Commits on May 10, 2003
  1. Increase the version number to 0.26 for release. Revise the README

    rcaputo committed May 10, 2003
    for clarity, conciseness, etc.
Commits on May 9, 2003
  1. Add a warning to explain the most likely cause of this module hanging.

    rcaputo committed May 9, 2003
    Thanks to Somni for reminding me of the problem.
Commits on May 4, 2003
  1. Increase version from 0.2502 (first release candidate for 0.26) to

    rcaputo committed May 4, 2003
    0.2503 (live CVS for any last minute bug fixes before 0.26).
  2. Increasing the version from 0.2501 (CVS after 0.25) to 0.2502 (first

    rcaputo committed May 4, 2003
    release candidate for 0.26).
Commits on May 3, 2003
  1. A previous patch to shutdown client sockets before they had closed

    rcaputo committed May 3, 2003
    used $self in closures which themselves were stored in $self,
    resulting in circular references that would prevent the patch from
    actually working.  This patch goes through some gyrations to break
    those circular references.
Commits on May 2, 2003
  1. Added the ability to shutdown() input/output on a ReadWrite wheel

    rcaputo committed May 2, 2003
    attached to a socket.  Thanks to Peter Chen for suggesting it.
Commits on Apr 30, 2003
  1. Applied Nick Williams' patch to avoid internal inconsistency problems

    rcaputo committed Apr 30, 2003
    with signals use:
    > When using the kernel's signal() method to send a user-defined
    > signal, it's possible to get a signal dispatched to a session which
    > is in the process of shutting down (i.e. about to call the _stop
    > event). However, the garbage collection in
    > _data_sig_free_terminated_sessions doesn't take this into account
    > and can cause POE to die screaming with an internal inconsistency
    > error in _data_ses_collect_garbage.
Commits on Apr 29, 2003
  1. Fix POE::NFA's stop() method. Previously it was using POE::Kernel's

    rcaputo committed Apr 29, 2003
    session_free() method, which was recently renamed.
Commits on Apr 21, 2003
  1. Work around the persistent double-GC problem by avoiding a second GC …

    rcaputo committed Apr 21, 2003
    …on a defunct session.
Commits on Apr 8, 2003
Commits on Apr 7, 2003
  1. Update to version 0.50.

    rcaputo committed Apr 7, 2003
  2. Fixed a leak discovered by dynweb. In client mode, SocketFactory

    rcaputo committed Apr 7, 2003
    would did release its grip on the socket until destroyed.  He found
    the leak in a program that passed the socket to $kernel->select_read()
    without destroying the SocketFactory.
Commits on Apr 5, 2003
Commits on Apr 3, 2003
  1. Patch the documentation for the Reuse parameter. Also support another

    rcaputo committed Apr 3, 2003
    reasonable alternative.  It might be best, however, to trust the user
    to supply a proper Perl boolean value.
Commits on Mar 24, 2003
  1. Applied and documented Dmitri Tikhonov's patch for the ProgramArgs

    rcaputo committed Mar 24, 2003
    feature.  This lets master programs specify parameters for
    code/programs that are forked off, avoiding closures and the possible
    memory leaks created by them.  The tests were extended to exercise
    the new feature.
  2. It seems impossible to avoid duplicate tags (identical timestamp) for

    rcaputo committed Mar 24, 2003
    the initial import of a source tree into CVS.  The usual convention is
    to tag the first import as "start", so we ignore any other tags that
    coincide with that.
Commits on Mar 16, 2003
  1. The two-second timeout was too slow for the 486 I'm using to test

    rcaputo committed Mar 16, 2003
    Windows NT.  Increased the timeout to five seconds so that the test
    doesn't trigger false failures on slow machines.
Commits on Mar 5, 2003
  1. Hachi's double-free fix leaves the code in this file, but it's

    rcaputo committed Mar 5, 2003
    commented out.  I'm commenting it so archaeologists will know what he
    was thinking.
  2. Clear the Client::TCP component's alias on shutdown. Otherwise, if

    rcaputo committed Mar 5, 2003
    something else in the program is active, the Client will disconnect
    but its internal session won't _stop.  "Buzz Darkyear" on MagNet #poe
    reported this one.
Commits on Mar 3, 2003
Commits on Mar 2, 2003
  1. Bug squashing attempt, sessions that used postbacks may in the past g…

    Jonathan Steinert committed Mar 2, 2003
    …et garbage collected twice, there was an internal consistency check to this effect which made a confess() call.
    I hope I wasn't supposed to increment the version to .2502 or something before committing.
Commits on Mar 1, 2003
  1. <sky> what happens is that the code is compiled twice, and the first

    rcaputo committed Mar 1, 2003
          time in the same scope as the ST_TIME
    <dngnand> This is the 0.22 bug?
    <sky> yes
    <sky> the bug exists in non threaded and threaded perl
    <sky> but because threaded perl keeps constants on the pad, it
          actually shows
    <dngnand> Oh!  Does it also do this in blead, then?
    <sky> no, but maintblead
    <sky> blead fixes it
    <sky> but blead has mucho rewritten pad handling
    <dngnand> Do you think this is something that will bite regular
              POE/Perl users?
    <sky> not in 0.23+
    <dngnand> I'll remove the warning from Makefile.PL, then.
Commits on Feb 27, 2003
  1. From the mailing list:

    cwest committed Feb 27, 2003
      Date: Tue, 25 Feb 2003 16:28:11 -0800
      From: Jim Westfall <>
      Subject: FollowTail behavior change with SeekBack
      I recently upgraded to 0.25 from 0.23 and noticed there is a
      behavior change to FollowTail with the SeekBack option.  A
      'SeekBack' => 0 is now causing it to seek back instead of seeking to
      the EOF like one would expect.  The code was as follows
         my $seek_back = ( (defined $params{SeekBack})
                            ? $params{SeekBack}
                            : 4096
      and has been changed to
        my $seek_back     = $params{SeekBack} || 4096;
      so now if $params{SeekBack} is undefined or set to 0 it will make
      $seek_back 4096.  simple workaround on my part is to set SeekBack to
      A simular situation exists with the PollInterval option, tho I dunno
      if 0 would be a valid option for it.
    This was a short sided change.  The old behaviour is correct and has
    been restored.
Commits on Feb 22, 2003
Commits on Feb 18, 2003
  1. Cygwin doesn't give us warnings when creating this bogus

    rcaputo committed Feb 18, 2003
    SocketFactory.  This patch works around Cygwin's strange behavior in a
    most unsatisfying way.
Commits on Feb 8, 2003
Commits on Feb 7, 2003
  1. Documentation cleanup in PoCo::Client::TCP and PoCo::Server::TCP

    Jonathan Steinert committed Feb 7, 2003
    First time I committed CVS (outside of my own hard drive) Yay me!
  2. This fix is against against revision 1.14. Forcing that new test to

    rcaputo committed Feb 7, 2003
    use "inet" pipes breaks on systems where localhost is firewalled.
    Tweaked the test to use whichever pipes it can under non-Windows
    operating systems, and forced it to "inet" for MSWin32.  A real
    solution would be to skip the test entirely if the pipe can't be
    built.  I suspect we'll need to revisit this later.
Commits on Feb 6, 2003
  1. It is unfortunate, but a bug in ActivePerl 5.8.0's version of Tk

    rcaputo committed Feb 6, 2003
    prevents us from supporting Tk at all in this build.  We regret the
    inconvenience and hope that support can resume in a future ActivePerl
    build.  Meanwhile, you may follow the status of the bug at