NAME
    Devel::PPAP - Push/Pop functions access patterns tracer

SYNOPSIS
        perl -MDevel::PPAP my_script.pl

        # yes, this is stupid, but it's 0.01
        mkdir ppap

        ppap_html
        browser ppap/index.html

DESCRIPTION
    Goal of this module is to collect low-level information about executed
    code, by tracing calls of PP functions storing relevant information
    about arguments. All data is stored in a report file and then can be
    processed by ppap_html script. At this moment file is plain readable
    text, so a person can analyze data as well.

AUDIENCE
    Audience of the results splits into groups: developers working on perl
    (p5p) and developers programming in Perl. How can each of these groups
    win from such reports?

  perl developers (p5p)
    I believe usage statistics will help developers of the perl itself make
    informed decisions for optimizations.

    During testing I found that unshift operation is not using slots
    available on the right side of an array. So after long unshift sequence
    array may end up with big free slot on the right side. See also samples/
    directory.

    I hope to interest people from p5p list in generating ideas that should
    be implemented. Main goal of this distribution is to help them improve
    perl 5 at the end.

  Perl developers
    Find abnormalies in thier code or third party modules they use. For
    example during development of this module I found that ~30% of push
    operations in a project deal with an array with more than 3000 elements,
    it wasn't something I expected to see.

    Now report doesn't show where above situation happen, but sure it's
    possible to implement. You can uncomment two lines in "describe_array"
    function in PPAP.xs to see a warning when an array with more than 1000
    elements is used in executed code.

TODO
    Things that would be cool to implement splitted by area of expertize.
    Pick anything you like to do.

  Java Script (jquery)
    At this point jquery and jquery.tablesorter are used to sort tables and
    there are plenty of things that can be improved:

    *   default sorting for numbers

        Default sorting for numbers is ascending, but it doesn't make much
        sense for counters. We usally want big numbers first.

    *   big tables split

        Part of a big table should be hidden by default. jquery.tablesorter
        has a pager plugin, it's very similar, but only two pages, for
        example 20 first and rest, with hide/show link somewhere.

    *   line coloring widget when table sorted by number

        Better visual distinction of offenders. Table rows marked with
        different CSS classes depending on number. For example three
        clusters: neutral, yellow and red. Some smart clusterization
        algorithm can be used. jquery.tablesorter has zebra plugin that can
        be used to implement this.

    *   ...

  CSS and nice look
    Add a lot of classes and styling to make page look pretty.

  Perl (reports processing)
    Lots and lots ways to represent the data.

    *   per PP function reports

        For example I'm working on push and other arrays related functions
        and the following reports may be interesting:

        *   stats on sizes of arrays we work with

        *   stats on sizes of varying arguments lists

        *   Number of cases when there is enough free slots on the right
            side of the array. Number of cases when there is enough slots on
            the left and rights. Not enough slots.

        *   correlation between time spent and above cases

    *   detecting code anomalies

        Big arrays usages or something like that.

    *   correlation analyzis

    *   graphs?

    *   ...

  XS, C and perl guts
    *   hash, string, scalar describer

        Something like describe_array function that reports interesting
        information about SV.

    *   check in Makefile for available clock functions

        At this moment MacOS fast timing functions are used by default :).
        Module makes no sense without a timer that support at least 100ns
        resolution.

    *   fork support

        After fork we should open a new file for report.

    *   functions to start and stop profiling

        Compilation may be quite heavy in big projects and you have to run a
        lot of actual code to neglect that.

    *   ...

CREDITS
    Thanks to authors and contributors of Devel::NYTProf for great pile of
    wisdom on overriding pp_* functions and other staffs I hope to borrow
    from them :)

REPOSITORY
    http://github.com/ruz/Devel-PPAP

AUTHOR
    Ruslan Zakirov <[email protected]>

LICENSE
    Under the same terms as perl itself.