I quickly looked at the code in question, and one obvious quick test would be to check the return value of the "unlink" calls (and $! / $ERRNO afterwards) at the two places in the respective .px files where 'needs-auth.txt' is handled.
I've been cleaning up the two tests mentioned but so far have been unable to reproduce the problem with the test file being left over, and I'd be happy to come up with a patch when I'm finished. Until then, I would suggest a quick change to if (-e 'needs-auth.txt') { unlink 'needs-auth.txt' or warn "Cannot remove 'needs-auth.txt: $!"; } would be enough to at least pinpoint the problem if it has to do with file permissions. There are some other potentially problematic constructs in the tests too, (the pipe()/select() calls, for instance, which _might_ cause race conditions in conjunction with the unlink() call), but this is all conjecture on my part; I have not managed to get the tests to fail yet. /Pär 2015-06-15 11:56 GMT+02:00 Darshit Shah <dar...@gmail.com>: > The two proxied https tests: > Test-proxied-https-auth-keepalive.px > Test-proxied-https-auth.px > > Implement a second HTTPS server inside the test itself and do not use > the standard WgetTest.pm class. In these implementations, we do not > actually chdir() to the correct test directory. I've been seeing some > failures in make distcheck, because the file needs-auth.txt sometimes > is left over after the test is run. Currently, I haven't been able to > debug the issue for why the file remains around, but I guess we should > fix the test directory for it first. > > My Perl is ridiculously horrible. So if someone who knows a bit of > Perl can help, I'd be very grateful! Else, I'll sit down over the > coming weekend and make the changes. > > -- > Thanking You, > Darshit Shah > >