Mithaldu was able to do some debugging, and made some very interesting discoveries:
16:26 <@ether> Mithaldu: are you still awake?
16:26 < Mithaldu> fsvo awake
16:26 <@ether> I just pushed this test branch --
https://github.com/karenetheridge/dist-zilla-plugin-test-compile/tree/topic/win32_debugging
16:26 <+dipsy> [ karenetheridge/dist-zilla-plugin-test-compile · GitHub ]
16:27 <@ether> or, copy/paste from the last message in
https://rt.cpan.org/Ticket/Display.html?id=88783
16:27 <+dipsy> [ Bug #88783 for Dist-Zilla-Plugin-Test-Compile: generated test sometimes hangs on Windows ]
16:27 < Mithaldu> useful, thank you :)
16:29 < Mithaldu> ether: so which part of this is not supposed to be changed?
16:29 < Mithaldu> also, it doesn't hang for me
16:32 < Mithaldu> ether: ok wait
16:32 < Mithaldu> so if i read your ticket comment right the issue happens when Foo.pm opens STDIN and tries to read from it until it gets EOF
16:32 < Mithaldu> is that the right ballpark?
16:32 <@ether> one commit back is the change I made that I hope will work
16:33 <@ether> yes I think so
16:33 <@ether> I didn't think that any processes woudl be reading from stdin, but apparently something is
16:33 < Mithaldu> see, i need that code
16:33 < Mithaldu> i need to test the hang
16:33 < Mithaldu> on which modules does this happen?
16:33 < Mithaldu> ok, i see
16:34 <@ether> chorny said "I just used WWW-Shorten-SCK repository to test this by adding "use No::Such::Module;". It does not help. "
16:34 <@ether> so I suppose he just added that line to one of the modules in the dist?
16:34 <@ether> which is what I'm simulating with Foo.pm
16:35 < Mithaldu> man, his information is really awkwardly unfocused
16:36 <@ether> it doesn't seem like the failure is always repeatable, either :(
16:38 < Mithaldu> i'm trying with the SCK thing he mentioned
16:40 <@xdg> ether, read the second bullet in the "BUGS" section of perlfork
16:42 <@ether> pseudo-processes being what we get in win32?
16:42 * ether reads the doc from the beginning; ok, this is just about emulation, not normal forks, so yeah
16:42 < Mithaldu> fork under windows gives you threads
16:43 < Mithaldu> but open3 doesn't fork on windows i think
16:43 < Mithaldu> so i don't understand why that's relevant
16:43 <@xdg> ether, fork on windows does what the document describes and IPC::Open3 uses fork internally
16:43 <@xdg> forks on windows are "pseudo-threads"
16:43 <@ether> I could always just drop the idea of using separate processes for testing on windows -- it may generate different errors, and mask others, but at least it shouldn't go kaboom
16:44 <@xdg> Capture::Tinier
16:44 <@ether> if ($^O eq 'MSWin32') { use_module($_) foreach @modules }
16:44 < Mithaldu> xdg: open3 doesn't
16:44 <@ether> xdg: or that. yes. :)
16:45 < Mithaldu> (i know this because i actually see the child process)
16:47 <@xdg> ah.
16:47 < Mithaldu> ether: i'll try to reproduce this with the two modules he pointed out
16:47 <@xdg> In any case, Capture::Tiny(ier) doesn't try to read from a pipe, which simplifies the IPC
16:47 <@ether> I just pushed an update, bypassing cmd.exe (which I already released yesterday; I copied this test from a different build that was created before that change.)
16:48 < Mithaldu> well that's not gonna help me help you, because i can't reproduce the error case yet :P
16:51 < Mithaldu> ok, reproduced
16:51 < Mithaldu> awesome
16:51 <@xdg> Mithaldu++
16:51 <@ether> o/
16:51 < Mithaldu> just downloaded this:
http://cpan.metacpan.org/authors/id/T/TR/TRCJR/MooseX-Role-Net-OpenSSH-0.001.tar.gz
16:51 <+dipsy> urgh. long url. Try
http://tinyurl.com/mt3fvum
16:51 < Mithaldu> and then make tested without having openssh
16:52 <@ether> I smell some new unit tests that I should be adding :)
16:52 < Mithaldu> open my $stdin, '<', File::Spec->devnull or die $!;
16:52 < Mithaldu> doesn't help
16:53 < Mithaldu> interestingly
16:53 < Mithaldu> perl -Ilib t\00-compile.t
16:53 <@ether> it's frustrating having the extra step in between for testing - I can't just put up a -TRIAL dist and see what the smokers say - I have to wait until people actually use the plugin to build their dists, and then find out about it secondhand... so having a test that I can actually use myself to reproduce these issues will be VERY helpful
16:53 < Mithaldu> ^- that works
16:53 < Mithaldu> prove -l t/*
16:53 < Mithaldu> ^- hangs
16:53 < Mithaldu> so the issue is processes wrapping around further outside
16:53 < Mithaldu> and i remember things like that
16:53 <@ether> prove closes stdout, iirc
16:54 <@ether> (well, Test::Harness actually..)
16:54 < Mithaldu> but i don't remember any details
16:54 < Mithaldu> it's more of a deja vu
16:54 < Mithaldu> anyhow, i'll mess with this and see what happens
16:54 < Mithaldu> xdg: where is tinier?
16:54 <@xdg> ether, if that's true then Capture::Tinier might be too tiny
16:54 < Mithaldu> ok, got tinier
16:54 < Mithaldu> trying that
16:55 < Mithaldu> ether: yes, i'm working on that :)
16:55 <@xdg> Mithaldu, you got it? If not:
https://gist.github.com/dagolden/6618205
16:55 <@ether> tinier:
https://gist.github.com/dagolden/6618205
16:55 <@ether> too slow
16:55 <+dipsy> [ 00-xdg-compile.t ]
16:55 <+dipsy> [ 00-xdg-compile.t ]
16:55 <@xdg> It was sort of a joke name, but with real intent
16:56 <@xdg> It's Capture::Tiny stripped of all the tee-ing, layer mgmt, and dealing with weirdly modified existing handles
16:56 <@ether> dammit, we should just get it into core :)
16:57 <@xdg> I do notice a line there about "# WIN32 hangs on tee without STDIN copied"
16:57 < Mithaldu> xdg: yes got it
16:57 <@xdg> shouldn't matter for Tinier, but it's indicative of the sort of problems we hit on Windows with process stuff
16:58 * ether sees that Meerkat is a real thing ;)
16:58 <@xdg> it is now
16:58 <@xdg> :-)
16:59 < Mithaldu> .... what the fuck
17:00 <@ether> that sounds ominous
17:00 < Mithaldu> deleting blib\arch\auto\MooseX\Role\Net\OpenSSH\.exists stops prove -l -v t/* from hanging
17:01 <@ether> !
17:01 <@ether> I didn't think anything at runtime looked at .exists files
17:02 < Mithaldu> blib/arch/auto/MooseX/Role/Net/OpenSSH/.exists
17:02 <@xdg> o_O
17:02 < Mithaldu> blib/lib/MooseX/Role/Net/OpenSSH.pm
17:02 < Mithaldu> t/00-compile.t
17:02 < Mithaldu> i run
17:02 < Mithaldu> prove -l -v t/*
17:02 < Mithaldu> and it hangs
17:02 < Mithaldu> i delete the .exists
17:02 < Mithaldu> and the test doesn't hang
17:02 < Mithaldu> let's see how many bits i can delete from OpenSSH.pm
17:03 <@xdg> what is generating .exists?
17:03 < Mithaldu> dzil
17:03 < Mithaldu> or eumm
17:03 < Mithaldu> unclear
17:03 <@xdg> if it's in blib, it's not dzil
17:03 <@xdg> eumm or a lib it's using to copy to blib
17:03 * xdg chdirs
17:04 <@xdg> eumm
17:04 <@xdg> I'm going to skim source
17:04 < Mithaldu> thanks
17:05 <@xdg> it's just getting touched by the makefile so it can have a dependency on the directory, which otherwise doesn't work well
17:06 <@xdg> (that's my tl;dr summary of a comment in the source)
17:08 * xdg afk
17:09 < Mithaldu> aHA
17:11 < Mithaldu> ether:
https://github.com/wchristian/winhangtest
17:11 <+dipsy> [ wchristian/winhangtest · GitHub ]
17:11 < Mithaldu> still making it smaller
17:13 <@ether> if $stderr isn't an IO::Handle there isn't a problem? wut?
17:13 <@ether> this is getting stranger and stranger
17:14 <@ether> in the meantime I'm whipping up a capture_tinier branch
17:14 < Mithaldu> my ( $stdout, $stderr, $rc ) = capture { system( qq{$^X -Mblib -e"require q[MooseX/Role/Net/OpenSSH.pm]"} ) };
17:15 < Mithaldu> this don't freeze and prints all the right things
17:15 < Mithaldu> so, tinier is the right path
17:16 < Mithaldu> output of:
17:16 < Mithaldu> warn "\n$rc - $stdout - $stderr\n";
17:16 < Mithaldu> 512 - - Can't locate Not/Exists.pm in @INC (@
17:16 < Mithaldu> etc.
17:17 <@ether> woo
17:18 < Mithaldu> and i got tinier working
17:18 < Mithaldu> ether:
https://github.com/wchristian/winhangtest/blob/master/t/00-compile.t
17:18 <+dipsy> [ winhangtest/t/00-compile.t at master · wchristian/winhangtest · GitHub ]
17:18 < Mithaldu> enjoy
17:18 < Mithaldu> also
17:19 < Mithaldu> do test that on linux to see if it blows up :)
17:19 < Mithaldu> xdg: splendid work, thanks a bunch!
17:19 <@ether> thank you so much to you both
17:19 < Mithaldu> who needs sleep when he can make cpan happy? :D
17:20 <@ether> so the "ssh library doesn't exist" issue was a red herring?
17:20 < Mithaldu> totally
17:20 <@ether> I'm wondering if there's something I can put into a .pm file to be tested that will tickle the right error
17:20 < Mithaldu> actually, let me try something dumb
17:21 < Mithaldu> ok
17:21 < Mithaldu> if i replace use Not::Exists; in OpenSSH.pm with "die 1;", then it doesn't hang
17:22 < Mithaldu> so this is some weird interaction between the child process failing to load a dep, and whatever prove invokes
17:22 < Mithaldu> and the .exists file
17:22 < Mithaldu> those are the three puzzle pieces
17:23 <@ether> I do do a perl Makefile.PL; make from within my tests..
17:23 <@ether> only right now I only have passing tests, because of the PITA that is Test::Tester
17:23 <@ether> and using it might further alter the harness in undesirable ways
17:24 <@ether> Mithaldu: thanks so much
17:25 < Mithaldu> i'm unclear on the implications of what you're saying
17:25 <@ether> I'd like to be able to have this particular type of test failure in the plugin's tests
17:25 <@ether> so I can put up a -TRIAL release and see results right away
17:25 < Mithaldu> yeah, that's gonna be tricky
17:25 <@ether> but that means capturing the test results in a way that the thing doing the testing doesn't also fail
17:25 <@ether> which generally implies Test::Tester
17:26 <@ether> which I have fought with before, and mostly beaten into submission, but it does funny things with the harness so I don't trust it too much
17:26 < Mithaldu> well
17:26 < Mithaldu> a test failure is freezing
17:26 < Mithaldu> that is dang hard to do anything with
17:26 < Mithaldu> i wouldn't bother
17:26 < Mithaldu> you're removing open3, which is the source of the immediate problem
17:28 < Mithaldu> none of the code in my perl dir should even touch .exists
17:28 <@ether> and hopefully the last (ha ha!) :)
RT for rt.cpan.org