Forum OpenACS Development: Announcement: NaviServer 4.99.25 available

Dear all,

I am glad to announce that the release of NaviServer 4.99.25 is available at SourceForge [1]. This release is mostly a bug-fix release. The forthcoming version 5.0 of NaviServer will contain several new features omitted in this bug-fix branch. In case, you are building NaviServer from the Bitbucket repository, please note that the release 4.99.25 is in the branch release/4.99 (bug fix branch for the NaviServer 4.99 family). New development happens in the "main" branch of the repository, leading to NaviServer 5.*.

See below for a summary of the changes.

Many thanks to the contributors of this release:

     Andrew Piskorski
     Antonio Pisano
     Brian Fenton
     Gustaf Neumann
     Hector Romojaro
     Joe Oldak
     Khy Huang
     Oleg Oleinick
     Zoran Vasiljevic
All the best!

-gustaf neumann

NaviServer 4.99.25, released 2023-05-01

 132 files changed, 3957 insertions(+), 2068 deletions(-)

New Features:

 - Added meta-information to configuration values

   NaviServer can now report, what configuration values provided in
   the configuration file were actually used, what their default
   values are, and whether these values were specified or not (using
   the default values). With this information, administration
   (e.g. migration) becomes easier. The NaviServer module "nsstats"
   shows this meta information via the web interface.

   This functionality is provided via the new option "-filter" for the
   command "ns_configsection ... section". When the "-filter" option
   is used, different kinds of information about the parameters is
   returned from the specified section.

    "-filter unread":

        Returns the parameters, which were set during configuration (i.e.,
        in the configuration file) but which were not read in the startup
        phase of the server. This option is useful to determine
        e.g. typographical errors of specified parameter names.

    "-filter defaulted":

        Returns the parameters, from which the default values were read
        (i.e., which were not explicitly set)

    "-filter defaults":

        Returns the defaults of the parameter. This is useful for
        contrasting the actual values with the default values for
        Parameters, e.g. in a web based interface.

 - ns_set reform (per default deactivated in 4.99, but activated in 5.*)

   The classical implementation for ns_sets uses separately malloced
   storage for every attribute name and attribute value. So, e.g., for
   1000 ns_sets with 20 members each, this means 1,000*20*2 = 40,000
   malloc/free operations, e.g., for a single db query! Although the
   malloc implementations have improved over the years, these will
   require many lock operations, especially under load, where many
   threads might perform many concurrent malloc operations. One other
   consequence is that the allocated memory will be scattered over
   address space, which has bad implications for CPU caching.

   The new implementation uses for one "ns_set" a single Tcl_DString
   keeping all attribute names and attribute values. This reduces the
   malloc operations and improves memory locality, such that cache
   hits will improve.

   One caveat of this change is that modules using "ns_set" have to be
   recompiled, since the full C-level data structure of the "ns_set"
   is exposed. Therefore, adding a member causes a binary
   incompatibility. One other potential problem is that C-level
   modules using the Ns_Set* API have to make sure that long-living
   string values are copied (this was necessary before as well, but
   was in many cases no problem, when the "ns_sets" were seldom

   For high compatibility, this feature is deactivated per default in
   the 4.99.* series and can be activated by setting the compile-time
   C macro "NS_SET_DSTRING".

API changes:

   API extensions:
    - Provide a new interface ending with *Sz to provide string sizes.
      This reduces the need of strlen() operations.
      * Ns_SetCreateSz()
      * Ns_SetIUpdateSz()
      * Ns_SetPutSz()
      * Ns_SetPutValueSz()
      * Ns_SetUpdateSz()

    - New API calls for "ns_set" reform
      * Ns_SetClearValues(): clear the values for all keys
      * Ns_SetDataPrealloc(): creating ns_sets with preallocated values
        to avoid resize operations
      * NsSetResize()
      * NsHeaderSetGet()

    - Ns_ConfigSet(const char *section, const char *key, const char *name)
      The last argument is new and allows one to create named sets
      (previously, all such sets were unnamed)

    - NsHexPrint(): Print the potentially binary content of a buffer
      in human-readable form.

    - Ns_RelativeTime(Ns_Time *relTimePtr, Ns_Time *timePtr)
      This call implements the inverse operation of Ns_AbsoluteTime(),
      and is used mostly to make debug messages eye-friendly.

Performance Improvements:

 - Replaced malloc operation per log entry by thread local variable in
   system log implementation.

 - When NaviServer 4.99.25 is compiled with NS_SET_DSTRING supportm
   the following preliminary performance results were measured from
   the "ns_set" reform (see above).  The tests were performed on (Xeon Gold 6226R CPU @ 2.90GHz, 32 cores,
   hyper-threading enabled). The test executes the SQL query

        select * from acs_objects limit 1000

   100 times in sequence. This test is run in 1 to 30 concurrent
   threads. With 30 threads, 3mio tuples are retrieved, and 72 mio
   malloc/free operations are needed alone for the retrieved values.

   Before (classical ns_set with many mallocs):

        threads 1 total 4606.787 ms avg 3285.25 ms
        threads 5 total 4595.358 ms avg 3493.07 ms
        threads 10 total 4804.193 ms avg 3755.93 ms
        threads 20 total 6279.524 ms avg 4569.16 ms
        threads 30 total 8966.427 ms avg 6618.58 ms

   After reform (using common Tcl_DString per tuple):

        threads 1 total 4524.645 ms avg 3242.54 ms
        threads 5 total 4251.266 ms avg 3450.09 ms
        threads 10 total 4656.795 ms avg 3665.31 ms
        threads 20 total 5934.105 ms avg 4671.38 ms
        threads 30 total 7384.591 ms avg 5642.76 ms

   To summarize, the improvement increases under higher load (with
   more parallel threads). E.g., with 30 threads, the total time
   improved by 17%.... leading also to a smaller RSS. These tests were
   not performed under "clinical" conditions.

   The new Tcl API call "ns_set size" can be used to pre-allocate
   larger ns_sets, such that the usual Tcl_DString growing policy does
   not kick in, reducing further realloc() operations.

Bug Fixes:

 - Fixed potential crash in "ns_accesslog extendedheaders XXXX".
   Setting extended headers via configuration file was correct, but
   changing it dynamically via "ns_accesslog extendedheaders .." was
   broken. (Issue

 - "ns_conn location": Fixed potential race condition

    It was possible that "ns_conn location" could return inconsistent
    results in a single request, when the underlying sockPtr was
    aggressively reused. Now, the value of the location member is
    copied to the connection structure instead of being shared with
    the socket structure (as before).

 - "ns_cache_eval -force": Fixed potential race condition

    There was a problem with "ns_cache_eval -force", where the system
    relied on the existence of a pre-existing entry, but in case the
    entry was flushed in the meantime problems a crash might have
    happened. Now the value during the "-force" call is cleared exactly
    like in the case of an unset operation. The null-value operations
    are already protected until these are finished in various places
    in the code.

 - "ns_socknread": Fix potentially wrong result for buffered channels

 - Bug fixes for problems showing up under MS Windows:

    * Make sure that the output variable of Ns_ObjvIndex() is always an
      integer. Previously, the output variable was in two places a
      character variable, causing crashes under MS Windows.

    * Handle incompatibility in setlocale() under MS Windows.  Under
      MS Windows, later calls to setlocale() overwrite the string
      returned by former calls. So, it is necessary to copy of the
      returned string of a setlocale() call under MS Windows.

    * Handling linking problems: MS Windows requires explicit handling
      when importing symbols from .dll files (Ns_LogSqlDebug,

    * Pass error codes from low-level function SockRecv() and
      SockSend() via variables.  This change makes sure the real error
      code (immediately after the I/O operation) is passed to the
      caller to avoid missed error cases and weird error message
      (e.g. under windows).

 - Bug fixes for ADP parser:

    * Support for greater than sign ">" inside attribute values.

      Previously, NaviServer determined the terminating end-of-tag
      character as literally the first greater than sign, no matter if
      this was used as attribute values between single or double

      The new version supports such values, since the "Living Standard
      of HTML" [1] allows the use of less than "<" and greater than
      ">" signs inside attribute values as long these are between
      single or double quotes.  The guide [2] just recommends using
      character escapes for "<", ">" and "&".


      This problem was reported by Wolfgang Winkler

    * Clear ADP flags in case of errors (this error was present for
      many years)

      Previously, The following ADP page could lead to a full
      breakdown of the server, since the error states of the server
      were never cleared, and subsequent requests served be the same
      interpreter could lead to the old error states.

      Many thanks to Oleg Oleinick for reporting and the great test

 - Provide better HTTP status code for early errors: When the driver
   terminates already a request, it might be the case that the error
   flags were not read out. In such cases, we provide now more
   specific status codes rather than 400.

 - ns_http:

   * Improved robustness with domain names resolving against many IP
     addresses: When a domain name is resolved against many IP
     addresses, and all these IP addresses block (connect returns "in
     progress") then the old code might have looped infinitely. Now
     the code respects the provided timeout (default set to 5s) per
     resolved IP address.

     Many thanks to Joe Oldak for pointing out the problem and
     leading us to the solution.

   * Fixed timeout handling during TLS handshake: This problem could
     appear, when the TCP connection to a server succeeded quickly,
     but the TLS handshake was taking a long time, without a raising
     a timeout exception, although the timeout time has expired.

 - Set the default server before the init-scripts is called to make it
   accessible from there

 - Provide compatibility between in-memory and file-based form-data
   handling for invalid characters: Since the file based
   implementation uses a fallback-charset of iso8859-1 when parsing
   form data in POST requests (in order to be able to extract
   "_charset_" data), the in-memory based variant does now the
   same. The decision of which parser is used is taken based on the
   size of the form data.... and should therefore be consistent.

 - nscp: Fixed problem, when the nscp module is activated but no
   "users" section is specified. Many thanks to Andrew Piskorski for
   reporting the problem.

 - Security Improvement: Added protection against sneaking in fake
   NAME.tmpfile entries into form fields, when performing file upload

Documentation improvements:

 - Improved the following man pages:


Configuration Files:

 - New feature for the sample configuration file openacs-config.tcl

   The following variables (with prefix oacs_) can be taken from the
   shell variables. This makes it easier to run similar variants of
   NaviServer instances from a single configuration file, while
   providing variables from the command line. This is e.g. useful for
   Docker or cluster setups.


   One can use e.g.the following command to change some ports and the
   log file during startup

    oacs_httpport=8000 oacs_httpsport=8443 oacs_smtpdport=2526 \
        oacs_logroot=/var/www/XXXX/log-node1/ \
        /usr/local/ns/bin/nsd -t /var/www/XXXX/etc/openacs-config.tcl ...

   - further updates for openacs-config.tcl:
      * Added sample nssmtpd configuration
      * Added placeholder for ClusterSecret
      * Reflect recent Oracle requirements (tested with Oracle 19c)
      * Added documentation for "StaticCSP", "CookieNamespace",
        "NsShutdownWithNonZeroExitCode", "LogIncludeUserId"

   - Updated all sample configuration files

Code Changes:

 - Added and updated predefined MIME types based on

 - Added compile time macro NS_VERBOSE_MALLOC to identify frequent
   *alloc operations

 - Print version of zlib during startup

 - OpenSSL maintenance:

    * Improved robustness for OCSP with OpenSSL 3.*

      The previous version crashed with OpenSSL 3.*, when OCSP was
      tried on self-signed certificates. Aside of the fact that OCSP
      does not make sense for self-signed certificates, the server
      should not crash in such situations.

    * Fixed ns_crypto::aead::encrypt/decrypt test under OpenSSL 1.1.1
      (as shipped per default, e.g. on Ubuntu 18.04.4)

      In short, the problem was that with this version of OpenSSL,
      setting empty additional authenticated data (AAD) behaved
      differently from other versions, namely it was clearing
      incorrectly (forgetting) the information that the initialization
      vector (IV) was already set. An upgrade of OpenSSL fixed the
      problem. However, with these changes, also the stock version of
      OpenSSL can be used. As a byproduct, better error messages are
      now produced, the code received more cleanup (e.g. explicit
      initialization, etc.)

 - Automated testing:

    * Setup if Bitbucket + GitHub pipelines for automated regression
      testing with multiple versions of components

      For NaviServer 4.99 the current setup performs tests with gcc-10
      + gcc-11, Tcl 8.6.13 + 8.7a5, NSF 2.4,0 + 2.4.0, tDOM 0.9.1 +
      0.9.3, extra modules: nsdbpg nsdbi nsdbipg nsudp nscoap nssmtpd.

    * Improved robustness of regression test when running with the
      docker networking setup

    * Force nonzero exit code when regression test fails

    * Added parameter "-timeout" to call of regression test cases
      Previously, the timeout was hard-wired to 3 seconds. One can now
      call a test with e.g. "nstest::http -timeout 1s ... GET ..."

    * adp_compress.test: removed trailing newline to ease
      cross-platform regression tests

    * Prefer standard Tcl test constraint "macOrUnix" over own solution

    * Extended regression tests with more test cases

 - Code management:
    * Changed name of branch from "master" to "main"

 - Code Cleanup
     * Improved type cleanness
     * Removed deprecated calls to "sprintf"
     * Improved portability for Tcl 8.7* (handling of binary data)

 - Improved comments, fixed typos

Changes in NaviServer Modules:

 39 files changed, 9658 insertions(+), 1781 deletions(-)

  Adjust to necessary API changes in NaviServer (contains as well
  support for the forthcoming release of NaviServer 5)


  - new pg-driver specific command: ns_pg_prepare /sql/

    Return a dict building a prepared statement for the passed-in SQL
    statement. The dict contains the keys "sql" and "args". The
    function is used by e.g. OpenACS to generate prepared statements
    from SQL commands with bind variables.

  - Raise exception when a value for a bind variable contains a NUL character.
    This value is explicitly forbidden in text strings passed to PostgreSQL.

  - Let "ns_pg" report available subcommands even when handle is not
    specified.  This makes the command compatible with the "icanuse"
    feature in OpenACS.


  - HTTP client log analysis:
     * Provide charts for performance (using highcharts via CDN)
     * Provide charts on request frequency (using highcharts via CDN)
     * Provide a summary table for HTTP client requests
     * Improved robustness against invalid URLs (containing unescaped
     * Added support for selection of different HTTP client log files
       via web interface

  - "Process" page:
      * Added percentage of request distribution over connection pools
      * Added information about the connected client
      * Added more detailed version information

  - Added cache configuration to output when looking at a single cache

  - Improved "log file" analysis
    * Automated stripping of color codes

    * Added filter option. The filter can be used to grep for (ID)
      strings in both the system and access logs, providing a summary
      for the traces of a request in the access log and system log on
      a single place.

  - Added default and usage information to "Config Parameters" page


  - Fixed bug when streaming LOB content to connection. The old code
    did not distinguish between binary and non-binary content.  This
    bug was discussed in

    Bumped version number to 2.9

- switched to plain Debug handling for debugging the driver

    The handling of Ns_LogSqlDebug is performed inside nsdb, including
    also the printout of (long) SQL statements. Previously, the driver
    was too chatty when Debug(sql) was turned on.

  - Added support for output columns of type SQLT_TIMESTAMP or SQLT_TIMESTAMP_TZ

    This change fixes a bug, where SQL queries of the form

         SELECT TO_TIMESTAMP(sysdate) FROM dual

    lead to errors for the form:

         Database operation "getrow" failed (exception 1406, "nsoracle.c:3659:Ns_OracleGetRow:
         error in `OCIStmtFetch ()': ORA-01406: fetched column value was truncated

    The driver needs for several output types special rules, where the
    timestamp cases were not supported so far. It is also recommended
    to set the according environment variables specifying the output
    format in the configuration server of NaviServer, such as e.g.


    For testing in you local Oracle installation, you might test the
    output formats (and the required sizes with the following snippet
    for sqlplus:

       COLUMN localtimestamp format a40
       COLUMN systimestamp format a40
       COLUMN ts_bytes format a80

       alter session set nls_timestamp_format='YYYY-MM-DD HH24:MI:SS.FF6';
       select localtimestamp, length(localtimestamp), dump(localtimestamp) ts_bytes from dual;

       alter session set nls_timestamp_tz_format='YYYY-MM-DD HH24:MI:SS.FF6 TZH:TZM';
       select systimestamp, length(systimestamp), dump(systimestamp) ts_bytes from dual;

       alter session set nls_timestamp_tz_format='YYYY-MM-DD HH24:MI:SS.FF6 TZR';
       select systimestamp, length(systimestamp), dump(systimestamp) ts_bytes from dual;


  - Added option to produce certificates with ECDSA:

    Prior to this change, all certificates were using RSA keys.
    Since a while, keys based on elliptic curves became the preference
    of letsencrypt.


   - Fixed a bug in "ns_conn" emulation, when the "kernel" was not correctly identified
Posted by Brian Fenton on
Well done Gustaf and team, this is seriously impressive. Also exciting to see first mention of an upcoming Naviserver 5.* - wondering what goodies you have in mind for us.

best wishes

Posted by Gustaf Neumann on
NaviServer 5 is more of an evolution than a revolution, therefore it will be fully compatible with the NaviServer 4.99 series. As Linus framed it many years ago, we are "running out of fingers and toes" for counting the version numbers.

The tip version on Bitbucket is what we have so far
- ns_set reform (better scalability, optionally also available for 4.99)
- ns_http with persistent connections
- Tcl9 compatibility (Tcl9 is in some respects still a moving target, we will probably wait with the official 5.0 release until Tcl9 is released)
- Improved crypto support (e.g. Argon2)
- Upgrade of Licensing (from Mozilla Public License 1.1 to Mozilla Public License 2.0, eases packaging for large distros)
... plus various other changes.

Posted by Brian Fenton on
All sounds great! Many thanks for your amazing work as always.