Bug 88436 – REGRESSION (r115654): Sometimes does not replace content for multipart/x-mixed-replace

• Component: Page Loading
• Status: RESOLVED FIXED

[Gustavo Noronha (kov)]

I have a work-in-progress patch for libsoup and the soup backend of WebKit to support multipart/x-mixed-replace. I noticed after I rebased it on top of a more recent WebKit it stopped working. I managed to bisect the problem to the revision mentioned in the summary. Interestingly some tests work right (like the multipart tests in our LayoutTests), but I managed to reproduce it with this toy: https://gitorious.org/kov-play/kov-play/blobs/master/nodejs/multipart-clock.js

I'll try to come up with a test. Would be good if we can confirm the problem in another port.

[Alexey Proskuryakov]

Please always add author and reviewer of an offending change to CC list for regressions.

[Nate Chapin]

(In reply to comment #0)
> I have a work-in-progress patch for libsoup and the soup backend of WebKit to support multipart/x-mixed-replace. I noticed after I rebased it on top of a more recent WebKit it stopped working. I managed to bisect the problem to the revision mentioned in the summary. Interestingly some tests work right (like the multipart tests in our LayoutTests), but I managed to reproduce it with this toy: https://gitorious.org/kov-play/kov-play/blobs/master/nodejs/multipart-clock.js
> 
> I'll try to come up with a test. Would be good if we can confirm the problem in another port.

I'll look at this tomorrow. Any hints you have or guidance as to how it's failing will be greatly appreciated :)

[Nate Chapin]

Created attachment 146384 [details]
Patch (without test)

Still looking for a proper layout test....

[Nate Chapin]

(In reply to comment #3)
> Created an attachment (id=146384) [details]
> Patch (without test)
> 
> Still looking for a proper layout test....

I have a test that works correctly in a full debug build of Safari, but asserts pretty regularly in WTR because of a null Internals* on the JS context during teardown.

I haven't figured out yet whether that's a test infrastructure bug or a symptom of something sinister. Is this familiar to anyone?

[Alexey Proskuryakov]

The code to remove Internals on teardown in WTR is new (I added it last week), and I seen an assertion below it a few times. Unfortunately, I cannot investigate this now.

If you have a mostly reproducible case, and can find out the reason for these assertions, that would be extremely helpful. Perhaps the method is called twice somehow?

[Nate Chapin]

Created attachment 147623 [details]
patch + test

[Brady Eidson]

In radar as <rdar://problem/11801704>

[WebKit Review Bot]

Comment on attachment 147623 [details]
patch + test

Clearing flags on attachment: 147623

Committed r121813: <http://trac.webkit.org/changeset/121813>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.

[WebKit Review Bot]

Re-opened since this is blocked by 90522

[Brady Eidson]

The rollout in bug 90522 was invalid as it wasn't based in any failures here at the WebKit project.

Since this patch fixed an important regression here in WebKit and did not cause any known new regressions here in Webkit, rolling this back in.

[WebKit Review Bot]

Comment on attachment 147623 [details]
patch + test

Clearing flags on attachment: 147623

Committed r121912: <http://trac.webkit.org/changeset/121912>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.