Bug 29329 – Tiger fails test http/tests/loading/redirect-with-no-location-crash.html added in r48413

• Status: RESOLVED
• Resolution: WONTFIX
• Component: New Bugs

[Eric Seidel (no email)]

REGRESSION: http/tests/loading/redirect-with-no-location-crash.html failing on Tiger bot

Example:
http://build.webkit.org/results/Tiger%20Intel%20Release/r48458%20(4387)/results.html

CCing the folks I see from:
http://trac.webkit.org/browser/trunk/LayoutTests/http/tests/loading/redirect-with-no-location-crash.html

[Eric Seidel (no email)]

Tiger may just need new results checked in.

[Eric Seidel (no email)]

http://trac.webkit.org/changeset/48413 is the regressing change (where the test was added).

[Xan Lopez]

(In reply to comment #2)
> http://trac.webkit.org/changeset/48413 is the regressing change (where the test
> was added).

How can it be a regression being this a new test? I think you were right in your previous comment where you say that probably Tiger just needs different results checked in.

[Gustavo Noronha (kov)]

It's not a regression, Tiger fails the new test, it needs to be fixed to pass it, having a bug report is important though, thanks!

[Gustavo Noronha (kov)]

(In reply to comment #3)
> How can it be a regression being this a new test? I think you were right in
> your previous comment where you say that probably Tiger just needs different
> results checked in.

I would say it needs to be either fixed or skipped. I don't think checking in a failed load as expected result is the way to go here.

[Eric Seidel (no email)]

Well, it's a regression in the sense that the bots were green before the checkin and are red afterwards. :) Either way, we need to check in new results or roll out the change.  I don't know which results are correct, or why Tiger's would be different from normal mac results.

[Eric Seidel (no email)]

Either way, we can't leave the bots green. :)  So action is needed asap.  Sounds like you guys are on it, which is great.  Thanks!

[Gustavo Noronha (kov)]

(In reply to comment #6)
> Well, it's a regression in the sense that the bots were green before the
> checkin and are red afterwards. :) Either way, we need to check in new results
> or roll out the change.  I don't know which results are correct, or why Tiger's
> would be different from normal mac results.

Or skip the test, which is what we do when we fail new tests in GTK+, I would
say - we don't roll them back because our code is wrong. Checking in different
results would be OK if the expected result for a load that results in a
Location-less 302 response is for it to fail in Tiger.

If this is actually the result of a different behavior in, say, Tiger's
CFNetwork that is not going to be fixed, or if a fix is not planned for the
short term, skipping would be the option, rolling out the change doesn't seem
sensible to me.

I have taken action to fix the other rednesses the change caused because they were caused by bad expected results, but I am not in a position to decide what to do for this test for Tiger, so we need input from a Mac/Tiger person.

[Brady Eidson]

This issue with 302 is a known bug with the CFNetwork from the Tiger era that was fixed by the time Leopard was released.

Apple folks can refer to <rdar://problem/5081884> for details.

My recommended course of action is to add this test to the Skipped list for Tiger until we can explore a workaround.

[Gustavo Noronha (kov)]

(In reply to comment #9)
> This issue with 302 is a known bug with the CFNetwork from the Tiger era that
> was fixed by the time Leopard was released.
> 
> Apple folks can refer to <rdar://problem/5081884> for details.
> 
> My recommended course of action is to add this test to the Skipped list for
> Tiger until we can explore a workaround.

I have landed r48487, skipping the test for Tiger, thanks for the comment, and the rubber stamp, and Eric for bringing this up! I'll leave the bug open.

[Eric Seidel (no email)]

Thank you for the quick response.

Tiger is no longer supported.