Summary: | check-webkit-style failing with "Path does not exist." | ||||||||
---|---|---|---|---|---|---|---|---|---|
Product: | WebKit | Reporter: | Philip Rogers <pdr> | ||||||
Component: | Tools / Tests | Assignee: | Dirk Pranke <dpranke> | ||||||
Status: | RESOLVED FIXED | ||||||||
Severity: | Normal | CC: | abarth, dpranke, levin, ojan, webkit.review.bot | ||||||
Priority: | P2 | ||||||||
Version: | 528+ (Nightly build) | ||||||||
Hardware: | Unspecified | ||||||||
OS: | Unspecified | ||||||||
Attachments: |
|
Description
Philip Rogers
2012-02-06 07:31:00 PST
fast/table/027.html at least does seem to exist. At a quick glance, it looks like the problem is that it's not correctly mapping the "win" in the test_expectations.txt path to a port. The "test" port is the fallback port when the port can't be identified. IIRC, Dirk already has a patch somewhere fixing this, but I don't know what the status of it is. (In reply to comment #2) > At a quick glance, it looks like the problem is that it's not correctly mapping the "win" in the test_expectations.txt path to a port. The "test" port is the fallback port when the port can't be identified. IIRC, Dirk already has a patch somewhere fixing this, but I don't know what the status of it is. Hm. I think Ojan might be referring to bug 76745, which is a bit different (at least, that's the only patch I've worked on that is ringing a bell for me). That said, fixing this is straightforward. The mapping from LayoutTests/platform/$name to port is loose at best; the code in lines 74-81 is wrong and what we should do is fetch all of the ports using PortFactory.all_port_names() and then grep the list until you find one whose path_to_test_expectations_file() matches. I will upload a patch for this shortly. Created attachment 125724 [details]
Patch
Here's a patch that fixes the lookup for all of the real test_expectations.txt files; however, I'm not sure what the right thing to do is if we attempt to check the style for (say) LayoutTests/platform/foo/test_expectations.txt (since there is no foo port) ... the existing code will use the 'test' port, which seems a bit weird at best? It seems like we should either raise an assertion that we're trying to check an unknown file, or have some other sort of no-op checker? Comment on attachment 125724 [details]
Patch
Seems fine.
Created attachment 126137 [details]
test bad paths better
Committed r107124: <http://trac.webkit.org/changeset/107124> I just wanted to thank you (Dirk) and Ojan for the quick fixes for these bugs! You guys rock |