|Summary:||[GTK] Regression in imported/w3c/web-platform-tests/fetch/api/basic/request-headers.any.html|
|Product:||WebKit||Reporter:||Alicia Boya García <aboya>|
|Component:||Tools / Tests||Assignee:||Nobody <webkit-unassigned>|
|Severity:||Normal||CC:||cdumez, lforschler, mcatanzaro, youennf|
|Version:||WebKit Nightly Build|
Description Alicia Boya García 2017-08-16 13:11:52 PDT
The following layout test is failing on GTK. imported/w3c/web-platform-tests/fetch/api/basic/request-headers.any.html Probable cause: r220779 introduced a workaround for the Fetch & Beacon API: > Use "application/octet-stream" Content-Type for payloads of type > ArrayBuffer / ArrayBufferView in Fetch & Beacon. If we don't set a > Content-Type header, our underlying network stack is going to add a > "application/x-www-form-urlencoded" Content-Type header, which is worse. > > Chrome is already using "application/octet-stream" Content-Type in Beacon, > but use no Content-Type in Fetch. The Fetch/Beacon specification says we > should have no Content-Type header in this case but this is unfortunately not > something we can support at the moment. Using "application/octet-stream" > Content-Type for now seems like the lesser evil. This workaround is unnecessary in WebKitGTK, which does not have the mentioned limitation. The related tests were passing in WebKitGTK, but this workaround has turned them into failures. We should either: a) Introduce the workaround only for platforms where it is needed. b) Update tests expectations in all ports. I think a) is preferrable in this case, as the workaround is for port-specific issues.
Comment 2 youenn fablet 2017-08-16 13:26:14 PDT
r220779 will probably be rolled out. If not, we should indeed restrict it to platform not being able to comply with the specs.
Comment 3 Chris Dumez 2017-08-16 13:41:45 PDT
Patch is being reworked via https://bugs.webkit.org/show_bug.cgi?id=175628. Will likely land today.