Fix windows

[schwern]

On a fresh Strawberry Perl 5.20.1 install (ie. AppVeyor) this is failing.

I found getting all the paths OS canonical a bit much, so I changed the tests. It's only human readable stuff and gets Windows passing.

I'm concerned that I couldn't find why this started failing. I see from CPAN Testers that it's been passing on Win32. Did something in the dependency chain change? Unfortunately CPAN Testers is erroring when I try to look at the passing results.

[karenetheridge]

Can you show me the output of the failing tests? There have been several win32-specific fixes that have gone in lately, which have seemed to satisfy everyone else, so I'm concerned that applying this change will cause a regression in other win32 systems. Let's instead figure out what's going on.

I know nothing about AppVeyor; what happens when this config file gets added?

[schwern]

Sorry to forget the failure, and to explain AppVeyor. Both should be explained here.
https://ci.appveyor.com/project/schwern/p5-dist-zilla-plugin-run/build/1.0.1

AppVeyor is like Travis, but for Windows and not as well integrated with Github yet. The config provided does a CI test against Strawberry Perl 5.20.1.1 (just happens to be the easiest Perl version to get).

Like Travis, nothing will happen until you go to appveyor.com and turn it on. Unlike Travis, you have to configure Github notifications manually. http://www.appveyor.com/docs/notifications#github-pull-request

Since you're maintaining a huge chunk of the tool chain now, and CPAN Windows testing is notoriously poor, it would be great to have you start using it!

[schwern]

I'm not sure why Travis is failing. It passes here on OS X. OIC it never worked.

[rivy]

I'm encountering similar issues.

Upon digging in, I think the problem is that the "%d" parameter is not using portable path separators.

It appears that Dist::Zilla is using paths in "natural" form (i.e., unix-like, using '/' path separators), not in OS-canonical form. Given that, I think that the paths should be made canonical as late as possible.

There are three repair sites:

1. For the "%d" parameter that would be in the +:::Run:Role::Runner::build_formatter() function.
2. The positional parameter in +::Run::AfterBuild::after_build() would have to be corrected at it's construction site.
3. And, lastly, $build_dir used the the test text within "./t/40_test_phase.t" would probably best be corrected at the point of test text construction.

I've made these repairs in PR#15 (which includes the changes introduced in PR#14 PR#16), for testing on travis-ci).

P.S. Sorry for all the commit noise. I was testing various configurations and didn't realize that the commit message reference would continually ad to the conversation. :\