Bug 22524

Summary:

XMLHttpRequest needs to have the static defines made threadsafe.

Product:

WebKit

Reporter:

David Levin <levin>

Component:

XML

Assignee:

Nobody <webkit-unassigned>

Status:

RESOLVED FIXED

Severity:

Normal

CC:

Priority:

Version:

528+ (Nightly build)

Hardware:

All

OS:

All

Attachments:

Description	Flags
Removes the non-thread safe static initializations in XMLHttpRequest.cpp	none
Very minor adjustment to the last patch (-- moves one brace for style purposes).	none
Changed the static initialization to happen in the XHR constructor.	none
Another revision. I think it is close.	ap: review+

David Levin

Reported 2008-11-26 20:41:45 PST

This is part of making XHR available to workers.

Attachments
Removes the non-thread safe static initializations in XMLHttpRequest.cpp (9.42 KB, patch) 2008-11-26 20:47 PST, David Levin	no flags	Details Formatted Diff Diff
Very minor adjustment to the last patch (-- moves one brace for style purposes). (9.42 KB, patch) 2008-11-27 01:00 PST, David Levin	no flags	Details Formatted Diff Diff
Changed the static initialization to happen in the XHR constructor. (9.14 KB, patch) 2008-11-27 23:13 PST, David Levin	no flags	Details Formatted Diff Diff
Another revision. I think it is close. (9.27 KB, patch) 2008-11-28 01:14 PST, David Levin	ap: review+	Details Formatted Diff Diff
Show Obsolete (3) View All Add attachment proposed patch, testcase, etc.

David Levin

Comment 1 2008-11-26 20:47:52 PST

Created attachment 25548 [details] Removes the non-thread safe static initializations in XMLHttpRequest.cpp

David Levin

Comment 2 2008-11-27 01:00:32 PST

Created attachment 25555 [details] Very minor adjustment to the last patch (-- moves one brace for style purposes).

Alexey Proskuryakov

Comment 3 2008-11-27 07:33:42 PST

Hmm... The patch is executed perfectly, but I think that it may be over-engineered a bit. I don't see how having classes for each function helps - is it better than just protecting initialization parts of these functions with a mutex?

David Levin

Comment 4 2008-11-27 23:13:35 PST

Created attachment 25566 [details] Changed the static initialization to happen in the XHR constructor. I didn't like doing a mutex in the functions due to one of them being called inside of a loop. As discussed, I've tried out a solution that initializes the static variables in the constructor. I looked at a few ways of doing this. The reason I put the variables in a struct and then initialized into a local static was to take advantage of AtomicallyInitializedStatic. If this doesn't seem that great, one alternative I can think of is to create another macro like AtomicallyInitializedStatic (in threading.h) that allowed me to call a function inside of this same mutex.

David Levin

Comment 5 2008-11-28 01:14:29 PST

Created attachment 25568 [details] Another revision. I think it is close.

Alexey Proskuryakov

Comment 6 2008-11-28 01:22:58 PST

Comment on attachment 25568 [details] Another revision. I think it is close. r=me

Alexey Proskuryakov

Comment 7 2008-11-28 02:24:35 PST

Committed revision 38827. I made a slight tweak I didn't think about earlier: made the global const.

Note You need to log in before you can comment on or make changes to this bug.