Support .only for all files

[naptowncode]

My suggested solution for #549. Needs unit tests, of course.

[mention-bot]

By analyzing the blame information on this pull request, we identified @novemberborn, @ariporad and @sotojuan to be potential reviewers

[naptowncode]

@sindresorhus @vdemedes @jamestalmage Added unit test. Should be ready to go.

[BarryThePenguin]

Is there a test for multiple exclusive tests in multiple files?

[naptowncode]

The test uses three files: One file that has only an exclusive test, one file that has an exclusive test and a non-exclusive test, one file that has two non-exclusive test. Total of five tests across three files with two exclusive tests in different files.

[novemberborn]

I'm wondering how this should work with --watch mode. Let's say we have two test files: foo.js and bar.js. The latter contains a .only.

Initially all test files are run and because bar.js contains a .only, only that test is actually executed.

Subsequently foo.js is changed. The watcher will run just foo.js which does not contain a .only. Now all tests in foo.js are executed.

Later a non-test file is changed and all test files are rerun. Only the .only test from bar.js is executed.

This isn't necessarily bad but it may be surprising to some. I wonder if .only should be sticky so that when changes are detected the watcher still only executes .only tests. But then if you remove the .only it starts executing all tests again. Similarly adding a .only switches back into the sticky mode.

This is a little tricky to implement though #544 requires similar per-test-file housekeeping.

Anyway not meant as a blocker to this PR landing. We can discuss in a separate issue if people think .only should stick across watcher runs.

[novemberborn]

Could you add a comment as to why the function is returning here? It's because if we're running exclusive tests, but the test file does not include exclusive tests, then we don't want those tests to be counted, right?

[naptowncode]

Correct, although I'm not super keen on just dropping this info on the floor. I just went for consistency with how non-exclusive tests are dropped within test-collection. I'll add a comment to that effect.

That said, I would favor changing the reporting completely as suggested by #471.

[novemberborn]

Solution makes sense to me. Great job given the amount of (necessary) indirection involved in coordinating test runs 👍

[naptowncode]

I'm not actually sure how it works with --watch. At the very least I should add some documentation for that.

[novemberborn]

I'm not actually sure how it works with --watch. At the very least I should add some documentation for that.

Watch itself isn't documented yet so don't worry about it. We can get this in and then discuss repercussions in a new issue.

[naptowncode]

@novemberborn I've incorporated those tweaks. If everything looks good, merging this would make it easier to work on #471. Thanks!

[sindresorhus]

LGTM. @vdemedes @jamestalmage ?

[vadimdemedes]

LGTM from me! @naptowncode could you please resolve the merge conflict? Big thanks for your work, this is an important change!

[novemberborn]

You no longer need the _reset() method now that run is always called.

[naptowncode]

@novemberborn @vdemedes @sindresorhus Updated and rebased. Also added a unit test that should have been there earlier for the .runOnlyExclusive option to Runner.run().

[sindresorhus]

@naptowncode Tests are failing on Travis it seems: https://travis-ci.org/sindresorhus/ava/builds/112949149

[novemberborn]

@sindresorhus that's just one test in Node 0.10 though.

[novemberborn]

@naptowncode LGTM!

[vadimdemedes]

Thank you so much, @naptowncode!

[novemberborn]

Please see #593 for how .only should interact in watch mode.