Bug 110039 – Code cache should be explicit about what it caches

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

[Geoffrey Garen]

Code cache should be explicit about what it caches

[Geoffrey Garen]

Created attachment 188741 [details]
Patch

[Geoffrey Garen]

No regression on PLT.

[Oliver Hunt]

Comment on attachment 188741 [details]
Patch

What does this do to memory usage?  It looks fine, but i recall membuster disliking the strong linkage between entries in the code cache and individual function bodies.

[Build Bot]

Comment on attachment 188741 [details]
Patch

Attachment 188741 [details] did not pass mac-wk2-ews (mac-wk2):
Output: http://queues.webkit.org/results/16593803

New failing tests:
media/video-controls-captions-trackmenu.html

[Geoffrey Garen]

> What does this do to memory usage?  It looks fine, but i recall membuster disliking the strong linkage between entries in the code cache and individual function bodies.

Because a low memory warning evicts the cache, I don't think Membuster will be affected / can be used to measure. Instead, I tuned using the PLT: this character count is what you need to get an equal PLT score to the old entry count. Based on that, and since we know that a larger cache makes the PLT faster, I think memory use should be equal.

[Geoffrey Garen]

Committed r143133: <http://trac.webkit.org/changeset/143133>

[Roger Fong]

This broke the windows build: http://build.webkit.org/builders/Apple%20Win%20Release%20%28Build%29/builds/44385/steps/compile-webkit/logs/stdio

The problem line is "typedef type name MapType::iterator iterator".
VS dislikes the use of "typename" anywhere outside a template class declaration. Is using "typename" required here if all that's happening is a typedef of MapType::iterator?

[Roger Fong]

(In reply to comment #7)
> This broke the windows build: http://build.webkit.org/builders/Apple%20Win%20Release%20%28Build%29/builds/44385/steps/compile-webkit/logs/stdio
> 
> The problem line is "typedef type name MapType::iterator iterator".
> VS dislikes the use of "typename" anywhere outside a template class declaration. Is using "typename" required here if all that's happening is a typedef of MapType::iterator?

I landed a build fix here: http://trac.webkit.org/changeset/143141 to get bots happy.
Let me know if I'm wrong and just blew up all of JSC (or just a little part of it).