Permalink
Commits on May 20, 2005
Commits on May 19, 2005
  1. Fall back to a single POE::Kernel->run() in this test because multiple

    rcaputo committed May 19, 2005
    entries were crashing ActivePerl + Tk.
Commits on May 17, 2005
  1. Fix a miscount in the number of skipped tests under Windows. Also

    rcaputo committed May 17, 2005
    skip all tests when using Event, as it seems that Event +
    thread-simulated fork() + re-entering the event loop = Breakage.
  2. Minor documentation gap, and some bad grammar.

    Jonathan Steinert committed May 17, 2005
Commits on May 16, 2005
  1. General cleanups during a Devel::Cover debugging session. Be more

    rcaputo committed May 16, 2005
    careful about the steps taken.  Display some debugging output.
    Explicitly specify an output directory.
  2. The restructured tests cause one of the sessions to receive a SIGCHLD

    rcaputo committed May 16, 2005
    when it doesn't have a wheel.  Now we don't try to call a method on
    that nonexistent wheel.
  3. Andrew A. Chen reported on the mailinglist that inet_aton wasn't impo…

    Larwan Berke committed May 16, 2005
    …rted, thanks!
Commits on May 15, 2005
  1. Fix SIGCHLD and SIGPIPE so they are IGNOREd by default. In SIGCHLD's

    rcaputo committed May 15, 2005
    case, this means Perl will reap child processes so you don't need to
    register a SIGCHLD handler if you don't care.  In SIGPIPE's case, it
    means programs won't die if they mistakenly write to a broken pipe.
    Both situations have occurred since POE stopped setting default
    handlers for these signals.
  2. Use better choices for the poll(2) flags when watching filehandles for

    rcaputo committed May 15, 2005
    normal (non-expedited/OOB) input and output.  Drop the minimum poll
    timeout hack for certain legacy Linux builds (it was 0 anyway).
  3. Skip the pty tests when IO::Poll or Event are used on Mac OS X. It

    rcaputo committed May 15, 2005
    appears that OS X 10.4 has broken poll(2) when used with pseudo-ttys.
    I haven't found a work-around other than to bypass the tests.
  4. Moved the POE::Pipe::{One,Two}Way test code out of the POE::Wheel::Run

    rcaputo committed May 15, 2005
    tests and into their own files.  This shouldn't sacrifice coverage
    since the pipe tests don't use event loops anyway.
  5. Duplicated test, same test is a couple lines up verbatim.

    Jonathan Steinert committed May 15, 2005
  6. Fix make distclean so test directories (not just the tests themselves…

    Jonathan Steinert committed May 15, 2005
    …) are cleaned up. This helps to stop CVS from screaming at you if you've run tests and then want to commit some changes (still have to run make distclean, but at least it works.)
Commits on Apr 29, 2005
  1. Compensate for machines with variable CPU rates. They cause

    rcaputo committed Apr 29, 2005
    hi-resolution time() to bobble backwards by a few milliseconds eveny
    few seconds.  This test requires events to run in a certain order, and
    they don't under those circumstances.  It may turn out that we need to
    use a timeless priority queue for FIFO events rather than relying on
    the clock to be monotonic.
Commits on Apr 25, 2005
  1. Put TkActiveState.pm's $VERSION in the correct package. Module::Build

    rcaputo committed Apr 25, 2005
    will soon start examining version numbers more strictly.
Commits on Apr 24, 2005
  1. Document that POE::Wheel::Run and POE do not reap child processes

    rcaputo committed Apr 24, 2005
    unless an event is registered for SIGCHLD.
Commits on Apr 23, 2005
  1. Remove tabs from source code. People (myself included) have been

    rcaputo committed Apr 23, 2005
    introducing them with edits and patches.
  2. Enable support for $kernel->run_one_timeslice() under Tk. I had to

    rcaputo committed Apr 23, 2005
    implement loop_do_timeslice() and replace Tk's MainLoop with a custom
    one in loop_run().  I haven't tested this against the memory leak test
    cases in rt.cpan.org, but I'm hoping this code (which is less twisty)
    will solve those issues as well.
  3. ActiveState Perl + Tk + POE::Wheel::Run + coderefs = Pain.

    rcaputo committed Apr 23, 2005
    ActiveState Perl emulates fork() with iThreads.  Only Perl's core
    exit() will properly "exit" a "process".  Tk overrides exit() with its
    own XS magic.  This seems to kill off the parent "process".  When
    POE::Wheel::Run is used to run subroutines (coderefs) in child
    processes, it must somehow exit() the child process, simulating exec.
    Under Windows, that's Perl's core exit(), which is usurped by Tk,
    which causes ALL MANNER OF EVIL!  DON'T GO THERE!  WE SURE AREN'T!
Commits on Apr 21, 2005
  1. Resolve rt.cpan.org ticket 11056: Socket6.pm can be installed under

    rcaputo committed Apr 21, 2005
    Cygwin, but Cygwin doesn't support IPv6.
  2. Apply Ed W's advice regarding non-blocking connect() in ActiveState

    rcaputo committed Apr 21, 2005
    Perl.  His advice works in stand-alone (non-POE) tests, and the change
    doesn't break anything in POE's test suite, but I haven't actually
    tested whether non-blocking connect() works under POE now.
Commits on Apr 18, 2005
  1. Re-entering Tk::MainLoop too many times, plus signals, seems to

    rcaputo committed Apr 18, 2005
    produce a very deterministic coredump, at least on my current FreeBSD
    perl build.  Removed one of three instances of POE::Loop->run(), and
    the crashy-dumpy seems to have gone away.
  2. Replace exit(0) in the test templates with POSIX::_exit(0). Some

    rcaputo committed Apr 18, 2005
    libraries (*cough*Tk*cough*) override Perl's exit(), which seems to
    cause trouble with certain tests' exit values.  So even if all the
    tests pass, the FILE fails!  Augh!  This seems to work around it
    nicely, and it's good mojo on OS X, FreeBSD, and ActivePerl.
Commits on Apr 13, 2005
  1. Replaced {STDIN,STDOUT,STDERR}_FILENO with checks for fileno() in the

    rcaputo committed Apr 13, 2005
    close-on-exec loop we have hand-rolled.  I'm hoping the regular
    close-on-exec works similarly and won't close STDIN/STDOUT/STDERR
    if their descriptors happen to be >2.
  2. Applied Nick Williams' patch to add Hostname to Server::TCP. This is

    rcaputo committed Apr 13, 2005
    a nonambiguous unpacked hostname or address that will be passed
    through inet_aton().  Useful when your hostnames can be four
    characters long, since they look suspiciously like packed addresses to
    length().
  3. Remove Test::More as a dependency. It has a tendency to want lots of

    rcaputo committed Apr 13, 2005
    other modules.  That dependency tree eventually leads to installing a
    new version of Perl.  Big can of worms.
  4. Uhlarik Ond�ej suggested that POE::Wheel::FollowTail should be allowed

    rcaputo committed Apr 13, 2005
    to follow files that don't exist yet.  This change adds that feature,
    resolving Uhlarik's rt.cpan.org ticket, number 6048.  I hope the utf-8
    in his name comes through.
  5. A cheap indent fix.

    rcaputo committed Apr 13, 2005
Commits on Apr 12, 2005
  1. Replace the hardcoded "make" utility with $Config::Config{make} so

    rcaputo committed Apr 12, 2005
    it's "nmake" or a cabbage or something on systems with nonstandard
    build tools.
  2. This guy named Jack has written a WebDAV server with POE and needs

    rcaputo committed Apr 12, 2005
    Filter::HTTPD not to bomb if the request method isn't GET, HEAD, or
    POST.  I've applied a modified version of his patch, resolving RT
    ticket #11821.