Permalink
Commits on Jul 8, 2014
Commits on Dec 11, 2013
  1. Avoid having two different HTTP version numbers in the response statu…

    …s line.
    
    Addresses a condition where response objects reported their status as
    "HTTP/0.9 200 OK HTTP/1.1".
    committed Dec 11, 2013
Commits on Sep 14, 2013
  1. Style changes.

    committed Sep 14, 2013
Commits on Sep 12, 2013
  1. Consolidate connection closing code.

    Makes connection shutdown a little more consistent.
    Also fixes a bug where the wrong object ID was used to look up part of
    the connection state.
    committed Sep 12, 2013
Commits on Sep 8, 2013
  1. Provide a dummy version number for development testing.

    The default connection string is based on the component version, which
    stopped being defined in-house since the move to Dist::Zilla.  Tests
    were noisy and ugly with moot "uninitialized value" warnings.
    committed Sep 8, 2013
  2. On a shutdown, deallocate the proper keep-alive connection ID.

    It was deallocating the PoCo::Client::HTTP request ID, which isn't
    coupled to the PoCo::Client::Keepalive ID at all.
    
    The problem wasn't apparent before because IDs were coincidentally in
    synch.  Recent changes finally desynchronized them.
    committed Sep 8, 2013
  3. Honor "Connection: close" response header by closing the connection.

    Without closing the connection, POE::Component::Client::Keepalive
    would put the connection back into its pool and detect closure when
    the server later shut down the socket.
    
    Nicolas Dehaine and Brad Sacks discovered this to be a problem when
    they tried to make a lot of rapid connections to a single address.
    The to-be-defunct connection didn't have enough time to shut down, so
    it would be returned to POE::Component::Client::HTTP.  The HTTP
    transaction would then fail, and an error would be reported.
    
    Their test case allowed me to repeat the problem and diagnose it.
    Otherwise this would still be broken.  Thanks, guys!
    committed Sep 8, 2013
Commits on Jul 16, 2013
  1. Bypass proxies for tests that make localhost connections.

    Localhost is relative, and it's often not what's expective when
    proxying is turned on without an exception for it.
    committed Jul 16, 2013
Commits on May 6, 2013
  1. Return a "400 Bad Request" error if we're redirected to a URI with a …

    …bad scheme.
    
    Sean Brunnock reported the issue when a link redirected his program to
    <htttp://...>.  This triggered an internal confess() we had put there
    to trap a strange case we couldn't reproduce.  His test case and patch
    were instrumental in helping me resolve this on a deeper level.
    committed May 6, 2013
Commits on Jan 13, 2013
Commits on Jun 3, 2012
  1. [rt.cpan.org 76542] Fix a double close in PoCo::Client::HTTP.

    Florian Schlichting found POE throwing an error on an undefined file
    descriptor.  POE::Component::Client::HTTP was closing a filehandle
    before destroying the wheel that was managing it.  When it was time
    for the wheel to remove the filehandle's watcher, fileno() failed, and
    trouble ensued.
    committed Jun 3, 2012
  2. [rt.cpan.org 77601] Switch to Socket's getaddrinfo().

    AVENJ reported that Socket::GetSockAddr's :newapi was deprecated.
    Switch to Socket for a more future-proof API.
    committed Jun 3, 2012
Commits on May 15, 2012
Commits on Apr 29, 2012
  1. [rt.cpan.org 76543] Help out Debian with a NAME and SYNOPSIS for an i…

    …nternal module.
    
    Florian sent in a patch written by Krzysztof Krzyzaniak to add NAME
    and SYNOPSIS sections to POE::Component::Client::HTTP::RequestFactory.
    This is reported to help Debian automatically build a "whatis" entry
    for the module. Thanks for the patch! Please send any others upstream.
    committed Apr 29, 2012
Commits on Mar 10, 2012
  1. [rt.cpan.org 70268] Some DNS providers redirect bad hosts.

    They redirect bad hosts instead of allowing outright failure.  I think
    it's supposed to be helpful, or improve domain sales, or both, or
    something.  But anyway, it meant that people using certain DNS
    providers wouldn't be able to cleanly test this distribution.
    
    Thanks to "blackcode" for reporting this.  Unfortunately I'm not sure
    of a foolproof solution other than to make the test accept 303s.
    
    Patches are welcome, dear reader from the future.
    committed Mar 10, 2012
Commits on Jan 25, 2012
  1. Allow response events to be objects.

    Reflex::POE::Event objects are intended to be used as POE events.  For
    it to work, components like this one must treat events as opaque data.
    committed Jan 25, 2012
Commits on Jul 30, 2011
Commits on Jul 22, 2011
  1. Merge pull request #3 from shlomif/master

    Convert http:// URLs to L<..> in the POD.
    committed Jul 22, 2011
  2. Add L<...> to URLs.

    They were displayed as plaintext in the web-interface.
    shlomif committed Jul 22, 2011
Commits on May 4, 2011
  1. Prepare for next release.

    committed May 4, 2011
Commits on Feb 21, 2011
  1. Bump the version for release.

    committed Feb 21, 2011
  2. Add tests for pending_requests_count() and fix underlying cause of RT…

    …#50231, thanks Jeff Bisbee
    apocalypse committed with Jan 30, 2011
Commits on Feb 14, 2011
  1. Bump the version for release.

    committed Feb 14, 2011
Commits on Feb 13, 2011
Commits on Jan 27, 2011
  1. Some machines time out before they can respond with a 500 error on ba…

    …d host. Accept this as a valid outcome.
    apocalypse committed with Jan 27, 2011
  2. Fix the X-PCCH-Errmsg header.

    committed Jan 27, 2011