RESOLVED FIXED 166626
DeferGC::~DeferGC should be super cheap
https://bugs.webkit.org/show_bug.cgi?id=166626
Summary DeferGC::~DeferGC should be super cheap
Filip Pizlo
Reported 2016-12-30 17:11:06 PST
Right now, it requires running the collector's full collectIfNecessaryOrDefer() hook, which is super big. The right thing for it to do is to check a boolean to see if collectIfNecessaryOrDefer() had ever deferred.
Attachments
the patch (11.66 KB, patch)
2016-12-30 17:18 PST, Filip Pizlo
no flags
the patch (12.41 KB, patch)
2017-01-02 13:29 PST, Filip Pizlo
ggaren: review-
buildbot: commit-queue-
Archive of layout-test-results from ews112 for mac-elcapitan (564.90 KB, application/zip)
2017-01-02 14:33 PST, Build Bot
no flags
the patch (13.54 KB, patch)
2017-01-05 16:18 PST, Filip Pizlo
buildbot: commit-queue-
Archive of layout-test-results from ews116 for mac-elcapitan (757.77 KB, application/zip)
2017-01-05 17:24 PST, Build Bot
no flags
the patch (14.51 KB, patch)
2017-01-05 19:15 PST, Filip Pizlo
no flags
the patch (14.82 KB, patch)
2017-01-05 19:57 PST, Filip Pizlo
saam: review+
Filip Pizlo
Comment 1 2016-12-30 17:18:57 PST
Created attachment 297864 [details] the patch
Filip Pizlo
Comment 2 2017-01-02 13:29:23 PST
Created attachment 297908 [details] the patch
Build Bot
Comment 3 2017-01-02 14:32:58 PST
Comment on attachment 297908 [details] the patch Attachment 297908 [details] did not pass mac-debug-ews (mac): Output: http://webkit-queues.webkit.org/results/2819823 Number of test failures exceeded the failure limit.
Build Bot
Comment 4 2017-01-02 14:33:01 PST
Created attachment 297912 [details] Archive of layout-test-results from ews112 for mac-elcapitan The attached test failures were seen while running run-webkit-tests on the mac-debug-ews. Bot: ews112 Port: mac-elcapitan Platform: Mac OS X 10.11.6
Geoffrey Garen
Comment 5 2017-01-03 11:01:22 PST
ASSERTION FAILED: m_didDeferGCWork == !!m_deferralDepth /Volumes/Data/EWS/WebKit/Source/JavaScriptCore/heap/HeapInlines.h(332) : void JSC::Heap::decrementDeferralDepthAndGCIfNeeded() 1 0x101de0d70 WTFCrash 2 0x1009e031d JSC::Heap::decrementDeferralDepthAndGCIfNeeded() 3 0x1009e0278 JSC::DeferGC::~DeferGC() 4 0x1009e0205 JSC::DeferGC::~DeferGC() 5 0x1009e01b1 JSC::GCSafeConcurrentJSLocker::~GCSafeConcurrentJSLocker()
Filip Pizlo
Comment 6 2017-01-05 16:18:27 PST
Created attachment 298150 [details] the patch
Filip Pizlo
Comment 7 2017-01-05 16:35:52 PST
Comment on attachment 298150 [details] the patch Still crashes on my machine.
Build Bot
Comment 8 2017-01-05 17:24:42 PST
Comment on attachment 298150 [details] the patch Attachment 298150 [details] did not pass mac-debug-ews (mac): Output: http://webkit-queues.webkit.org/results/2840818 Number of test failures exceeded the failure limit.
Build Bot
Comment 9 2017-01-05 17:24:45 PST
Created attachment 298157 [details] Archive of layout-test-results from ews116 for mac-elcapitan The attached test failures were seen while running run-webkit-tests on the mac-debug-ews. Bot: ews116 Port: mac-elcapitan Platform: Mac OS X 10.11.6
Filip Pizlo
Comment 10 2017-01-05 17:36:07 PST
It turns out that this assertion was just wrong! I've replaced it with a comment explaining how all of the unexpected cases arise and why they are benign.
Filip Pizlo
Comment 11 2017-01-05 19:15:04 PST
Created attachment 298165 [details] the patch
Filip Pizlo
Comment 12 2017-01-05 19:57:56 PST
Created attachment 298167 [details] the patch
Saam Barati
Comment 13 2017-01-06 11:54:11 PST
Comment on attachment 298167 [details] the patch View in context: https://bugs.webkit.org/attachment.cgi?id=298167&action=review r=me > Source/JavaScriptCore/heap/Heap.cpp:2092 > +void Heap::decrementDeferralDepthAndGCIfNeededSlow() Name nit: This doesn't actually decrement the deferral depth, that's done by the caller. Maybe change the name accordingly? > Source/JavaScriptCore/heap/HeapInlines.h:345 > + // Only m_didDeferGCWork is true: Impossible. We stopped for GC and the GC did DeferGC. If this is impossible why can't we assert this?
Filip Pizlo
Comment 14 2017-01-06 11:58:02 PST
(In reply to comment #13) > Comment on attachment 298167 [details] > the patch > > View in context: > https://bugs.webkit.org/attachment.cgi?id=298167&action=review > > r=me > > > Source/JavaScriptCore/heap/Heap.cpp:2092 > > +void Heap::decrementDeferralDepthAndGCIfNeededSlow() > > Name nit: This doesn't actually decrement the deferral depth, that's done by > the caller. Maybe change the name accordingly? It's the slow path of decrementDeferralDepthAndGCIfNeeded(), so I think the name is right. > > > Source/JavaScriptCore/heap/HeapInlines.h:345 > > + // Only m_didDeferGCWork is true: Impossible. We stopped for GC and the GC did DeferGC. > > If this is impossible why can't we assert this? The word-sentence "Impossible." should not be there. I removed it.
Filip Pizlo
Comment 15 2017-01-06 12:24:42 PST
Radar WebKit Bug Importer
Comment 16 2017-01-06 15:26:31 PST
Note You need to log in before you can comment on or make changes to this bug.