Use nonexistent.127.0.0.1 instead of nonexistent.localhost in our copy of wpt xhr tests
Created attachment 401453 [details] Patch
This patch modifies the imported WPT tests. Please ensure that any changes on the tests (not coming from a WPT import) are exported to WPT. Please see https://trac.webkit.org/wiki/WPTExportProcess
Comment on attachment 401453 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=401453&action=review > LayoutTests/imported/w3c/ChangeLog:3 > + Use nonexistent.127.0.0.1 instead of nonexistent.localhost in our copy of wpt xhr tests Why the "127.0.0.1" bit? This is not being treated as an IP address, so it seems that "nonexistent.0" would also work just as well, no?
It would, but this makes it closer to what exists in the upstream repo, and easier to see the intent of this fake host.
Comment on attachment 401453 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=401453&action=review > LayoutTests/imported/w3c/ChangeLog:10 > + We have some bots where a connection to nonexistent.localhost times out instead of fails. > + To cover this up, use a domain that will actually fail. The tests still won't work as designed, because nonexistent.127.0.0.1 doesn't resolve to loopback on macOS. I think that WPT tests are written with this expired draft in mind <https://tools.ietf.org/html/draft-west-let-localhost-be-localhost>, and work on platforms where localhost subdomains resolve to loopback. This patch will break that, because nonexistent.127.0.0.1 isn't supposed to work this way.
I think that skipping these tests would be OK, even though they found something semi-relevant for us in <rdar://problem/63684261>.
The intent of these tests is to check what happens when an XHR fails. This keeps that. Skipping these tests would just reduce test coverage.
But the intent is certainly not that they are blocked as "external URL" by our script machinery.