Bug #78297 for File-DataClass: dist pretends to work on Win32 when it cannot

Tue Jul 10 12:58:05 2012 MITHALDU [...] cpan.org - Ticket created

Subject:

dist pretends to work on Win32 when it cannot

Please don't just make your tests skip on win32 if the module cannot work on win32. Instead just do this in your Build.PL: exit if $^O eq "MSWin32";

Tue Jul 10 13:10:36 2012 Support [...] RoxSoft.co.uk - Correspondence added

The distribution can and does work on MSWin32 if *ALL* the dependencies are installed. CPAN testing and only CPAN testing is skipped on this platform because MSWin32 smokers randomly fail to install dependencies

Tue Jul 10 13:10:36 2012 The RT System itself - Status changed from 'new' to 'open'

Tue Jul 10 13:10:37 2012 Support [...] RoxSoft.co.uk - Status changed from 'open' to 'rejected'

Tue Jul 10 13:15:57 2012 MITHALDU [...] cpan.org - Correspondence added

Can you give me a list of dependencies that randomly fail to install so i can investigate reasons for that? Also, even in such a case the result of a test should be NA (generated by an exit in the build phase) to indicate that installing it might be harder than usual, instead of a PASS.

Tue Jul 10 13:15:58 2012 The RT System itself - Status changed from 'rejected' to 'open'

Tue Jul 10 13:40:36 2012 Support [...] RoxSoft.co.uk - Correspondence added

On Tue Jul 10 13:15:57 2012, MITHALDU wrote: Show quoted text

> Can you give me a list of dependencies that randomly fail to install so i > can investigate reasons for that? > > Also, even in such a case the result of a test should be NA (generated by > an exit in the build phase) to indicate that installing it might be > harder than usual, instead of a PASS.

You can get a list of random failure by working backwards through the test results for all of the versions of this distribution. The failures that you cannot explain, those are the ones you are looking for.

Tue Jul 10 13:40:37 2012 Support [...] RoxSoft.co.uk - Status changed from 'open' to 'rejected'

Fri Aug 03 20:55:47 2012 dagolden [...] cpan.org - Correspondence added

I understand how frustrating it is to get random Win32 failures. I still get spurious failures for Capture::Tiny that I've never been able to track down. That said, for the purpose of test reporting, it's much more helpful to end users to report NA than PASS, because that signals that tests didn't actually run on that platform and people won't falsely expect things to just work and then have a nasty surprise. They'll go in more forewarned. You can do that by having the phrase "OS unsupported" in the test output for Windows. E.g. diag("OS unsupported -- you must manually confirm all dependencies for this to work") if $^O eq 'MSWin32' That means that if tests *work* then a PASS will get reported, but if tests fail, the report will get downgraded from FAIL to NA. More aggressively, you could just skip CPAN Testers reporting entirely: # in Makefile.PL/Build.PL exit 0 if $ENV{AUTOMATED_TESTING} && $^O eq 'MSWin32'; That means that *real* users can still work with your module as usual, but you won't get hassled by CPAN Testers on Windows. I encourage you to consider those options instead of just skipping and reporting a misleading PASS.

Fri Aug 03 20:55:49 2012 The RT System itself - Status changed from 'rejected' to 'open'

Sat Aug 04 08:32:11 2012 Support [...] RoxSoft.co.uk - Correspondence added

On Fri Aug 03 20:55:47 2012, DAGOLDEN wrote: Show quoted text

> "dist pretends to work on Win32 when it cannot"

No. If you force install IPC::SRLock this distribution should work on MSWin32 You have to force install IPC::SRLock because you cannot express conditional dependencies in META.json: if ($^O ne 'MSWin32') { # Require IPC::ShareLite } IPC::ShareLite is obviously not supported on MSWin32 Show quoted text

> You can do that by having the phrase "OS unsupported" in the test output > That means that if tests *work* then a PASS will get reported, but if > tests fail, the report will get downgraded from FAIL to NA.

The text of the reason for skipping tests will be changed in the next release to contain "OS unsupported". This will not effect the outcome Show quoted text

> More aggressively, you could just skip CPAN Testers reporting entirely: > > # in Makefile.PL/Build.PL > exit 0 if $ENV{AUTOMATED_TESTING} && $^O eq 'MSWin32'; > > That means that *real* users can still work with your module as usual, > but you won't get hassled by CPAN Testers on Windows.

We have had this discussion before. Doing this resulted in FAIL reports from Oliver Paukstadt. His explanation was (if memory serves) "user error, hehe". It happened more than once and I have no reason to believe that this situation has changed If you wish to continue discussing ways in which the CPAN Testing service can be improved please do so by email and not via RT

Sat Aug 04 08:32:12 2012 Support [...] RoxSoft.co.uk - Status changed from 'open' to 'rejected'