It's unfortunate that it's so easy to check in a test_expectations.txt file that doesn't work either due to a typo or due to two rules that conflict with each other in specificity. It'd be nice if folks were warned that they were about to do such a thing.
Another approach is to simply test_expectations to the point where it cannot have parse errors.
(In reply to comment #1) > Another approach is to simply test_expectations to the point where it cannot have parse errors. This bug was prompted by http://trac.webkit.org/changeset/85950, which fixed what was called a parse error, but it was really a cuplicate expectation where one was more specific than the other. On further inspection, it looks like check-webkit-style will complain about "duplicate or ambiguous expectation" but only if both offending lines are in the patch. Maybe it needs to consider the whole file?
I don't believe it is possible to have parse errors in the Skipped files, which are another approach to the same problem. Yes, I know Skipped files do not have all the features of test_expectations.
(In reply to comment #2) > (In reply to comment #1) > > Another approach is to simply test_expectations to the point where it cannot have parse errors. > > This bug was prompted by http://trac.webkit.org/changeset/85950, which fixed what was called a parse error, but it was really a cuplicate expectation where one was more specific than the other. > > On further inspection, it looks like check-webkit-style will complain about "duplicate or ambiguous expectation" but only if both offending lines are in the patch. Maybe it needs to consider the whole file? In general, check-webkit-style only complains about the line that you changed. However, I'm sure one could change this for a particular checker. Patches welcome :).
I think this should probably be merged into bug 51548, and I agree that the checker needs to consider the whole file, not just the lines that are modified.
This particular issue (of filtering errors) came up for another checker as well, so I'm fixing both in one bug. *** This bug has been marked as a duplicate of bug 60466 ***