Windows Unicode Character Prop fix, update appveyor, travis

[MSP-Greg]

I ended up here while trying to test the rdoc repo under windows. It uses kpeg to build two files.

There is a quirk in windows regexp where a Unicode character property like \p{**} can be used in a hard coded regexp, but one can't create one unless the encoding of it is UTF-8. So, this commit fixes that, in addition to several minor fixes:

1. .travis.yml - added current 2.2 thru 2.4.

2. appveyor.yml - added it, passes 2.0 thru trunk. See here. Going forward, if you don't want to set it up but would like a check, ping me and I can run one.

3. lib/kpeg/format_parser.rb - commented out two unused variables.

4. lib/kpeg/grammar.rb - this contains the windows patch re encoding. Tried to keep the constraint on it pretty tight.

5. test/test_kpeg.rb, test/test_kpeg_code_generator.rb - changed some tests over to assert_nil from assert_equal.

6. test/test_kpeg_string_escape.rb - ends with two blank lines, removed one...

Thanks, Greg

[MSP-Greg]

Update - with this PR, RDoc builds on Windows for all versions tested (2.2 - trunk). Without the patch, all versions fail.

[aycabta]

@evanphx This Pull Request is very important for RDoc because of test on AppVeyor fails. Please review this.

[graaff]

This is actually a wider issue that does not affect only Windows. Trying to use kpeg on Linux with a non-UTF-8 locale results in similar issues. This breaks building the rdoc literals. Downstream bug report: https://bugs.gentoo.org/640150

Simple reproducer based on the rdoc source code:

LANG=C ruby -S kpeg -fsv -o lib/rdoc/markdown/literals.rb lib/rdoc/markdown/literals.kpeg

/usr/lib64/ruby/gems/2.3.0/gems/kpeg-1.1.0/lib/kpeg/grammar.rb:133:in `initialize': invalid character property name {Zl}: /\n|\r\n?|\p{Zl}|\p{Zp}/ (RegexpError)
	from /usr/lib64/ruby/gems/2.3.0/gems/kpeg-1.1.0/lib/kpeg/grammar.rb:133:in `new'
	from /usr/lib64/ruby/gems/2.3.0/gems/kpeg-1.1.0/lib/kpeg/grammar.rb:133:in `initialize'

I have generalized the patch to exclude the windows test and instead force encoding when it is not Encoding::UTF_8. This works as expected and allows things to work with LANG=C.

[evanphx]

I realize it's been absolutely forever since this PR was opened. I'm happy to fix it up if we still need it, please let me know @MSP-Greg.

[MSP-Greg]

No problem. Obviously, GitHub provides a pretty good UI for reviewing one's open PR's, and I need to use it.

This came out of running CI for Windows in RDoc, which is being done currently. Let me look into it. If it's still needed, maybe change to GitHub Actions?

[evanphx]

Let me know, happy to help figure it out.

[graaff]

I've just checked that my reproducer mentioned earlier still causes the same issue with kpeg 1.3.2 and the rdoc 6.4.0 sources on ruby 3.0.