Forum OpenACS Development: RFC: registering filters ...

I want to propose a change to ad_register_filter so that it actually does register the filter (well, rp_invoke_filter in fact) *on the fly*. Currently "all" that ad_register_filter does is nsv_lappends a job to the "rp_filters"-nsv making sure that it will be ns_register_filter'ed at the next restart.

I always assumed that ad_* was the equivalent to ns_*, only better ...

Well, here is the proposed addition to ad_register_filter - just tuck this at the end:

    # Figure out how to invoke the filter, based on the number of arguments.
    if { [llength [info procs $proc]] == 0 } {
        # [info procs $proc] returns nothing when the procedure has been
        # registered by C code (e.g., ns_returnredirect). Assume that neither
        # "conn" nor "why" is present in this case.
        set arg_count 1
    } else {
        set arg_count [llength [info args $proc]]
    }

    set filter_index ?
    ns_register_filter $kind $method $path rp_invoke_filter [list $filter_index $debug $arg_count $proc $arg]

(BTW, filter_index doesn't seem to be used)

Thoughts?

Oh, I think I see: ad_register_filter is supposed to only be called from *.init.tcl scripts, I guess (haven't found code to confirm this yet). Does this sound right? Then, after all all the package's init scripts have run, the nsv is scanned and the filters get registered with ns_register_filter in the proper priority order. After all, the nsv doesn't persist after restart so it would have to be populated at boot time.

In my case I was calling ad_register_filter from an after_instantiate callback, and was surprised that the filter was never registered. It did get registered (by ad_register_filter) after a restart, though, due to the fact that I am looping my app's package_ids in an *init.tcl script ...

I guess you should use the ns_ variant outside the init scripts, unless we come up with a persistent storage mechanism. But i wonder if I should still register rp_invoke_filter [list ...] instead of the filter directly ...

Hm, it looks as though some more thinking would be in order, if I can manage that on such a hot day as this.

My experiments suggest that "apm_first_time_loading" works:

    if { [apm_first_time_loading_p] } {
        # Append the filter to the list.
        nsv_lappend rp_filters . \
            [list $priority $kind $method $path $proc $arg $debug $critical $description [info script]]
    } else {
        # Figure out how to invoke the filter, based on the number of arguments.
        if { [llength [info procs $proc]] == 0 } {
            # [info procs $proc] returns nothing when the procedure has been
            # registered by C code (e.g., ns_returnredirect). Assume that neither
            # "conn" nor "why" is present in this case.
            set arg_count 1
        } else {
            set arg_count [llength [info args $proc]]
        }

        set filter_index {}
        ns_register_filter $kind $method $path rp_invoke_filter [list $filter_index $debug $arg_count $proc $arg]
    }

I will test this some more and, unless I hear otherwise, commit it to HEAD in a couple of days or so.

Thanks Jeff.