[tests] Make ninja check-swift 4.1x faster on high-end workstations

[davezarzycki]

No description provided.

[davezarzycki]

@swift-ci please smoke test

[jckarter]

Will PR testing still run this test? Is there any way we can make the test itself faster and still get useful coverage from it?

[davezarzycki]

There might be a better solution. This was just the quickest. Who owns this test?

[davezarzycki]

PS – This test is almost as long as the entire validation test suite on my Linux box. :-(

[benlangmuir]

@nkcsgexi Most of the time in this test is spent in the python code that diffs the results. If I skip the diff, this test takes 10 seconds on my machine. With the diff it's around 80 seconds.

[davezarzycki]

What’s wrong with classic-and-fast diff? Also, 10 seconds without the diffing is okay but not great. That’s still almost half of the entire ninja check-swift time on a fast workstation.

[benlangmuir]

This test could also be trivially parallelized since it's running independently over each file and just reducing at the end to set the return code.

[harlanhaskins]

@davezarzycki IIRC the switch from diff was for Windows support.

[nkcsgexi]

The change looks reasonable to me! In longer term, we plan to rely more on the stress-tester to run SwiftSyntax on user projects; the importance of running on the stdlib source will decrease.

[benlangmuir]

Here's a fix for the difflib slowness by first comparing for == before doing the slow diff. #22653

[davezarzycki]

Abandoned in favor of #22653