno $error
    if { $errno == 2 } {
        return $error
    }
    return -code $errno -errorinfo $::errorInfo -errorcode $::errorCode $error
}


ad_proc -public db_exec { {-subst all} type db statement_name pre_sql {ulevel 2} args } {

    A helper procedure to execute a SQL statement, potentially binding
    depending on the value of the $bind variable in the calling environment
    (if set).

} {
    set start_time [expr {[clock clicks -microseconds]/1000.0}]
    set driverkey [db_driverkey -handle_p 1 $db]

    set sql [db_qd_replace_sql \
                 -ulevel [expr {$ulevel +1 }] \
                 -subst $subst \
                 $statement_name \
                 $pre_sql]

    set errno [catch {
        upvar bind bind

        if { [info exists bind] && [llength $bind] != 0 } {
            if { [llength $bind] == 1 } {
                # $bind is an ns_set id:

                switch -- $driverkey {
                    oracle {
                        return [ns_ora $type $db -bind $bind $sql {*}$args]
                    }
                    postgresql {
                        return [ns_pg_bind $type $db -bind $bind $sql]
                    }
                    nsodbc {
                        return [ns_odbc_bind $type $db -bind $bind $sql]
                    }
                    default {
                        error "Unknown database driver.  Bind variables not supported for this database."
                    }
                }

            } else {
                # $bind is a Tcl list, convert it to an ns_set:
                set bind_vars [ns_set create]
                foreach { name value } $bind {
                    ns_set put $bind_vars $name $value
                }
            }

            switch -- $driverkey {
                oracle {
                    # TODO: Using $args outside the list is
                    # potentially bad here, depending on what is in
                    # args and if the items contain any embedded
                    # whitespace.  Or maybe it works fine.  But it's
                    # hard to know.  Document or fix.
                    # --atp@piskorski.com, 2003/04/09 15:33 EDT

                    return [ns_ora $type $db -bind $bind_vars $sql {*}$args]
                }
                postgresql {
                    return [ns_pg_bind $type $db -bind $bind_vars $sql]
                }
                nsodbc {
                    return [ns_odbc_bind $type $db -bind $bind_vars $sql]
                }
                default {
                    error "Unknown database driver.  Bind variables not supported for this database."
                }
            }

        } else {
            # Bind variables, if any, are defined solely as individual
            # Tcl variables:

            switch -- $driverkey {
                oracle {
                    return [uplevel $ulevel [list ns_ora $type $db $sql$args]
                }
                postgresql {
                    return [uplevel $ulevel [list ns_pg_bind $type $db $sql]]
                }
                nsodbc {
                    return [uplevel $ulevel [list ns_odbc_bind $type $db $sql]]
                }
                default {
                    # Using plain ns_db like this will work ONLY if
                    # the query is NOT using bind variables:
                    # --atp@piskorski.com, 2001/09/03 08:41 EDT
                    return [uplevel $ulevel [list ns_db $type $db $sql$args]
                }
            }
        }
    } error]

    #
    # If db_exec took more than a threshold, yack about it. We have to
    # be careful there, since this might be called during bootstrap,
    # where "parameter::get_from_package_key" is not yet defined. We
    # cannot use "parameter::get_from_package_key" directly, since
    # this needs an SQL query, leading to an infinite recursion. So,
    # we use a per-thread variable, which is set at startup and then
    # updated, whenever the parameter changes.
    #
    set complain_time [expr {[info exists ::acs::DbLogMinDuration] ? $::acs::DbLogMinDuration : 2000 }]
    if { [clock clicks -milliseconds] - $start_time > $complain_time} {
        set duration [format %.2f [expr {[clock clicks -milliseconds] - $start_time}]]
        ns_log Warning "db_exec: longdb $duration ms $db $type $statement_name"
    } else {
        #set duration [format %.2f [expr {[clock clicks -milliseconds] - $start_time}]]
        #ns_log Debug "db_exec: timing $duration seconds $db $type $statement_name"
    }

    ds_collect_db_call $db $type $statement_name $sql $start_time $errno $error
    if { $errno == 2 } {
        return $error
    }

    return -code $errno -errorinfo $::errorInfo -errorcode $::errorCode $error
}


d_proc -public db_string {
    {-dbn ""}
    -cache_key
    {-cache_pool db_cache_pool}
    {-subst all}
    statement_name
    sql
    -default
    -bind
} {

    @return the first column of the result of the SQL query <i>sql</i>.  If the query doesn't return a row, returns <i>default</i> or raises an error if no <i>default</i> is provided.

    @param dbn The database name to use.  If empty_string, uses the default database.
    @param cache_key Cache the result using given value as the key.  Default is to not cache.
    @param cache_pool Override the default db_cache_pool
    @param subst Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands
    @param default Return value in case the SQL query returns no value
    @param bind bind variables, passed either as an ns_set id, or via bind value list
    
    @param statement_name name of the SQL query
    @param sql SQL query to be executed

} {
    # Query Dispatcher (OpenACS - ben)
    set full_name [db_qd_get_fullname $statement_name]

    set code {
        db_with_handle -dbn $dbn db {
            set selection [db_exec -subst $subst 0or1row $db $full_name $sql]
        }
        if { $selection eq ""} {
            if { [info exists default] } {
                return $default
            }
            error "Selection did not return a value, and no default was provided"
        }
        return [ns_set value $selection 0]
    }

    if { [info exists cache_key] } {
        return [ns_cache eval $cache_pool $cache_key $code]
    } else {
        return [eval $code]
    }
}


d_proc -public db_list {
    {-dbn ""}
    -cache_key
    {-cache_pool db_cache_pool}
    {-subst all}
    statement_name
    sql
    -bind
} {

    @return a Tcl list of the values in the first column of the result of SQL query <tt>sql</tt>.
    If the SQL query doesn't return any rows, returns an empty list.

    @param dbn The database name to use.  If empty_string, uses the default database.
    @param cache_key Cache the result using given value as the key.  Default is to not cache.
    @param cache_pool Override the default db_cache_pool
    @param subst Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands
    @param bind bind variables, passed either as an ns_set id, or via bind value list

    @param statement_name name of the SQL query.
    @param sql SQL query to be executed.
} {

    # Query Dispatcher (OpenACS - SDW)
    set full_statement_name [db_qd_get_fullname $statement_name]

    # Can't use db_foreach in this proc, since we need to use the ns_set directly.

    set code {
        db_with_handle -dbn $dbn db {
            set selection [db_exec -subst $subst select $db $full_statement_name $sql]
            set result [list]
            while { [db_getrow $db $selection] } {
                lappend result [ns_set value $selection 0]
            }
        }
        return $result
    }
    if { [info exists cache_key] } {
        return [ns_cache eval $cache_pool $cache_key $code]
    } else {
        return [eval $code]
    }
}

d_proc -public db_list_of_ns_sets {
    {-dbn ""}
    {-subst all}
    {-columns_var ""}
    statement_name
    sql
    -bind
} {
    @return a list of ns_sets with the values of each column of each row
    returned by the SQL query specified.

    @param bind bind variables, passed either as an ns_set id, or via bind value list
    @param dbn The database name to use.  If empty_string, uses the default database.
    @param statement_name name of the SQL query.
    @param sql SQL query to be executed.

    @return list of ns_sets, one per each row return by the SQL query

} {
    set full_statement_name [db_qd_get_fullname $statement_name]

    #
    # For large queries, "db_list_of_ns_sets" might be suboptimal,
    # since it requires to held the results twice, needing more memory
    # compared to other variants. Often, this can be replaced easily,
    # sometimes replacing it might require some refactoring.
    #
    # This is e.g. the case for "select_notifications"; don't
    # complain about it for now.
    #
    if {$statement_name ne "select_notifications"} {
        ns_log notice "consider replacing: db_list_of_ns_sets $full_statement_name $sql"
    }

    db_with_handle -dbn $dbn db {
        set result [list]
        set selection [db_exec -subst $subst select $db $full_statement_name $sql]

        while { [db_getrow $db $selection] } {
            lappend result [ns_set copy $selection]
        }
        if {$columns_var ne ""} {
            upvar 1 $columns_var __columns
            if {[acs::icanuse "ns_set keys"]} {
                set __columns [ns_set keys $selection]
            } else {
                set __columns [dict keys [ns_set array $selection]]
            }
        }
    }

    return $result
}

d_proc -public db_list_of_lists {
    {-dbn ""}
    -cache_key
    {-cache_pool db_cache_pool}
    -with_headers:boolean
    {-subst all}
    {-columns_var ""}
    statement_name
    sql
    -bind
} {
    @param with_headers when specified, first line of returned list of
    lists will always be the list of column names as reported by the
    database. Useful when you want to dynamically assign variables to
    values returned in the list of lists.

    @return a Tcl list, each element of which is a list of all column
    values in a row of the result of the SQL query<tt>sql</tt>. If
    <tt>sql</tt> doesn't return any rows, returns an empty list,
    unless with_headers flag was specified and in this case the only
    element in the list will be the list of headers.

    It checks if the element is I18N and replaces it, thereby
    reducing the need to do this with every single package

    @param dbn The database name to use.  If empty_string, uses the default database.
    @param cache_key Cache the result using given value as the key.  Default is to not cache.
    @param cache_pool Override the default db_cache_pool
    @param subst Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands
    @param bind bind variables, passed either as an ns_set id, or via bind value list
    @param statement_name name of the SQL query.
    @param sql SQL query to be executed.

} {
    set full_statement_name [db_qd_get_fullname $statement_name]

    db_with_handle -dbn $dbn db {

        set code {
            set result {}
            set selection [db_exec -subst $subst select $db $full_statement_name $sql]
            #ns_log notice "ll2: $sql -> $selection"
            #ns_log notice "ll2: $sql -> [ns_set array $selection]"

            if {$with_headers_p || $columns_var ne ""} {
                if {$columns_var ne ""} {
                    upvar 1 $columns_var headers
                }
                if {[acs::icanuse "ns_set keys"]} {
                    set headers [ns_set keys $selection]
                } else {
                    set headers [dict keys [ns_set array $selection]]
                }
                if {$with_headers_p} {
                    set result [list $headers]
                }
            }

            if {[acs::icanuse "ns_set values"]} {
                while { [db_getrow $db $selection] } {
                    lappend result [ns_set values $selection]
                }
            } else {
                while { [db_getrow $db $selection] } {
                    lappend result [dict values [ns_set array $selection]]
                }
            }
            set result
        }

        if { [info exists cache_key] } {
            return [ns_cache eval $cache_pool $cache_key $code]
        } else {
            return [eval $code]
        }
    }

    return $result
}


d_proc -public db_foreach {
    {-dbn ""}
    {-subst all}
    statement_name
    sql
    args
} {

    Usage:
    <blockquote>
    db_foreach <em><i>statement-name sql</i></em> [ -bind <em><i>bind_set_id</i></em> | -bind <em><i>bind_value_list</i></em> ] \
        [ -column_array <em><i>array_name</i></em> | -column_set <em><i>set_name</i></em> ] \
        <em><i>code_block</i></em> [ if_no_rows <em><i>if_no_rows_block</i> ]</em>

    </blockquote>

    <p>Performs the SQL query <em><i><tt>sql</tt></i></em>, executing
    <em><i><tt>code_block</tt></i></em> once for each row with variables set to
    column values (or a set or array populated if <tt>-column_array</tt> or
                   <tt>column_set</tt> is specified). If the query returns no rows, executes
    <em><i><tt>if_no_rows_block</tt></i></em> (if provided). In place of 'if_no_rows' also the 'else' keyword can be used.</p>

    <p>Example:

    <blockquote><pre>db_foreach greeble_query "select foo, bar from greeble" {
        ns_write "&lt;li&gt;foo=$foo; bar=$bar\n"
    } if_no_rows {
        # This block is optional.
        ns_write "&lt;li&gt;No greebles!\n"
    }</pre></blockquote>

    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    ad_arg_parser { bind column_array column_set args } $args

    # Do some syntax checking.
    set arglength [llength $args]
    if { $arglength == 1 } {
        # Have only a code block.
        set code_block [lindex $args 0]
    } elseif$arglength == 3 } {
        # Should have code block + if_no_rows + code block.
        if { [lindex $args 1] ni {"if_no_rows" "else"}} {
            return -code error "Expected if_no_rows or else as second-to-last argument"
        }
        lassign $args code_block . if_no_rows_code_block
    } else {
        return -code error "Expected 1 or 3 arguments after switches"
    }

    if { [info exists column_array] && [info exists column_set] } {
        return -code error "Can't specify both column_array and column_set"
    }

    if { [info exists column_array] } {
        upvar 1 $column_array array_val
    }

    if { [info exists column_set] } {
        upvar 1 $column_set selection
    }

    set bindArg [expr {[info exists bind] ? [list -bind $bind] : ""}]
    set counter 0
    set result [uplevel [list db_list_of_lists \
                             -with_headers \
                             -dbn $dbn \
                             -subst $subst \
                             $statement_name \
                             $sql \
                             {*}${bindArg}]]
    #ns_log notice "RESULT $result"
    set columns [lindex $result 0]
    #ns_log notice "columns <$columns>"
    foreach tuple [lrange $result 1 end] {
        incr counter

        #
        # Result will be provided in different formats to the code
        # block depending on the flags...
        #
        if { [info exists column_set] } {
            #
            # ns_set
            #
            if { [info exists selection] } {
               ns_set free $selection
            }
            set selection [ns_set create]
            foreach a $columns v $tuple { ns_set put $selection $a $v }
        } elseif { [info exists column_array] } {
            #
            # array
            #
            unset -nocomplain array_val
            array set array_val [concat {*}[lmap a $columns v $tuple {list $a $v}]]
        } else {
            #
            # plain variables
            #
            foreach a $columns v $tuple { uplevel [list set $a $v] }
        }

        set errno [catch { uplevel 1 $code_block } error]

        #
        # Handle or propagate the error.
        #
        switch -- $errno {
            0 {
                # TCL_OK
            }
            1 {
                # TCL_ERROR
                error $error $::errorInfo $::errorCode
            }
            2 {
                # TCL_RETURN
                error "Cannot return from inside a db_foreach loop"
            }
            3 {
                # TCL_BREAK
                break
            }
            4 {
                # TCL_CONTINUE - just ignore and continue looping.
            }
            default {
                error "Unknown return code: $errno"
            }
        }
    }
    # If the if_no_rows_code is defined, go ahead and run it.
    if { $counter == 0 && [info exists if_no_rows_code_block] } {
        uplevel 1 $if_no_rows_code_block
    }
}

ad_proc -private db_multirow_helper {} {

    Helper function for db_multirow, performing the actual DB queries.

} {
    uplevel 1 {
        if { !$append_p || ![info exists counter]} {
            set counter 0
        }

        set local_counter -1
        #
        # Make sure 'next_row' dict doesn't exist.
        #
        # The variables 'this_row' and 'next_row' are used to always
        # execute the code block one result set row behind, so that we
        # have the opportunity to peek ahead, which allows us to do
        # group by's inside the multirow generation.
        #
        # Also make the 'next_row' dict available as a magic __db_multirow__next_row variable
        #
        upvar 1 __db_multirow__next_row next_row
        unset -nocomplain next_row

        #
        # Execute the query in one sweep, similar to 'db_foreach'.
        #
        upvar 1 __db_multirow__local_columns local_columns
        set __selections [uplevel 1 [list db_list_of_lists -dbn $dbn \
                                         -subst $subst \
                                         -columns_var __db_multirow__local_columns \
                                         $full_statement_name $sql]]

        lappend local_columns {*}$extend

        if { !$append_p || ![info exists columns] } {
            # store the list of columns in the var_name:columns variable
            set columns $local_columns
        } else {
            # Check that the columns match, if not throw an error
            if { [join [lsort -ascii $local_columns]] ne [join [lsort -ascii $columns]] } {
                error "Appending to a multirow with differing columns.
    Original columns     : [join [lsort -ascii $columns] ""].
    Columns in this query: [join [lsort -ascii $local_columns] ""]" "" "ACS_MULTIROW_APPEND_COLUMNS_MISMATCH"
            }
        }

        if {[llength $__selections] == 0} {
            return
        }

        set more_rows_p 1
        while { 1 } {
            incr local_counter

            if { $more_rows_p } {
                set more_rows_p [expr {$local_counter < [llength $__selections]}]
                set selection [lindex $__selections $local_counter]
            } else {
                break
            }
            #ns_log notice "$local_counter: $selection"

            #
            # Setup the 'columns' part, now that we know the columns
            # in the result set in the first iteration (when
            # $local_counter == 0).
            #
            if { $local_counter == 0 } {

                # In case the '-unclobber' switch is specified, save
                # variables which we might clobber.
                #
                if { $unclobber_p && $code_block ne "" } {
                    foreach col $columns {
                        upvar 1 $col column_value __saved_$col column_save

                        if { [info exists column_value] } {
                            if { [array exists column_value] } {
                                array set column_save [array get column_value]
                            } else {
                                set column_save $column_value
                            }

                            # Clear the variable
                            unset column_value
                        }
                    }
                }
            }

            if { $code_block eq "" } {
                #
                # There is no code block - pull values directly into
                # the var_name array.
                #
                # The extra loop after the last row is only for when
                # there's a code block.
                #
                if { !$more_rows_p } {
                    break
                }

                incr counter
                upvar $level_up "$var_name:$counter" array_val
                set array_val(rownum) $counter
                array set array_val [join [lmap __column $local_columns __value $selection {
                    list $__column $__value
                }]]
            } else {
                #
                # There is a code block to execute.
                # Copy next_row to this_row, if it exists
                #
                unset -nocomplain this_row
                if {[info exists next_row]} {
                    set this_row $next_row
                }

                # Pull values from the query into next_row
                unset -nocomplain next_row
                if { $more_rows_p } {
                    set next_row $selection
                }

                # Process the row
                if { [info exists this_row] } {
                    # Pull values from this_row into local variables
                    foreach name $local_columns __value $this_row {
                        upvar 1 $name column_value
                        set column_value $__value
                        # ns_log notice "... [list set $name $__value]"
                    }

                    # Initialize the "extend" columns to the empty string
                    foreach column_name $extend {
                        upvar 1 $column_name column_value
                        set column_value ""
                    }

                    # Execute the code block
                    set errno [catch { uplevel 1 $code_block } error]
                    #ns_log notice ".... code block returns errno $errno"

                    # Handle or propagate the error. Can't use the usual
                    # "return -code $errno..." trick due to the db_with_handle
                    # wrapped around this loop, so propagate it explicitly.
                    #
                    switch -- $errno {
                        0 {
                            # TCL_OK
                        }
                        1 {
                            # TCL_ERROR
                            error $error $::errorInfo $::errorCode
                        }
                        2 {
                            # TCL_RETURN
                            error "Cannot return from inside a db_multirow loop"
                        }
                        3 {
                            # TCL_BREAK
                            #### CHECK? #ns_db flush $db
                            break
                        }
                        4 {
                            # TCL_CONTINUE
                            continue
                        }
                        default {
                            error "Unknown return code: $errno"
                        }
                    }

                    # Pull the local variables back out and into the array.
                    incr counter
                    upvar $level_up "$var_name:$counter" array_val
                    set array_val(rownum) $counter
                    foreach column_name $columns {
                        upvar 1 $column_name column_value
                        set array_val($column_name$column_value
                    }
                }
            }
        }

        # Restore values of columns which we've saved
        if { $unclobber_p && $code_block ne "" && $local_counter > 0 } {
            foreach col $columns {
                upvar 1 $col column_value __saved_$col column_save

                # Unset it first, so the road's paved to restoring
                unset -nocomplain column_value

                # Restore it
                if { [info exists column_save] } {
                    if { [array exists column_save] } {
                        array set column_value [array get column_save]
                    } else {
                        set column_value $column_save
                    }

                    # And then remove the saved col
                    unset column_save
                }
            }
        }
        # Unset the next_row variable, just in case
        unset -nocomplain next_row
    }
}


d_proc -public db_multirow {
    -local:boolean
    -append:boolean
    {-upvar_level 1}
    -unclobber:boolean
    {-extend {}}
    {-dbn ""}
    -cache_key
    {-cache_pool db_cache_pool}
    {-subst all}
    var_name
    statement_name
    sql
    args
} {
    <p>Performs the SQL query <code>sql</code>, saving results in variables
    of the form
    <code><i>var_name</i>:1</code>, <code><i>var_name</i>:2</code>, etc,
    setting <code><i>var_name</i>:rowcount</code> to the total number
    of rows, and setting <code><i>var_name</i>:columns</code> to a
    list of column names.

    Usage:
    
    <blockquote>
    db_multirow [ -local ] [ -upvar_level <em><i>n_levels_up</i></em> ] [ -append ] [ -extend <em><i>column_list</i></em> ] \
        <em><i>var-name statement-name sql</i></em> [ -bind <em><i>bind_set_id</i></em> | -bind <em><i>bind_value_list</i></em> ] \
        <em><i>code_block</i></em> [ if_no_rows <em><i>if_no_rows_block</i> ]</em>

    </blockquote>


    <p>

    If "cache_key" is set, cache the array that results from the query *and*
    any code block for future use.  When this result is returned from cache,
    THE CODE BLOCK IS NOT EXECUTED.  Therefore, any values calculated by the
    code block that aren't listed as arguments to "extend" will
    not be created.  In practice this impacts relatively few queries, but do
    take care.

    <p>

    You can not simultaneously append to and cache a nonempty multirow.

    <p>

    Each row also has a column, rownum, automatically
    added and set to the row number, starting with 1. Note that this will
    override any column in the SQL statement named 'rownum', also if you're
    using the Oracle rownum pseudo-column.

    <p>

    If the <code>-local</code> is passed, the variables defined by
    db_multirow will be set locally (useful if you're compiling
    dynamic templates in a function or similar situations). Use the
    <code>-upvar_level</code> switch to specify how many levels up the
    variable should be set.

    The default behavior (i.e., when no "-local" is specified) depends
    on the calling environment: when "db_multirow" is called from an
    ADP file the variables are set in the ADP environment.  Otherwise,
    the default behavior is "-local".

    <p>

    You may supply a code block, which will be executed for each row in
    the loop. This is very useful if you need to make computations that
    are better done in Tcl than in SQL, for example using ns_urlencode
    or ns_quotehtml, etc. When the Tcl code is executed, all the columns
    from the SQL query will be set as local variables in that code. Any
    changes made to these local variables will be copied back into the
    multirow.

    <p>

    You may also add additional, computed columns to the multirow, using the
    <code>-extend { <i>col_1</i> <i>col_2</i> ... }</code> switch. This is
    useful for things like constructing a URL for the object retrieved by
    the query.

    <p>

    If you're constructing your multirow through multiple queries with the
    same set of columns, but with different rows, you can use the
    <code>-append</code> switch. This causes the rows returned by this query
    to be appended to the rows already in the multirow, instead of starting
    a clean multirow, as is the normal behavior. The columns must match the
    columns in the original multirow, or an error will be thrown.

    <p>

    Your code block may call <code>continue</code> in order to skip a row
    and not include it in the multirow. Or you can call <code>break</code>
    to skip this row and quit looping.

    <p>

    Notice the nonstandard numbering (everything else in Tcl starts at
    0); the reason is that the graphics designer, a non-programmer,
    may wish to work with row numbers.

    <p>

    Example:
    <pre>db_multirow -extend { user_url } users users_query {
        select user_id first_names, last_name, email from cc_users
    } {
        set user_url [acs_community_member_url -user_id $user_id]
    }</pre>

    @param dbn The database name to use.  If empty_string, uses the default database.
    @param cache_key Cache the result using given value as the key.  Default is to not cache.
    @param cache_pool Override the default db_cache_pool
    @param subst Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands

    @param unclobber If set, will cause the proc to not overwrite local variables. Actually, what happens
    is that the local variables will be overwritten, so you can access them within the code block. However,
    if you specify -unclobber, we will revert them to their original state after execution of this proc.

    @param statement_name name of the SQL query
    @param sql SQL query to be executed
    @param var_name name of the Tcl multirow array
    @param code_block code block to be executed for every tuple reurned

    @see template::multirow
} {
    # Query Dispatcher (OpenACS - ben)
    set full_statement_name [db_qd_get_fullname $statement_name]

    #
    # When this function is called outside ADP, fall back to "-local"
    # behavior.
    #
    set adpLevel [template::adp_level]

    if { $local_p || $adpLevel eq ""} {
        set level_up $upvar_level
    } else {
        set level_up \#$adpLevel
    }

    ad_arg_parser { bind args } $args

    # Do some syntax checking.
    set arglength [llength $args]
    if { $arglength == 0 } {
        # No code block.
        set code_block ""
    } elseif$arglength == 1 } {
        # Have only a code block.
        set code_block [lindex $args 0]
    } elseif$arglength == 3 } {
        # Should have code block + if_no_rows + code block.
        if { [lindex $args 1] ne "if_no_rows"
             && [lindex $args 1] ne "else"
         } {
            return -code error "Expected if_no_rows as second-to-last argument"
        }
        lassign $args code_block . if_no_rows_code_block
    } else {
        return -code error "Expected 1 or 3 arguments after switches"
    }

    upvar $level_up "$var_name:rowcount" counter
    upvar $level_up "$var_name:columns" columns

    if { [info exists cache_key]
         && $append_p
         && [info exists counter] && $counter > 0
     } {
        return -code error "Can't append and cache a nonempty multirow datasource simultaneously"
    }

    if { [info exists cache_key] } {
        #
        # Call helper with cache key
        #
        set value [ns_cache eval $cache_pool $cache_key {
            db_multirow_helper

            set values [list]

            for { set count 1 } { $count <= $counter } { incr count } {
                upvar $level_up "$var_name:[expr {$count}]" array_val
                lappend values [array get array_val]
            }

            return [list $counter $columns $values]
        }]

        lassign $value counter columns values

        set count 1
        foreach value $values {
            upvar $level_up "$var_name:[expr {$count}]" array_val
            array set array_val $value
            incr count
        }
    } else {
        #
        # Call helper without cache key
        #
        db_multirow_helper
    }

    # If the if_no_rows_code is defined, go ahead and run it.
    if { $counter == 0 && [info exists if_no_rows_code_block] } {
        uplevel 1 $if_no_rows_code_block
    }
}

d_proc -public db_multirow_group_last_row_p {
    {-column:required}
} {
    Used inside the code_block to db_multirow to ask whether this row is the last row
    before the value of 'column' changes, or the last row of the result set.

    <p>

    This is useful when you want to build up a multirow for a master/slave table pair,
    where you only want one row per row in the master table, but you want to include
    data from the slave table in a column of the multirow.

    <p>

    Here's an example:

    <pre>
    # Initialize the lines variable to hold a list of order line summaries
    set lines [list]

    # Start building the multirow. We add the dynamic column 'lines_pretty', which will
    # contain the pretty summary of the order lines.
    db_multirow -extend { lines_pretty } orders select_orders_and_lines {
        select o.order_id,
        o.customer_name,
        l.item_name,
        l.quantity
        from   orders o,
        order_lines l
        where  l.order_id = o.order_id
        order  by o.order_id, l.item_name
    } {
        lappend lines "$quantity $item_name"
        if { [db_multirow_group_last_row_p -column order_id] } {
            # Last row of this order, prepare the pretty version of the order lines
            set lines_pretty [join $lines ", "]

            # Reset the lines list, so we start from a fresh with the next row
            set lines [list]
        } else {
            # There are yet more order lines to come for this order,
            # continue until we've collected all the order lines
            # The 'continue' keyword means this line will not be added to the resulting multirow
            continue
        }
    }
    </pre>

    @author Lars Pind (lars@collaboraid.biz)

    @param column The name of the column defining the groups.

    @return 1 if this is the last row before the column value changes, 0 otherwise.
} {
    upvar 1 __db_multirow__next_row next_row
    if { ![info exists next_row] } {
        # If there is no next row, this is the last row
        return 1
    }
    upvar 1 __db_multirow__local_columns columns
    upvar 1 $column column_value
    set pos [lsearch $columns $column]
    if {$pos == -1} {
        error "column '$column' not found in columns list '$columns'"
    }
    # Otherwise, it's the last row in the group if the next row has a
    # different value than this row
    set next_value [lindex $next_row $pos]
    return [expr {$next_value ne $column_value}]
}


d_proc -public db_dml {
    {-dbn ""}
    {-subst all}
    statement_name
    sql
    args
} {
    Do a DML statement.

    <p>

    args can be one of: -clobs, -blobs, -clob_files or -blob_files.
    See the db-api doc referenced below for more information.

    @param dbn The database name to use.  If empty_string, uses the default database.
    @param subst Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands

    @see <a href="/doc/db-api-detailed">/doc/db-api-detailed</a>
} {
    ad_arg_parser { clobs blobs clob_files blob_files bind } $args
    set driverkey [db_driverkey $dbn]

    switch -- $driverkey {
        postgresql {
            set postgres_p 1
        }
        oracle -
        nsodbc -
        default {
            set postgres_p 0
        }
    }

    # Query Dispatcher (OpenACS - ben)
    set full_statement_name [db_qd_get_fullname $statement_name]

    # This "only one of..." check didn't exist in the PostgreSQL
    # version, but it shouldn't't hurt anything: --atp@piskorski.com,
    # 2003/04/08 06:19 EDT

    # Only one of clobs, blobs, clob_files, and blob_files is allowed.
    # Remember which one (if any) is provided:

    set lob_argc 0
    set lob_argv [list]
    set command "dml"
    if { [info exists clobs] } {
        set command "clob_dml"
        set lob_argv $clobs
        incr lob_argc
    }
    if { [info exists blobs] } {
        set command "blob_dml"
        set lob_argv $blobs
        incr lob_argc
    }
    if { [info exists clob_files] } {
        set command "clob_dml_file"
        set lob_argv $clob_files
        incr lob_argc
    }
    if { [info exists blob_files] } {
        set command "blob_dml_file"
        set lob_argv $blob_files
        incr lob_argc
    }
    if { $lob_argc > 1 } {
        error "Only one of -clobs, -blobs, -clob_files, or -blob_files may be specified as an argument to db_dml"
    }

    if { ! $postgres_p } {
        # Oracle:
        db_with_handle -dbn $dbn db {
            if { $lob_argc == 1 } {
                # Bind :1, :2, ..., :n as LOBs (where n = [llength $lob_argv])
                set bind_vars [list]
                for { set i 1 } { $i <= [llength $lob_argv] } { incr i } {
                    lappend bind_vars $i
                }
                eval [list db_exec -subst $subst "${command}_bind" $db $full_statement_name $sql 2 $bind_vars$lob_argv
            } else {
                eval [list db_exec -subst $subst $command $db $full_statement_name $sql$lob_argv
            }
        }

    } elseif {$command eq "blob_dml_file"} {
        # PostgreSQL:
        db_with_handle -dbn $dbn db {
            # another ugly hack to avoid munging Tcl files.
            # __lob_id needs to be set inside of a query (.xql) file for this
            # to work.  Say for example that you need to create a lob. In
            # Oracle, you would do something like:

            # db_dml update_photo  "update foo set bar = empty_blob()
            #                       where bar = :bar
            #                       returning foo into :1" -blob_files [list $file]
            # for PostgreSQL we can do the equivalent by placing the following
            # in a query file:
            # update foo set bar = [set __lob_id [db_string get_id "select empty_lob()"]]
            # where bar = :bar

            # __lob_id acts as a flag that signals that blob_dml_file is
            # required, and it is also used to pass along the lob_id.  It
            # is unsert afterwards to avoid name clashes with other invocations
            # of this routine.
            # (DanW - Openacs)

            db_exec -subst $subst dml $db $full_statement_name $sql
            if {[uplevel {info exists __lob_id}]} {
                ns_pg blob_dml_file $db [uplevel {set __lob_id}] $blob_files
                uplevel {unset __lob_id}
            }
        }

    } else {
        # PostgreSQL:
        db_with_handle -dbn $dbn db {
            db_exec -subst $subst dml $db $full_statement_name $sql
        }
    }
}




d_proc -public db_0or1row {
    {-dbn ""}
    -cache_key
    {-cache_pool db_cache_pool}
    {-subst all}
    statement_name
    sql
    -bind
    -column_array
    -column_set
} {

    Performs the specified SQL query. If a row is returned, sets variables
    to column values (or a set or array populated if -column_array
                      or column_set is specified) and returns 1. 

    @return 1 if variables are set, 0 if no rows are returned.
              If more than one row is returned, throws an error.

    @param dbn The database name to use.  If empty_string, uses the default database.
    @param cache_key Cache the result using given value as the key.  Default is to not cache.
    @param cache_pool Override the default db_cache_pool
    @param subst Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands
    @param bind bind variables, passed either as an ns_set id, or via bind value list
    @param column_array array to be populated with values
    @param column_set ns_set to be populated with values
    @param statement_name name of the SQL query
    @param sql SQL query to be executed
} {
    # Query Dispatcher (OpenACS - ben)
    set full_statement_name [db_qd_get_fullname $statement_name]

    if { [info exists column_array] && [info exists column_set] } {
        return -code error "Can't specify both column_array and column_set"
    }

    if { [info exists column_array] } {
        upvar 1 $column_array array_val
        unset -nocomplain array_val
    }

    if { [info exists column_set] } {
        upvar 1 $column_set selection
    }

    if { [info exists cache_key] } {
        set values [ns_cache eval $cache_pool $cache_key {
            db_with_handle -dbn $dbn db {
                set selection [db_exec -subst $subst 0or1row $db $full_statement_name $sql]
            }

            set values [list]

            if { $selection ne "" } {
                set values [ns_set array $selection]
            }

            set values
        }]

        if { $values eq "" } {
            set selection ""
        } else {
            set selection [ns_set create s {*}$values]
        }
    } else {
        db_with_handle -dbn $dbn db {
            set selection [db_exec -subst $subst 0or1row $db $full_statement_name $sql]
        }
    }

    if { $selection eq "" } {
        return 0
    }

    if { [info exists column_array] } {
        array set array_val [ns_set array $selection]
    } elseif { ![info exists column_set] } {
        foreach {key value} [ns_set array $selection] {
            uplevel 1 [list set $key $value]
        }
    }

    return 1
}


ad_proc -public db_1row { {-subst all} args } {

    A wrapper for db_0or1row, which produces an error if no rows are returned.

    @param args Arguments to be passed to db_0or1row. Check db_0or1row proc doc
                for details.

    @see db_0or1row

    @return 1 if variables are set, otherwise an exception is thrown.

} {
    if { ![uplevel ::db_0or1row -subst $subst $args] } {
        return -code error "Query did not return any rows."
    }
}

if {[namespace which ns_cache_transaction_begin] eq ""} {
    #
    # When the server has no support for ns_cache_transaction_*,
    # provide dummy procs to avoid run time "if" statements.
    #
    proc ns_cache_transaction_begin args {;}
    proc ns_cache_transaction_commit args {;}
    proc ns_cache_transaction_rollback args {;}
}

ad_proc -public db_transaction {{ -dbn ""} transaction_code args } {
    Usage: <b><i>db_transaction</i></b> <i>transaction_code</i> [ on_error { <i>error_code_block</i> } ]

    Executes transaction_code with transactional semantics.  This
    means that either all of the database commands within
    transaction_code are committed to the database or none of them
    are.  Multiple <code>db_transaction</code>s may be nested (end
    transaction is transparently ns_db dml'ed when the outermost
    transaction completes).<p>

    To handle errors, use <code>db_transaction {transaction_code}
    on_error {error_code_block}</code>.  Any error generated in
    <code>transaction_code</code> will be caught automatically and
    process control will transfer to <code>error_code_block</code>
    with a variable <code>errmsg</code> set.  The error_code block can
    then clean up after the error, such as presenting a usable error
    message to the user.  Following the execution of
    <code>error_code_block</code> the transaction will be aborted.  If
    you want to explicitly abort the transaction, call
    <code>db_abort_transaction</code> from within the transaction_code
    block or the error_code block.<p>

    Example 1:<br> In this example, db_dml triggers an error, so
    control passes to the on_error block which prints a readable
    error.

    <pre>
    db_transaction {
        db_dml test "nonsense"
    } on_error {
        ad_return_error "Error in blah/foo/bar" "The error was: $errmsg"
    }
    </pre>

    Example 2:<br> In this example, the second command, "nonsense"
    triggers an error.  There is no on_error block, so the transaction
    is immediately halted and aborted.

    <pre>
    db_transaction {
        db_dml test {insert into footest values(1)}
        nonsense
        db_dml test {insert into footest values(2)}
    }
    </pre>

    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    upvar "#0" [db_state_array_name_is -dbn $dbn] db_state

    set syn_err "db_transaction: Invalid arguments. Use db_transaction { code } \[on_error { error_code_block }\] "
    set arg_c [llength $args]

    if { $arg_c != 0 && $arg_c != 2 } {
        # Either this is a transaction with no error handling or there
        # must be an on_error { code } block.
        error $syn_err
    }  elseif$arg_c == 2 } {
        # We think they're specifying an on_error block
        if {[lindex $args 0] ne "on_error"  } {
            # Unexpected: they put something besides on_error as a
            # connector.
            error $syn_err
        } else {
            # Success! We got an on_error code block.
            set on_error [lindex $args 1]
        }
    }
    # Make the error message and database handle available to the
    # on_error block.
    upvar errmsg errmsg

    db_with_handle -dbn $dbn db {
        # Preserve the handle, since db_with_handle kills it after
        # executing this block.
        set dbh $db
        # Remember that there's a transaction happening on this handle.
        if { ![info exists db_state(transaction_level,$dbh)] } {
            set db_state(transaction_level,$dbh) 0
        }
        set level [incr db_state(transaction_level,$dbh)]
        if { $level == 1 } {
            ns_db dml $dbh "begin transaction"
            ns_cache_transaction_begin
        }
    }
    # Execute the transaction code.
    set errno [catch {
        uplevel 1 $transaction_code
    } errmsg]
    incr db_state(transaction_level,$dbh) -1

    set err_p 0
    switch -- $errno {
        0 {
            # TCL_OK
        }
        2 {
            # TCL_RETURN
        }
        3 {
            # TCL_BREAK - Abort the transaction and do the break.
            ns_db dml $dbh "abort transaction"
            ns_cache_transaction_rollback
            db_release_unused_handles -dbn $dbn
            break
        }
        4 {
            # TCL_CONTINUE - just ignore.
        }
        default {
            # TCL_ERROR or unknown error code: Its a real error.
            set err_p 1
        }
    }

    if { $err_p || [db_abort_transaction_p -dbn $dbn]} {
        # An error was triggered or the transaction has been aborted.
        db_abort_transaction -dbn $dbn

        if { [info exists on_error] && $on_error ne "" } {

            if {"postgresql" eq [db_type]} {

                # JCD: with postgres we abort the transaction prior to
                # executing the on_error block since there is nothing
                # you can do to "fix it" and keeping it meant things like
                # queries in the on_error block would then fail.
                #
                # Note that the semantics described in the proc doc
                # are not possible to support on PostgreSQL.

                # DRB: I removed the db_release_unused_handles call that
                # this patch included because additional aborts further
                # down triggered an illegal db handle error.  I'm going to
                # have the code start a new transaction as well.  If we
                # don't, if a transaction fails and the on_error block
                # fails, the on_error block DML will have been committed.
                # Starting a new transaction here means that DML by both
                # the transaction and on_error clause will be rolled back.
                # On the other hand, if the on_error clause doesn't fail,
                # any DML in that block will be committed.  This seems more
                # useful than simply punting ...

                ns_db dml $dbh "abort transaction"
                ns_cache_transaction_rollback
                ns_db dml $dbh "begin transaction"
                ns_cache_transaction_begin

            }

            # An on_error block exists, so execute it.

            set errno  [catch {
                uplevel 1 $on_error
            } on_errmsg]

            # Determine what do with the error.
            set err_p 0
            switch -- $errno {
                0 {
                    # TCL_OK
                }

                2 {
                    # TCL_RETURN
                }
                3 {
                    # TCL_BREAK
                    ns_db dml $dbh "abort transaction"
                    ns_cache_transaction_rollback
                    db_release_unused_handles
                    break
                }
                4 {
                    # TCL_CONTINUE - just ignore.
                }
                default {
                    # TCL_ERROR or unknown error code: Its a real error.
                    set err_p 1
                }
            }

            if { $err_p } {
                # An error was generated from the $on_error block.
                if { $level == 1} {
                    # We're at the top level, so we abort the transaction.
                    set db_state(db_abort_p,$dbh) 0
                    ns_db dml $dbh "abort transaction"
                    ns_cache_transaction_rollback
                }
                #
                # We throw this error because it was thrown from the
                # error handling code that the programmer must fix.
                #
                error $on_errmsg $::errorInfo $::errorCode
            } else {
                # Good, no error thrown by the on_error block.
                if { [db_abort_transaction_p -dbn $dbn] } {
                    # This means we should abort the transaction.
                    if { $level == 1 } {
                        set db_state(db_abort_p,$dbh) 0
                        ns_db dml $dbh "abort transaction"
                        ns_cache_transaction_rollback
                        #
                        # We still have the transaction generated
                        # error.  We don't want to throw it, so we log
                        # it, unless it is "rollback tests"
                        #
                        if {$errmsg ne "rollback tests"} {
                            ns_log Error "Aborting transaction due to error:\n$errmsg"
                        }
                    } else {
                        # Propagate the error up to the next level.
                        error $errmsg $::errorInfo $::errorCode
                    }
                } else {
                    #
                    # The on_error block has resolved the transaction
                    # error.  If we're at the top, commit and exit.
                    # Otherwise, we continue on through the lower
                    # transaction levels.
                    #
                    if { $level == 1} {
                        ns_db dml $dbh "end transaction"
                        ns_cache_transaction_commit
                    }
                }
            }
        } else {
            # There is no on_error block, yet there is an error, so we propagate it.
            if { $level == 1 } {
                set db_state(db_abort_p,$dbh) 0
                ns_db dml $dbh "abort transaction"
                ns_cache_transaction_rollback
                error "Transaction aborted: $errmsg" $::errorInfo $::errorCode
            } else {
                db_abort_transaction -dbn $dbn
                error $errmsg $::errorInfo $::errorCode
            }
        }
    } else {
        # There was no error from the transaction code.
        if { [db_abort_transaction_p -dbn $dbn] } {
            # The user requested the transaction be aborted.
            if { $level == 1 } {
                set db_state(db_abort_p,$dbh) 0
                ns_db dml $dbh "abort transaction"
                ns_cache_transaction_rollback
            }
        } elseif$level == 1 } {
            # Success!  No errors and no requested abort.  Commit.
            ns_db dml $dbh "end transaction"
            ns_cache_transaction_commit
        }
    }
}


ad_proc -public db_abort_transaction {{-dbn ""}} {

    Aborts all levels of a transaction. That is if this is called within
    several nested transactions, all of them are terminated. Use this
    instead of db_dml "abort" "abort transaction".

    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    upvar "#0" [db_state_array_name_is -dbn $dbn] db_state

    db_with_handle -dbn $dbn db {
        # We set the abort flag to true.
        set db_state(db_abort_p,$db) 1
    }
}


ad_proc -private db_abort_transaction_p {{-dbn ""}} {
    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    upvar "#0" [db_state_array_name_is -dbn $dbn] db_state

    db_with_handle -dbn $dbn db {
        if { [info exists db_state(db_abort_p,$db)] } {
            return $db_state(db_abort_p,$db)
        } else {
            # No abort flag registered, so we assume everything is ok.
            return 0
        }
    }
}


ad_proc -public db_name {{-dbn ""}} {

    @return the name of the database as reported by the driver.

    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    db_with_handle -dbn $dbn db {
        set dbtype [ns_db dbtype $db]
    }
    return $dbtype
}


ad_proc -public db_get_username {{-dbn ""}} {
    @return the username parameter from the driver section of the
    first database pool for the dbn.

    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    set pool [lindex [db_available_pools $dbn] 0]
    return [ns_config "ns/db/pool/$pool" User]
}

ad_proc -public db_get_password {{-dbn ""}} {
    @return the password parameter from the driver section of the
    first database pool for the dbn.

    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    set pool [lindex [db_available_pools $dbn] 0]
    return [ns_config "ns/db/pool/$pool" Password]
}

ad_proc -public db_get_sql_user {{-dbn ""}} {
    <strong>Oracle only.</strong>

    <p>
    @return a valid Oracle user@database/password string to access a
    database through sqlplus.

    <p>
    This proc may well <em>work</em> for databases other than Oracle,
    but its return value won't really be of any use.

    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    set pool [lindex [db_available_pools $dbn] 0]
    set datasource [ns_config "ns/db/pool/$pool" DataSource]
    if { $datasource ne "" && ![string is space $datasource] } {
        return "[ns_config ns/db/pool/$pool User]/[ns_config ns/db/pool/$pool Password]@$datasource"
    } else {
        return "[ns_config ns/db/pool/$pool User]/[ns_config ns/db/pool/$pool Password]"
    }
}

ad_proc -public db_get_pgbin {{-dbn ""}} {
    <strong>PostgreSQL only.</strong>

    <p>
    @return the pgbin parameter from the driver section of the first database pool.

    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    #
    # First, we try to get the postgres folder from the conf.
    #
    set pool [lindex [db_available_pools $dbn] 0]
    set driver [ns_config "ns/db/pool/$pool" Driver]
    set pgbin [ns_config "ns/db/driver/$driver" pgbin]

    if {$pgbin eq ""} {
        #
        # When the pgbin conf is missing, we guess the folder from the
        # psql location.
        #
        set pgbin [file dirname [util::which psql]]
    }

    return $pgbin
}


ad_proc -public db_get_port {{-dbn ""}} {
    <strong>PostgreSQL only.</strong>

    <p>
    @return the port number from the first database pool.  It assumes the
    datasource is properly formatted since we've already verified that we
    can connect to the pool.
    It returns an empty string for an empty port value.

    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    set pool [lindex [db_available_pools $dbn] 0]
    set datasource [ns_config "ns/db/pool/$pool" DataSource]
    set last_colon_pos [string last ":" $datasource]
    if { $last_colon_pos == -1 } {
        ns_log Error "datasource contains no \":\"? datasource = $datasource"
        return ""
    }
    set first_colon_pos [string first ":" $datasource]

    if { $first_colon_pos == $last_colon_pos || ($last_colon_pos - $first_colon_pos) == 1 } {
        # No port specified
        return ""
    }

    return [string range $datasource $first_colon_pos+1 $last_colon_pos-1]
}


ad_proc -public db_get_database {{-dbn ""}} {
    <strong>PostgreSQL and NSDB only.</strong>

    Return the database name from the first database pool.  It assumes
    the datasource is properly formatted since we've already verified
    that we can connect to the pool.

    On the longer range, it might be better to use SQL queries, at
    least in cases, where database is already connected.

    PostgreSQL:

        SELECT current_database()

    Oracle:

        SELECT name from v$database;
        SELECT ora_database_name FROM dual

    @param dbn The database name to use.  If empty_string, uses the default database.
    @return database name
} {
    set pool [lindex [db_available_pools $dbn] 0]
    set datasource [ns_config "ns/db/pool/$pool" DataSource]
    set last_colon_pos [string last ":" $datasource]
    if { $last_colon_pos == -1 } {
        ns_log Error "datasource contains no \":\"? datasource = $datasource"
        return ""
    }
    set dbname [string range $datasource $last_colon_pos+1 end]
    set equal_pos [string first "=" $dbname]
    if {$equal_pos > -1} {
        # The value after the last colon is a
        regexp {dbname=([^ ]+) ?} $dbname . dbname
    }
    return $dbname
}


d_proc -public db_get_dbhost {
    {-dbn ""}
} {
    <strong>PostgreSQL only.</strong>

    <p>
    @return the name of the database host from the first database pool.
    It assumes the datasource is properly formatted since we've already
    verified that we can connect to the pool.

    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    set pool [lindex [db_available_pools $dbn] 0]
    set datasource [ns_config "ns/db/pool/$pool" DataSource]
    set first_colon_pos [string first ":" $datasource]
    if { $first_colon_pos == -1 } {
        ns_log Error "datasource contains no \":\"? datasource = $datasource"
        return ""
    }
    return [string range $datasource 0 $first_colon_pos-1]
}

d_proc -public db_source_sql_file {
    {-dbn ""}
    {-callback apm_ns_write_callback}
    file
} {
    Sources a SQL file into Oracle (SQL*Plus format file) or
    PostgreSQL (psql format file).

    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    set proc_name {db_source_sql_file}
    set driverkey [db_driverkey $dbn]

    switch -- $driverkey {

        oracle {
            set user_pass [db_get_sql_user -dbn $dbn]
            cd [ad_file dirname $file]
            set fp [open "|[ad_file join $::env(ORACLE_HOME) bin sqlplus] $user_pass @$file" "r+"]
            fconfigure $fp -buffering line
            puts $fp "exit"

            while { [gets $fp line] >= 0 } {
                # Don't bother writing out lines which are purely whitespace.
                if { ![string is space $line] } {
                    apm_callback_and_log $callback "[ns_quotehtml $line]\n"
                }
            }
            close $fp
        }

        postgresql {
            set file_name [ad_file tail $file]

            set pguser [db_get_username]
            if { $pguser ne "" } {
                set pguser "-U $pguser"
            }

            set pgport [db_get_port]
            if { $pgport ne "" } {
                set pgport "-p $pgport"
            }

            set pgpass [db_get_password]
            if { $pgpass ne "" } {
                set pgpass "<<$pgpass"
            }

            #
            # GN: windows requires $pghost "-h ..."
            #
            if { ([db_get_dbhost] eq "localhost" || [db_get_dbhost] eq "")
                 && $::tcl_platform(platform) ne "windows"
             } {
                set pghost ""
            } else {
                set pghost "-h [db_get_dbhost]"
            }

            set dir [ad_file dirname $file]
            set cmd "[file join [db_get_pgbin] psql] \
                         $pghost $pgport $pguser \
                         -f $file \
                         [db_get_database$pgpass"

            try {
                if {[info commands proxy::exec] ne ""} {
                    ns_log notice [list ::proxy::exec -call $cmd -cd $dir]
                    ::proxy::exec -call $cmd -cd $dir
                } {
                    cd $dir
                    set fp [open "|$cmd" r]
                    set result ""
                    catch {set result [read $fp]}
                    close $fp
                    set result
                }
            } on error {result} {
            } on ok {result} {
            }
            set error_found 0
            foreach line [split $result \n] {
                #
                # Don't bother writing out lines which are purely
                # whitespace.
                #
                if { ![string is space $line] } {
                    apm_callback_and_log $callback "[ns_quotehtml $line]\n"
                }
                #
                # PSQL dumps errors and notice information on
                # stderr, and has no option to turn this off.  So
                # we have to chug through the "error" lines
                # looking for those that really signal an error.
                #
                if { [string first NOTICE $line] == -1 } {
                    append error_lines "$line\n"
                    set error_found [expr { $error_found
                                            || [string first ERROR $line] != -1
                                            || [string first FATAL $line] != -1 } ]
                }
            }
            ns_log notice "ERROR_FOUND=$error_found"
            if { $error_found } {
                return -code error -errorinfo $error_lines -errorcode $::errorCode $error_lines
            }
        }

        nsodbc {
            error "$proc_name is not supported for this database."
        }
        default {
            error "$proc_name is not supported for this database."
        }
    }
}

d_proc -public db_load_sql_data {
    {-dbn ""}
    {-callback apm_ns_write_callback}
    file
} {
    Loads a CSV formatted file into a table using PostgreSQL's COPY command or
    Oracle's SQL*Loader utility.  The filename format consists of a sequence
    number used to control the order in which tables are loaded, and the table
    name with "-" replacing "_".  This is a bit of a kludge but greatly speeds
    the loading of large amounts of data, such as is done when various "ref-*"
    packages are installed.

    @param dbn The database name to use.  If empty_string, uses the default database.
    @param file Filename in the format dd-table-name.ctl where 'dd' is a sequence number
    used to control the order in which data is loaded.  This file is an
    RDBMS-specific data loader control file.

} {

    switch [db_driverkey $dbn] {

        oracle {
            set user_pass [db_get_sql_user -dbn $dbn]

            set fd [open $file r]
            set file_contents [read $fd]
            close $fd
            set file_contents [subst $file_contents]

            set fd1 [file tempfile tmpnam [ns_config ns/parameters tmpdir]/oacs-XXXXXX.ctl]
            puts $fd1 $file_contents
            close $fd1

            cd [ad_file dirname $file]

            set fd [open "|[ad_file join $::env(ORACLE_HOME) bin sqlldr] userid=$user_pass control=$tmpnam" "r"]

            while { [gets $fd line] >= 0 } {
                # Don't bother writing out lines which are purely whitespace.
                if { ![string is space $line] } {
                    apm_callback_and_log $callback "[ns_quotehtml $line]\n"
                }
            }
            close $fd
        }

        postgresql {
            set pguser [db_get_username]
            if { $pguser ne "" } {
                set pguser "-U $pguser"
            }

            set pgport [db_get_port]
            if { $pgport ne "" } {
                set pgport "-p $pgport"
            }

            set pgpass [db_get_password]
            if { $pgpass ne "" } {
                set pgpass "<<$pgpass"
            }

            if { [db_get_dbhost] eq "localhost" || [db_get_dbhost] eq "" } {
                set pghost ""
            } else {
                set pghost "-h [db_get_dbhost]"
            }

            set fd [open $file r]
            set copy_command [subst -nobackslashes [read $fd]]
            close $fd
            set fd [file tempfile copy_file [ad_tmpdir]/psql-copyfile-XXXXXX]
            puts $fd $copy_command
            close $fd

            if { $::tcl_platform(platform) eq "windows" } {
                set fp [open "|[file join [db_get_pgbin] psql] -f $copy_file $pghost $pgport $pguser [db_get_database]" "r"]
            } else {
                set fp [open "|[file join [db_get_pgbin] psql] -f $copy_file $pghost $pgport $pguser [db_get_database$pgpass" "r"]
            }

            while { [gets $fp line] >= 0 } {
                # Don't bother writing out lines which are purely whitespace.
                if { ![string is space $line] } {
                    apm_callback_and_log $callback "[ns_quotehtml $line]\n"
                }
            }

            # PSQL dumps errors and notice information on stderr, and has no option to turn
            # this off.  So we have to chug through the "error" lines looking for those that
            # really signal an error.

            set errno [ catch {
                close $fp
            } error]

            # remove the copy file.
            file delete -force -- $copy_file

            if { $errno == 2 } {
                return $error
            }

            # Just filter out the "NOTICE" lines, so we get the stack dump along with real
            # ERRORs.  This could be done with a couple of opaque-looking regexps...

            set error_found 0
            foreach line [split $error "\n"] {
                if { [string first NOTICE $line] == -1 } {
                    append error_lines "$line\n"
                    set error_found [expr { $error_found
                                            || [string first ERROR $line] != -1
                                            || [string first FATAL $line] != -1 } ]
                }
            }

            if { $error_found } {
                return -code error -errorinfo $error_lines -errorcode $::errorCode $error_lines
            }

        }

        nsodbc {
            error "db_load_sql_data is not supported for this database."
        }
        default {
            error "db_load_sql_data is not supported for this database."
        }
    }
}

d_proc -public db_source_sqlj_file {
    {-dbn ""}
    {-callback apm_ns_write_callback}
    file
} {
    <strong>Oracle only.</strong>
    <p>
    Sources a SQLJ file using loadjava.

    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    set user_pass [db_get_sql_user -dbn $dbn]
    set fp [open "|[ad_file join $::env(ORACLE_HOME) bin loadjava] -verbose -user $user_pass $file" "r"]

    # Despite the fact that this works, the text does not get written to the stream.
    # The output is generated as an error when you attempt to close the input stream as
    # done below.
    while { [gets $fp line] >= 0 } {
        # Don't bother writing out lines which are purely whitespace.
        if { ![string is space $line] } {
            apm_callback_and_log $callback "[ns_quotehtml $line]\n"
        }
    }
    if { [catch {
        close $fp
    } errmsg] } {
        apm_callback_and_log $callback "[ns_quotehtml $errmsg]\n"
    }
}


d_proc -public db_tables {
    -pattern
    {-dbn ""}
} {
    @return a Tcl list of all the tables owned by the connected user.

    @param pattern Will be used as LIKE 'pattern%' to limit the number of tables returned.

    @param dbn The database name to use.  If empty_string, uses the default database.

    @author Don Baccus (dhogaza@pacifier.com)
    @author Lars Pind (lars@pinds.com)

    @change-log yon@arsdigita.com 20000711 changed to return lowercase table names
} {
    set proc_name {db_tables}
    set driverkey [db_driverkey $dbn]

    switch -- $driverkey {
        oracle {
            set sql_table_names_with_pattern {
                select lower(table_name) as table_name
                from user_tables
                where table_name like upper(:pattern)
            }
            set sql_table_names_without_pattern {
                select lower(table_name) as table_name
                from user_tables
            }
        }

        postgresql {
            set sql_table_names_with_pattern {
                select relname as table_name
                from pg_class
                where relname like lower(:pattern) and
                relname !~ '^pg_' and relkind = 'r'
            }
            set sql_table_names_without_pattern {
                select relname as table_name
                from pg_class
                where relname !~ '^pg_' and relkind = 'r'
            }
        }

        nsodbc -
        default {
            error "$proc_name is not supported for this database."
        }
    }

    set tables [list]
    if { [info exists pattern] } {
        db_foreach -dbn $dbn table_names_with_pattern \
            $sql_table_names_with_pattern {
                lappend tables $table_name
            }
    } else {
        db_foreach -dbn $dbn table_names_without_pattern \
            $sql_table_names_without_pattern {
                lappend tables $table_name
            }
    }

    return $tables
}


ad_proc -public db_table_exists {{-dbn ""} table_name } {
    @return 1 if a table with the specified name exists in the database, otherwise 0.

    @param dbn The database name to use.  If empty_string, uses the default database.

    @author Don Baccus (dhogaza@pacifier.com)
    @author Lars Pind (lars@pinds.com)
} {
    set proc_name {db_table_exists}
    set driverkey [db_driverkey $dbn]

    switch -- $driverkey {
        oracle {
            set n_rows [db_string -dbn $dbn table_count {
                select count(*) from user_tables
                where table_name = upper(:table_name)
            }]
        }

        postgresql {
            set n_rows [db_string -dbn $dbn table_count {
                select count(*) from pg_class
                where relname = lower(:table_name) and
                relname !~ '^pg_' and relkind = 'r'
            }]
        }

        nsodbc -
        default {
            error "$proc_name is not supported for this database."
        }
    }

    return $n_rows
}


ad_proc -public db_columns {{-dbn ""} table_name } {
    @return a Tcl list of all the columns in the table with the given name.

    @param dbn The database name to use.  If empty_string, uses the default database.

    @author Lars Pind (lars@pinds.com)

    @change-log yon@arsdigita.com 20000711 changed to return lowercase column names
} {
    set columns [list]

    # Works for both Oracle and PostgreSQL:
    db_foreach -dbn $dbn table_column_names {
        select lower(column_name) as column_name
        from user_tab_columns
        where table_name = upper(:table_name)
    } {
        lappend columns $column_name
    }

    return $columns
}


ad_proc -public db_column_exists {{-dbn ""} table_name column_name } {
    @return 1 if the row exists in the table, 0 if not.

    @param dbn The database name to use.  If empty_string, uses the default database.

    @author Lars Pind (lars@pinds.com)
} {
    set columns [list]

    # Works for both Oracle and PostgreSQL:
    set n_rows [db_string -dbn $dbn column_exists {
        select count(*)
        from user_tab_columns
        where table_name = upper(:table_name)
        and column_name = upper(:column_name)
    }]

    return [expr {$n_rows > 0}]
}


ad_proc -public db_column_type {{-dbn ""} {-complain:boolean} table_name column_name } {

    @return the Oracle Data Type for the specified column.
    @return -1 if the table or column doesn't exist.
    @return an error if table or column doesn't exist and -complain flag was specified

    @param dbn The database name to use.  If empty_string, uses the default database.
    @param complain throw an error when datatype is not found

    @author Yon Feldman (yon@arsdigita.com)

} {
    # Works for both Oracle and PostgreSQL:
    set datatype [db_string -dbn $dbn column_type_select {
        select data_type as data_type
        from user_tab_columns
        where upper(table_name) = upper(:table_name)
        and upper(column_name) = upper(:column_name)
    } -default -1]
    if {$complain_p && $datatype == -1} {
        error "Datatype for $table_name.$column_name not found."
    } else {
        return $datatype
    }
}


ad_proc -deprecated ad_column_type {{-dbn ""} table_name column_name } {

    @return 'numeric' for number type columns, 'text' otherwise
    Throws an error if no such column exists.

    @param dbn The database name to use.  If empty_string, uses the default database.

    @author Yon Feldman (yon@arsdigita.com)

    DEPRECATED: it is unclear what the purpose of this proc is. For
                instance, on a Linux/Postgres installation,
                ad_column_type acs_objects object_type ->
                'numeric'. When things should happen based on the
                column type, maybe a better approach is to rely on
                more complete or consistent api, or on the information
                schema.

    @see db_column_type, https://wikipedia.org/wiki/Information_schema

} {
    set column_type [db_column_type -dbn $dbn $table_name $column_name]

    if { $column_type == -1 } {
        return "Either table $table_name doesn't exist or column $column_name doesn't exist"
    } elseif {$column_type ne "NUMBER"  } {
        return "numeric"
    } else {
        return "text"
    }
}


ad_proc -public db_write_clob {{-dbn ""} statement_name sql args } {
    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    ad_arg_parser { bind } $args
    set proc_name {db_write_clob}
    set driverkey [db_driverkey $dbn]

    # TODO: Below, is db_qd_get_fullname necessary?  Why this
    # difference between Oracle and Postgres code?
    # --atp@piskorski.com, 2003/04/09 10:00 EDT

    switch -- $driverkey {
        oracle {
            set full_statement_name [db_qd_get_fullname $statement_name]
            db_with_handle -dbn $dbn db {
                db_exec write_clob $db $full_statement_name $sql
            }
        }

        postgresql {
            db_with_handle -dbn $dbn db {
                db_exec write_clob $db $statement_name $sql
            }
        }

        nsodbc -
        default {
            error "$proc_name is not supported for this database."
        }
    }
}


ad_proc -public db_write_blob {{-dbn ""} statement_name sql args } {
    @param dbn The database name to use.  If empty_string, uses the default database.
} {
    ad_arg_parser { bind } $args
    set full_statement_name [db_qd_get_fullname $statement_name]
    db_with_handle -dbn $dbn db {
        db_exec_lob write_blob $db $full_statement_name $sql
    }
}


ad_proc -public db_blob_get_file {{-dbn ""} statement_name sql args } {
    @param dbn The database name to use.  If empty_string, uses the default database.

    <p>
    <strong>TODO:</strong>
    This proc should probably be changed to take a final
    <code>file</code> argument, <em>only</em>, rather than the current
    <code>args</code> variable length argument list.  Currently, it is
    called only 4 places in OpenACS, and each place <code>args</code>,
    if used at all, is always "<code>-file $file</code>".  However,
    such a change might break custom code...  I'm not sure.
    --atp@piskorski.com, 2003/04/09 11:39 EDT

} {
    ad_arg_parser { bind file args } $args
    set proc_name {db_blob_get_file}
    set driverkey [db_driverkey $dbn]

    set full_statement_name [db_qd_get_fullname $statement_name]

    switch -- $driverkey {
        oracle {
            db_with_handle -dbn $dbn db {
                db_exec_lob blob_get_file $db $full_statement_name $sql $file
            }
        }

        postgresql {
            db_with_handle -dbn $dbn db {
                db_exec_lob blob_select_file $db $full_statement_name $sql $file
            }
        }

        nsodbc -
        default {
            error "$proc_name is not supported for this database."
        }
    }
}


ad_proc -public db_blob_get {{-dbn ""} {-subst all} statement_name sql args } {
    PostgreSQL only.

    @param dbn The database name to use.  If empty_string, uses the default database.
    @param subst Perform Tcl substitution in xql-files. Possible values: all, none, vars, commands

} {
    ad_arg_parser { bind } $args
    set proc_name {db_blob_get}
    set driverkey [db_driverkey $dbn]

    switch -- $driverkey {

        postgresql {
            set full_statement_name [db_qd_get_fullname $statement_name]
            db_with_handle -dbn $dbn db {
                set data [db_exec_lob blob_get $db $full_statement_name $sql]
            }
            return $data
        }

        oracle {
            set pre_sql $sql
            set full_statement_name [db_qd_get_fullname $statement_name]
            set sql [db_qd_replace_sql -ulevel 3 -subst $subst $full_statement_name $pre_sql]
            set data [db_string dummy_statement_name $sql]
            return $data
        }

        nsodbc -
        default {
            error "$proc_name is not supported for this database."
        }
    }
}


d_proc -private db_exec_lob {
    {-ulevel 2}
    type
    db
    statement_name
    pre_sql
    {file ""}
} {
    A helper procedure to execute a SQL statement, potentially binding
    depending on the value of the $bind variable in the calling environment
    (if set).
} {
    set proc_name {db_exec_lob}
    set driverkey [db_driverkey -handle_p 1 $db]

    # Note: db_exec_lob is marked as private and in the entire
    # toolkit, is ONLY called from a few other procs defined in this
    # same file.  So we definitely could change it to take a -dbn
    # switch and remove the passed in db handle altogether, and call
    # 'db_driverkey -dbn' rather than 'db_driverkey -handle'.  But,
    # db_exec NEEDS 'db_driverkey -handle', so we might as well use it
    # here too.  --atp@piskorski.com, 2003/04/09 12:13 EDT

    # TODO: Using this as a wrapper for the separate _oracle and
    # _postgresql versions of this proc is ugly.  But also simplest
    # and safest at this point, as it let me change as little as
    # possible of those two relatively complex procs.
    # --atp@piskorski.com, 2003/04/09 11:55 EDT

    switch -- $driverkey {
        oracle {
            set which_proc {db_exec_lob_oracle}
        }
        postgresql {
            set which_proc {db_exec_lob_postgresql}
        }

        nsodbc -
        default {
            error "$proc_name is not supported for this database."
        }
    }

    ns_log Debug "$proc_name: $which_proc -ulevel [expr {$ulevel +1}] $type $db $statement_name $pre_sql $file"
    return [$which_proc -ulevel [expr {$ulevel +1}] $type $db $statement_name $pre_sql $file]
}


d_proc -private db_exec_lob_oracle {
    {-ulevel 2}
    {-subst all}
    type
    db
    statement_name
    pre_sql
    {file ""}
} {
    A helper procedure to execute a SQL statement, potentially binding
    depending on the value of the $bind variable in the calling environment
    (if set).
} {
    set start_time [expr {[clock clicks -microseconds]/1000.0}]

    set sql [db_qd_replace_sql \
                 -ulevel [expr {$ulevel + 1}] \
                 -subst $subst \
                 $statement_name \
                 $pre_sql]

    set file_storage_p 0
    upvar $ulevel storage_type storage_type

    if {[info exists storage_type] && $storage_type eq "file"} {
        set file_storage_p 1
        set original_type $type
        set qtype 1row
        ns_log Debug "db_exec_lob: file storage in use"
    } else {
        set qtype $type
        ns_log Debug "db_exec_lob: blob storage in use"
    }

    set errno [catch {
        upvar bind bind

        # Below, note that 'ns_ora blob_get_file' takes 3 parameters,
        # while 'ns_ora write_blob' takes only 2.  So if file is empty
        # string (which it always will/should be for $qtype
        # write_blob), we must not pass any 3rd parameter to the
        # ns_ora command: --atp@piskorski.com, 2003/04/09 15:10 EDT

        if { [info exists bind] && [llength $bind] != 0 } {
            if { [llength $bind] == 1 } {
                if { $file eq "" } {
                    # gn: not sure, why the eval was ever needed (4 times)
                    set selection [eval [list ns_ora $qtype $db -bind $bind $sql]]
                } else {
                    set selection [eval [list ns_ora $qtype $db -bind $bind $sql $file]]
                }

            } else {
                set bind_vars [ns_set create]
                foreach { name value } $bind {
                    ns_set put $bind_vars $name $value
                }
                if { $file eq "" } {
                    set selection [eval [list ns_ora $qtype $db -bind $bind_vars $sql]]
                } else {
                    set selection [eval [list ns_ora $qtype $db -bind $bind_vars $sql $file]]
                }
            }

        } else {
            if { $file eq "" } {
                set selection [uplevel $ulevel [list ns_ora $qtype $db $sql]]
            } else {
                set selection [uplevel $ulevel [list ns_ora $qtype $db $sql $file]]
            }
        }

        if {$file_storage_p} {
            set content_index [ns_set find $selection "content"]
            if {$content_index == -1} {
                set content_index 0
            }
            set content [ns_set value $selection $content_index]

            switch -- $original_type {

                blob_get_file {
                    if {[ad_file exists $content]} {
                        file copy -- $content $file
                        return $selection
                    } else {
                        error "file: $content doesn't exist"
                    }
                }

                write_blob {

                    if {[ad_file exists $content]} {
                        set ofp [open $content r]
                        fconfigure $ofp -encoding binary
                        ns_writefp $ofp
                        close $ofp
                        return $selection
                    } else {
                        error "file: $content doesn't exist"
                    }
                }
            }
        } else {
            return $selection
        }

    } error]

    ds_collect_db_call $db $type $statement_name $sql $start_time $errno $error
    if { $errno == 2 } {
        return $error
    }

    return -code $errno -errorinfo $::errorInfo -errorcode $::errorCode $error
}


d_proc -private db_exec_lob_postgresql {
    {-ulevel 2}
    {-subst all}
    type
    db
    statement_name
    pre_sql
    {file ""}
} {
    A helper procedure to execute a SQL statement, potentially binding
    depending on the value of the $bind variable in the calling environment
    (if set).

    Low level replacement for db_exec which emulates blob handling.

} {
    set start_time [expr {[clock clicks -microseconds]/1000.0}]

    # Query Dispatcher (OpenACS - ben)
    set sql [db_qd_replace_sql \
                 -ulevel [expr {$ulevel + 1}] \
                 -subst $subst \
                 $statement_name \
                 $pre_sql]

    # create a function definition statement for the inline code
    # binding is emulated in tcl. (OpenACS - Dan)

    set errno [catch {
        upvar bind bind
        if { [info exists bind] && [llength $bind] != 0 } {
            if { [llength $bind] == 1 } {
                set lob_sql [db_bind_var_substitution $sql [ns_set array $bind]]
            } else {
                set lob_sql [db_bind_var_substitution $sql $bind]
            }
        } else {
            set lob_sql [uplevel $ulevel [list db_bind_var_substitution $sql]]
        }

        # get the content - assume it is in column 0, or optionally it can
        # be returned as "content" with the storage type indicated by the
        # "storage_type" column.

        set selection [ns_db 1row $db $lob_sql]
        set content [ns_set value $selection 0]

        foreach var {storage_type content} {
            set i [ns_set find $selection $var]
            if {$i != -1} {
                set $var [ns_set value $selection $i]
            }
        }

        # this is an ugly hack, but it allows content to be written
        # to a file/connection if it is stored as a lob or if it is
        # stored in the content-repository as a file. (DanW - Openacs)

        switch -- $type {

            blob_get {

                if {[info exists storage_type]} {
                    switch -- $storage_type {
                        file {
                            if {[ad_file exists $content]} {
                                set ifp [open $content r]

                                # DRB: this could be made faster by setting the buffersize
                                # to the size of the file, but for very large files allocating
                                # that much more memory on top of that needed by Tcl for storage
                                # of the data might not be wise.

                                fconfigure $ifp -translation binary

                                set data [read $ifp]
                                close $ifp
                                return $data
                            } else {
                                error "file: $content doesn't exist"
                            }
                        }

                        lob {
                            if {[regexp {^[0-9]+$} $content match]} {
                                return [ns_pg blob_get $db $content]
                            } else {
                                error "invalid lob_id: should be an integer"
                            }
                        }

                        default {
                            error "invalid storage type"
                        }
                    }
                } elseif {[ad_file exists $content]} {
                    set ifp [open $content r]
                    fconfigure $ifp -translation binary
                    set data [read $ifp]
                    close $ifp
                    return $data
                } elseif {[regexp {^[0-9]+$} $content match]} {
                    return [ns_pg blob_get $db $content]
                } else {
                    error "invalid query"
                }
            }

            blob_select_file {

                if {[info exists storage_type]} {
                    switch -- $storage_type {
                        file {
                            if {[ad_file exists $content]} {
                                file copy -- $content $file
                            } else {
                                error "file: $content doesn't exist"
                            }
                        }

                        lob {
                            if {[regexp {^[0-9]+$} $content match]} {
                                ns_pg blob_select_file $db $content $file
                            } else {
                                error "invalid lob_id: should be an integer"
                            }
                        }

                        default {
                            error "invalid storage type"
                        }
                    }
                } elseif {[ad_file exists $content]} {
                    file copy -- $content $file
                } elseif {[regexp {^[0-9]+$} $content match]} {
                    ns_pg blob_select_file $db $content $file
                } else {
                    error "invalid query"
                }
            }

            write_blob {

                if {[info exists storage_type]} {
                    switch -- $storage_type {
                        file {
                            if {[ad_file exists $content]} {
                                set ofp [open $content r]
                                fconfigure $ofp -encoding binary
                                ns_writefp $ofp
                                close $ofp
                            } else {
                                error "file: $content doesn't exist"
                            }
                        }

                        text {
                            ns_write $content
                        }

                        lob {
                            if {[regexp {^[0-9]+$} $content match]} {
                                ns_pg blob_write $db $content
                            } else {
                                error "invalid lob_id: should be an integer"
                            }
                        }

                        default {
                            error "invalid storage type"
                        }
                    }
                } elseif {[ad_file exists $content]} {
                    set ofp [open $content r]
                    fconfigure $ofp -encoding binary
                    ns_writefp $ofp
                    close $ofp
                } elseif {[regexp {^[0-9]+$} $content match]} {
                    ns_pg blob_write $db $content
                } else {
                    ns_write $content
                }
            }
        }

        return

    } error]

    set errinfo $::errorInfo
    set errcode $::errorCode

    ds_collect_db_call $db 0or1row $statement_name $sql $start_time $errno $error

    if { $errno == 2 } {
        return $error
    }

    return -code $errno -errorinfo $errinfo -errorcode $errcode $error
}

d_proc -public db_flush_cache {
    {-cache_key_pattern *}
    {-cache_pool db_cache_pool}
} {

    Flush the given cache of entries with keys that match the given pattern.

    @param cache_key_pattern The "string match" pattern used to flush keys (default is to flush all entries)
    @param cache_pool The pool to flush (default is to flush db_cache_pool)
    @author Don Baccus (dhogasa@pacifier.com)

} {
    #
    # If the key pattern has meta characters, iterate over the entries.
    # Otherwise, make a direct lookup, without retrieving the all keys
    # from the cache, which can cause large mutex lock times.
    #
    if {[regexp {[*\]\[]} $cache_key_pattern]} {
        if {[namespace which ns_cache_eval] ne ""} {
            #
            # NaviServer variant
            #
            ::acs::clusterwide ns_cache_flush -glob $cache_pool $cache_key_pattern
        } else {
            #
            # AOLserver variant
            #
            foreach key [ns_cache names $cache_pool $cache_key_pattern] {
                ns_cache flush $cache_pool $key
            }
        }
    } else {
        acs::clusterwide ns_cache flush $cache_pool $cache_key_pattern
    }
}

ad_proc -public db_bounce_pools {{-dbn ""}} {
    @return Call ns_db bouncepool on all pools for the named database.
    @param dbn The database name to use.  Uses the default database if not supplied.
} {
    foreach pool [db_available_pools $dbn] {
        ns_db bouncepool $pool
    }
}

d_proc -public -callback subsite::parameter_changed -impl acs-tcl {
    -package_id:required
    -parameter:required
    -value:required
} {
    Implementation of subsite::parameter_changed for acs-tcl.


    @param package_id the package_id of the package the parameter was changed for
    @param parameter  the parameter name
    @param value      the new value

} {
    if {$parameter eq "DbLogMinDuration"} {
        set new_value [expr {$value/1000.0}]
        foreach pool [ns_db pools] {
            set ns_db_old_value [ns_time format [ns_db logminduration $pool]]
            set ns_db_old_ms [expr {int($ns_db_old_value * 1000)}]
            set old_ms $::acs::DbLogMinDuration
            ns_log notice "... pool $pool db old value $ns_db_old_value old_ms $old_ms ns_db_old_ms $ns_db_old_ms -> $new_value"
            if {$ns_db_old_value > $new_value || $old_ms == $ns_db_old_ms} {
                #
                # If the "ns_db_old_value" is larger (less SQL
                # logging) and the user wants more logging, then
                # reduce it.  If there is already a more detailed
                # logging turned on then leave it as it is.
                #
                # If the old value was probably set via such parameter
                # settings, adjust as well.
                #
                # Note that "ns_db logminduration ..." has only an
                # effect when SQL logging is turned on.
                #
                #     ns_logctl severity "Debug(sql)" on
                #
                ns_log notice "... adjust pool $pool old value $ns_db_old_value -> $new_value"
                ns_db logminduration $pool $new_valuex
            }
        }
        #
        # We could use nsv instead of the per-thread variable, such as
        # "nsv_set acs_properties DbLogMinDuration", which would be
        # faster to change, but slower to test.
        #
        ns_eval [list set ::acs::DbLogMinDuration $value]
    }
}



# Local variables:
#    mode: tcl
#    tcl-indent-level: 4
#    indent-tabs-mode: nil
# End: