REOPENED177411
REGRESSION(r214201): WebProcess hangs during policy decisions
https://bugs.webkit.org/show_bug.cgi?id=177411
Summary REGRESSION(r214201): WebProcess hangs during policy decisions
Alex Christensen
Reported 2017-09-23 12:50:37 PDT
REGRESSION(r214201): WebProcess hangs during policy decisions
Attachments
Patch (44.31 KB, patch)
2017-09-23 16:44 PDT, Alex Christensen
no flags
Patch (42.41 KB, patch)
2017-10-06 13:57 PDT, Alex Christensen
no flags
Alex Christensen
Comment 1 2017-09-23 16:44:28 PDT
Geoffrey Garen
Comment 2 2017-09-23 18:08:23 PDT
Comment on attachment 321641 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=321641&action=review r=me > Source/WebKit/UIProcess/WebPageProxy.cpp:3756 > + m_decidePolicyForResponseRequest = 0; Should be nullptr. Why did m_decidePolicyForResponseRequest change to a pointer?
Alex Christensen
Comment 3 2017-09-23 23:21:13 PDT
That's how it used to be. It was an awful mess.
Alex Christensen
Comment 4 2017-09-23 23:23:16 PDT
Alex Christensen
Comment 5 2017-09-25 12:03:16 PDT
Radar WebKit Bug Importer
Comment 6 2017-09-27 12:19:44 PDT
Alex Christensen
Comment 7 2017-10-06 13:57:52 PDT
Reopening to attach new patch.
Alex Christensen
Comment 8 2017-10-06 13:57:53 PDT
mitz
Comment 9 2017-10-06 14:03:28 PDT
The API must behave consistently across platforms and correctly on all platforms.
Alex Christensen
Comment 10 2017-10-06 14:22:38 PDT
(In reply to mitz from comment #9) > The API must behave consistently across platforms and correctly on all > platforms. I agree wholeheartedly. Unfortunately, that would require https://bugs.webkit.org/show_bug.cgi?id=160677 to land to allow us to wait during redirects and that's WAAAAY too risky for the branch.
mitz
Comment 11 2017-10-06 16:35:27 PDT
(In reply to Alex Christensen from comment #10) > (In reply to mitz from comment #9) > > The API must behave consistently across platforms and correctly on all > > platforms. > I agree wholeheartedly. Unfortunately, that would require > https://bugs.webkit.org/show_bug.cgi?id=160677 to land to allow us to wait > during redirects and that's WAAAAY too risky for the branch. Not at all. Your change log says “[…] Mac, which depends on the ability to asynchronously call the completion handlers” but that’s incorrect. There is only one client of the API on macOS that, because it adapted to the changes introduced in r214201, depends on that behavior. All other software expects and works with the long-standing behavior of the API as it was prior to r214201 on all platforms. That behavior must be reinstated. Because one client has come to depend on incorrect behavior, that one client can be accommodated by maintaining the some of all of the incorrect behavior as needed for that client.
Note You need to log in before you can comment on or make changes to this bug.