Well, additional testing on my part has found that the LWP::Parallel::UserAgent is more than happy to associate an empty HTTP::Response object (status code 0, content="<empty response>") to the current request entry. To circumvent this, I added an additional test for the empty response before sending it back.
The attached patch entirely replaces the previously supplied patch. I'm sorry for any inconveniences caused if you've applied the other patch. I promise to be more careful next time :-) As always, I'm interested in any feedback from those who may test this patch. Humbly, Andy Levine > -----Original Message----- > From: Andy Levine [mailto:[EMAIL PROTECTED] > Sent: Monday, March 08, 2004 3:50 PM > To: [EMAIL PROTECTED] > Subject: PATCH:PoCoCl::UserAgent and error response handling during connec > t phase > > Attached is a patch which fixes some problems we are having using > PoCoCl::UserAgent with the parallel ftp: and file: protocol handlers. Also > attached is a test script ("borrowed" liberally from the perldoc page for > PoCoCl::UserAgent) that illustrates the problem. I am soliciting input as > to > the general suitability of the patch and whether I missed anything. > > Versions: > --------- > POE - v 0.2802 > POE::Component::Client::UserAgent - v 0.05 > LWP - 5.76 > ParallelUserAgent - 2.57 > > Description of problem: > ----------------------- > During the request initiation stage, handled in _pococ_ua_write, it is > possible for the underlying protocol handler to return a response. In > cases > where the responses are non-error responses, all will be handled > appropriately. However, in cases where the returned responses are error > conditions, a couple of problems occur: > 1) By not processing the error, eventually the request will time out. This > will cause an error response with return code RC_REQUEST_TIMEOUT to be > generated, thus masking the real error returned by the underlying protocol > handler. > 2) If the timeout handler (_pococ_ua_timeout) is called by lack of > response > processing, the process will die(). This is because the write handler, > _write_request in LWP::Parallel::UserAgent has already cleared the current > connection for the entry. When the timeout occurs, the timeout handler > again > tries to clear the current connection for the entry, causing > LWP::Parallel::UserAgent to Carp::confess("No connections for '$netloc'") > in > _remove_current_connection. > > To replicate problem: > --------------------- > Use attached test script as follows: > ./test-ua FTP-URI [debuglevel] > > Make sure the FTP URI points to a valid FTP server that REQUIRES > authentication, but pass either no credentials OR pass invalid ones. > Debuglevel can be turned on using level 1-9. > > Example: ./test-us ftp://baduser:[EMAIL PROTECTED]/foo.txt 9 > > > Example output of failed run: > ----------------------------- > $ ./test-ua.pl ftp://foo:[EMAIL PROTECTED]/foo.txt > Retrieving: ftp://foo:[EMAIL PROTECTED]/foo.txt > No connections for 'core-001.somecompany.com:21' at > /usr/lib/perl5/site_perl/5.8.0/LWP/Parallel/UserAgent.pm line 776 > > LWP::Parallel::UserAgent::_remove_current_connection('POE::Component::Clie > nt > ::UserAgent=HASH(0xa4456cc)','LWP::Parallel::UserAgent::Entry=HASH(0xa477b > 34 > )') called at > /usr/lib/perl5/site_perl/5.8.0/POE/Component/Client/UserAgent.pm line 300 > > POE::Component::Client::UserAgent::_pococ_ua_timeout('POE::Component::Clie > nt > ::UserAgent=HASH(0xa4456cc)','POE::Session=ARRAY(0xa4480e8)','POE::Kernel= > AR > RAY(0xa1a7c5c)','HASH(0xa4480a0)','timeout','POE::Session=ARRAY(0xa4480e8) > ', > 'undef','/usr/lib/perl5/site_perl/5.8.0/POE/Component/Client/UserAgent.pm' > ,1 > 24,...) called at /usr/lib/perl5/site_perl/5.8.0/POE/Session.pm line 629 > > POE::Session::_invoke_state('POE::Session=ARRAY(0xa4480e8)','POE::Session= > AR > RAY(0xa4480e8)','timeout','ARRAY(0xa579aec)','/usr/lib/perl5/site_perl/5.8 > .0 > /POE/Component/Client/UserAgent.pm',124) called at > /usr/lib/perl5/site_perl/5.8.0/POE/Kernel.pm line 843 > > POE::Kernel::_dispatch_event('POE::Kernel=ARRAY(0xa1a7c5c)','POE::Session= > AR > RAY(0xa4480e8)','POE::Session=ARRAY(0xa4480e8)','timeout',512,'ARRAY(0xa57 > 9a > ec)','/usr/lib/perl5/site_perl/5.8.0/POE/Component/Client/UserAgent.pm',12 > 4, > 1078776576.92054,...) called at > /usr/lib/perl5/site_perl/5.8.0/POE/Resource/Events.pm line 239 > POE::Kernel::_data_ev_dispatch_due('POE::Kernel=ARRAY(0xa1a7c5c)') > called at /usr/lib/perl5/site_perl/5.8.0/POE/Loop/Select.pm line 321 > POE::Kernel::loop_do_timeslice('POE::Kernel=ARRAY(0xa1a7c5c)') > called at /usr/lib/perl5/site_perl/5.8.0/POE/Loop/Select.pm line 329 > POE::Kernel::loop_run('POE::Kernel=ARRAY(0xa1a7c5c)') called at > /usr/lib/perl5/site_perl/5.8.0/POE/Kernel.pm line 1009 > POE::Kernel::run('POE::Kernel=ARRAY(0xa1a7c5c)') called at > ./test-ua.pl line 24 > > Problem resolution: > ------------------- > The attached patch fixes the above issue by checking for error responses > in > _pococ_us_write before returning. If an error response exists, it is > forwarded on to the failure handler, on_failure. > > The patch also sets the 'use_eval' flag within the UserAgent object so all > calls to the protocol handlers will be wrapped. Without this, there > appears > to be no way to catch calls to die() emanating from the handlers. > > Please consider this patch for inclusion in some future release of > PoCoCl::UserAgent. > > Thanks, > > Andy Levine > Principal Software Engineer > dynamicsoft, Inc. >
UserAgent.pm.patch
Description: Binary data
