Make extract-tests.py use the same test directives as rustdoc's test extractor

[brson]

Right now to rustdoc supports ignore for marking tests as broken (same as rustc) and should_fail for indicating tests that are supposed to fail. The script we use for extracting tests from the standalone docs supports xfail-test for marking broken tests, ignore and notrust for completely ignoring the code block, and doesn't support should_fail at all.

Make extract-tests.py behave the same as rustdoc.

Alternately, just process all our standalone docs through rustdoc.

[wting]

When you mention should_fail, are you referring to code block directives (~~~ { .should_fail }) or function attribute (#[should_fail])?

If should_fail is found, what do you want done?

[brson]

@wting I'd like to add { .should_fail } to the directives that extract-tests.py understands, to match #[should_fail] in unit tests and ````rust,should_fail` in rustdoc tests. Right now there are a number of tests in the docs that are disabled because they are expected to fail.

I'm not sure the best way to fix this actually because of the way these tests are run. Currently what happens is that they are extracted to a directory and then the compiletest test driver is run on them in 'run-pass' mode, which means they are all expected to pass.

Probably the simplest way to get this working would be to modify compiletest so that it understand a // should_fail (or //should-fail might be more consistent with other compiletest directives) directive and then when it runs the test it checks for a failure error code. extract-tests.py would add //should-fail to the tests it outputs.

Ultimately though I would like all the standalone documentation infrastructure to be folded into rustdoc.

[huonw]

Supporting should_fail is #3509, fwiw.

[wting]

Alright, I've updated extract-tests.py to support { .should_fail } and output //should-fail into the fragment .rs files.