⚠ Archived content — this site is no longer maintained.   Current WebKit documentation is at docs.webkit.org.

Timeline



May 29, 2017:

10:42 PM Changeset in webkit [217543] by bshafiei@apple.com
  • 7 edits in tags/Safari-604.1.23.0.2/Source

Versioning.

10:37 PM Changeset in webkit [217542] by bshafiei@apple.com
  • 1 copy in tags/Safari-604.1.23.0.2

New tag.

10:03 PM Changeset in webkit [217541] by rniwa@webkit.org
  • 5 edits in trunk/Websites/perf.webkit.org

Fix UI glitches with a custom analysis test group with a patch
​https://bugs.webkit.org/show_bug.cgi?id=172694

Reviewed by Sam Weinig.

Fix the following UI glitches with perf try bots:

  • Retrying an A/B testing with a patch fails.
  • A patch specified in an test group does not get specified in the configurator.
  • Drag & dropping a patch doesn't work.
  • Results for custom analysis tasks don't get shown.
  • public/api/test-groups.php:

(main): Fix a bug that test group's platform does not match that of the request'ed platform. Since each test
group is associated with platform, just use that instead of querying test_configurations. This resulted in
the configurator not being able to find a triggerable in some cases.

  • public/v3/components/custom-analysis-task-configurator.js:

(CustomAnalysisTaskConfigurator):
(CustomAnalysisTaskConfigurator.prototype.setCommitSets): Add patches in the commit set.
(CustomAnalysisTaskConfigurator.prototype._setUploadedFilesToUploader): Now clears the exiting uploaded files
Also renamed from _setUploadedFilesIfEmpty.
(CustomAnalysisTaskConfigurator.prototype._setPatchFiles): Added.
(CustomAnalysisTaskConfigurator.prototype.didConstructShadowTree): We no longer update the list of roots
for the comparsion when a new root is added to the baseline.
(CustomAnalysisTaskConfigurator.prototype._configureComparison): Copy over the list of patches and roots when
starting to configure the comparsion.

  • public/v3/components/instant-file-uploader.js:

(InstantFileUploader.prototype.clear): Added.
(InstantFileUploader.prototype.didConstructShadowTree): Added event handlers for dragover & drop events to
allow specifying a patch and root using drag & drop. Unfortunately, this still doesn't work in WebKit due to
a bug in our shadow DOM implementation.
(InstantFileUploader.prototype._didFileInputChange):
(InstantFileUploader.prototype._uploadFiles): Extracted from _didFileInputChange.

  • public/v3/pages/analysis-task-page.js:

(AnalysisTaskTestGroupPane.prototype.setAnalysisResults): No longer takes metric.
(AnalysisTaskTestGroupPane.cssTemplate): Removed unused rules. Also disallow flexing on the list of test groups
to avoid the name of a test froup from overflowing on top of the results pane.
(AnalysisTaskPage.prototype._assignTestResultsIfPossible): Set setAnalysisResults even when metric is not set
as is the case for a custom analysis task.
(AnalysisTaskPage.prototype._retryCurrentTestGroup): Use createWithCustomConfiguration to allow retrying of
an A/B testing with a patch in a custom analysis task.
(AnalysisTaskPage.prototype._createTestGroupAfterVerifyingCommitSetList):

11:20 AM Changeset in webkit [217540] by Claudio Saavedra
  • 2 edits in trunk/LayoutTests

[WPE] Mark a couple of tests as slow/timing out

Unreviewed gardening. These are all known ones.

  • platform/wpe/TestExpectations:
10:35 AM Changeset in webkit [217539] by Claudio Saavedra
  • 2 edits in trunk/LayoutTests

[WPE] Marking workers/bomb.html test as slow.

Unreviewed gardening.

  • platform/wpe/TestExpectations:
9:17 AM Changeset in webkit [217538] by Claudio Saavedra
  • 2 edits in trunk/Source/WebKit2

[GTK] Remove unneeded forward declaration

Rubber-stamped by Zan Dobersek.

  • UIProcess/API/C/gtk/WKAPICastGtk.h: Remove WebGrammarDetail as

its header is included in the same file.

9:05 AM Changeset in webkit [217537] by zandobersek@gmail.com
  • 2 edits in trunk/Tools

Unreviewed. Bumping the WPEBackend-mesa dependency version.

  • wpe/jhbuild.modules: Use the latest version that uses proper GLib

priorities for IPC communication.

7:09 AM Changeset in webkit [217536] by commit-queue@webkit.org
  • 4 edits in trunk

Use the parent box style to adjust RenderStyle for alignment.
​https://bugs.webkit.org/show_bug.cgi?id=172215

Patch by Emilio Cobos Álvarez <​ecobos@igalia.com> on 2017-05-29
Reviewed by Antti Koivisto.

Source/WebCore:

The css-flexbox spec defined align-self in terms of the parent
element, which is what this code did.

The css-align spec defines these properties in terms of the style of
the containing box instead, which means display: contents styles
should not be used for this adjustment, but the parent box style
instead.

For example, align-items is defined as:

This property specifies the default align-self for all of the boxes
(including anonymous boxes) participating in this box’s formatting
context.

Note that the css-align spec was recently updated to align (no pun
intended) with Gecko, and make the |auto| value compute to itself.
This patch puts us in a more recent spec than before, but not totally
up-to-date.

Tests: imported/w3c/web-platform-tests/css/css-display-3/display-contents-alignment-001.html

imported/w3c/web-platform-tests/css/css-display-3/display-contents-alignment-002.html

  • css/StyleResolver.cpp:

(WebCore::StyleResolver::adjustRenderStyle):
(WebCore::StyleResolver::adjustStyleForAlignment):

LayoutTests:

3:37 AM Changeset in webkit [217535] by Yusuke Suzuki
  • 2 edits in trunk/Source/WTF

Unreviewed, disable faster Interpreter::getOpcodeID for ARM_THUMB2 with non-Darwin OSes
​https://bugs.webkit.org/show_bug.cgi?id=172686

Because of test failures.

  • wtf/Platform.h:
2:03 AM Changeset in webkit [217534] by pvollan@apple.com
  • 2 edits in trunk/LayoutTests

Unreviewed Windows test gardening, update expected results after r217418.

  • platform/win/fast/block/float/032-expected.txt:
1:41 AM Changeset in webkit [217533] by Claudio Saavedra
  • 2 edits in trunk/Source/WebKit2

[WPE] Build fix after r217531

Unreviewed.

  • UIProcess/API/C/wpe/WKAPICastWPE.h:

(WebKit::toAPI): Add missing WebGrammarDetail API cast.

12:15 AM Changeset in webkit [217532] by pvollan@apple.com
  • 7 edits in trunk/LayoutTests

Unreviewed Windows test gardening, update expected results after r217418.

  • platform/win/fast/forms/input-appearance-spinbutton-expected.txt:
  • platform/win/fast/forms/input-appearance-spinbutton-up-expected.txt:
  • platform/win/fast/forms/number/number-appearance-rtl-expected.txt:
  • platform/win/fast/forms/number/number-appearance-spinbutton-disabled-readonly-expected.txt:
  • platform/win/fast/forms/number/number-appearance-spinbutton-layer-expected.txt:
  • platform/win/fast/forms/search-vertical-alignment-expected.txt:

May 28, 2017:

10:30 PM Changeset in webkit [217531] by mitz@apple.com
  • 39 edits in trunk/Source

[Xcode] ALWAYS_SEARCH_USER_PATHS is set to YES
​https://bugs.webkit.org/show_bug.cgi?id=172691

Reviewed by Tim Horton.

  • Configurations/Base.xcconfig: Set ALWAYS_SEARCH_USER_PATHS to NO.

Source/JavaScriptCore:

Source/WebCore:

  • WebCore.xcodeproj/project.pbxproj: Added DateTimeChooser.h, DateTimeChooserClient.h, PerformanceMark.h, PerformanceMeasure.h, SVGUnknownElement.h, and MathMLUnknownElement.h to the WebCore target.

Source/WebKit2:

  • UIProcess/API/C/WKAPICast.h: Moved GTK-only definitions that used WebGrammarDetail.h to WKAPICastGtk.h. This had the effect of no longer including APIArray.h from this header.
  • UIProcess/API/C/gtk/WKAPICastGtk.h: Moved GTK-only definitions to here.

(WebKit::toAPI):

  • UIProcess/WebGrammarDetail.h: Replaced include of APIArray.h with a forward declaration.
  • Shared/API/c/WKRenderLayer.cpp:
  • Shared/API/c/WKRenderObject.cpp:
  • UIProcess/API/C/WKApplicationCacheManager.cpp:
  • UIProcess/API/C/WKContext.cpp:
  • UIProcess/API/C/WKContextConfigurationRef.cpp:
  • UIProcess/API/C/WKCookieManager.cpp:
  • UIProcess/API/C/WKKeyValueStorageManager.cpp:
  • UIProcess/API/C/WKNotificationManager.cpp:
  • UIProcess/API/C/WKOpenPanelResultListener.cpp:
  • UIProcess/API/C/WKPageGroup.cpp:
  • UIProcess/API/C/WKResourceCacheManager.cpp:
  • UIProcess/API/C/WKUserContentControllerRef.cpp:
  • UIProcess/API/gtk/WebKitBackForwardList.cpp:
  • UIProcess/WebContextMenuListenerProxy.cpp:

Added #include "APIArray.h" to these files now that WKAPICast.h does not include it.

10:09 PM Changeset in webkit [217530] by Yusuke Suzuki
  • 2 edits in trunk/Source/JavaScriptCore

[JSC] Provide better type information of toLength and tighten bytecode
​https://bugs.webkit.org/show_bug.cgi?id=172690

Reviewed by Sam Weinig.

In this patch, we carefully leverage operator + in order to

  1. tighten bytecode

operator+ emits to_number bytecode. What this bytecode does is the same
to @Number() call. It is more efficient, and it is smaller bytecode
than @Number() call (load global variable @Number, set up arguments, and
call it).

  1. offer better type prediction data

Now, we have code like

length > 0 ? (length < @MAX_SAFE_INTEGER ? length : @MAX_SAFE_INTEGER) : 0

This is not good because DFG prediction propagation phase predicts as Double
since @MAX_SAFE_INTEGER is double. But actually it rarely becomes Double.
Usually, the result becomes Int32. This patch leverages to_number in a bit
interesting way: to_number has value profiling to offer better type prediction.
This value profiling can offer a chance to change the prediction to Int32 efficiently.
It is a bit tricky. But it is worth doing to speed up our builtin functions,
which should leverage all the JSC's tricky things to be optimized.

Related microbenchmarks show performance improvement.

baseline patched

array-prototype-forEach 50.2348+-2.2331 49.7568+-2.3507
array-prototype-map 51.0574+-1.8166 47.9531+-2.1653 might be 1.0647x faster
array-prototype-some 52.3926+-1.8882 48.3632+-2.0852 definitely 1.0833x faster
array-prototype-every 52.7394+-2.0712 50.2896+-2.1480 might be 1.0487x faster
array-prototype-reduce 54.9994+-2.3638 51.8716+-2.6253 might be 1.0603x faster
array-prototype-reduceRight 209.7594+-9.2594 51.5867+-2.5745 definitely 4.0662x faster

  • builtins/GlobalOperations.js:

(globalPrivate.toInteger):
(globalPrivate.toLength):

6:30 PM Changeset in webkit [217529] by commit-queue@webkit.org
  • 7 edits
    2 adds in trunk

[WebIDL] @@iterator should only be accessed once when disambiguating a union type
​https://bugs.webkit.org/show_bug.cgi?id=172684

Patch by Sam Weinig <​sam@webkit.org> on 2017-05-28
Reviewed by Yusuke Suzuki.

Source/JavaScriptCore:

  • runtime/IteratorOperations.cpp:

(JSC::iteratorMethod):
(JSC::iteratorForIterable):

  • runtime/IteratorOperations.h:

(JSC::forEachInIterable):
Add additional iterator helpers to allow union + sequence conversion code
to check for iterability by getting the iterator method, and iterate using
that method later on.

Source/WebCore:

WebIDL specifies that when determining if the value you are converting to a union
is a sequence, you must get the @@iterator property and, should it exist, use it
to iterate the sequence. While we correctly accessing the property to make the
determination, we were not passing it into the sequence conversion code, and thus
the sequence conversion code re-accessed it, which is observable and wrong.

This patch pipes the @@iterator method through the sequence conversion code to avoid
this.

Test: js/dom/sequence-in-union-iterator-access.html

  • bindings/js/JSDOMConvertSequences.h:

(WebCore::Detail::GenericSequenceConverter::convert):
(WebCore::Detail::NumericSequenceConverter::convertArray):
(WebCore::Detail::NumericSequenceConverter::convert):
(WebCore::Detail::SequenceConverter::convertArray):
(WebCore::Detail::SequenceConverter::convert):
(WebCore::Detail::SequenceConverter<IDLLong>::convert):
(WebCore::Detail::SequenceConverter<IDLFloat>::convert):
(WebCore::Detail::SequenceConverter<IDLUnrestrictedFloat>::convert):
(WebCore::Detail::SequenceConverter<IDLDouble>::convert):
(WebCore::Detail::SequenceConverter<IDLUnrestrictedDouble>::convert):
(WebCore::Converter<IDLSequence<T>>::convert):
(WebCore::Converter<IDLFrozenArray<T>>::convert):
Add variants of convert that take a JSObject* (sequence) / JSValue (iterator method)
rather than just the JSValue (sequence). To avoid too much duplication, split some
parts of SequenceConverter and NumericSequenceConverter up so they could be reused.

  • bindings/js/JSDOMConvertUnion.h:
  • Fix incorrect step 3 (WebIDL got updated at some point and we didn't notice) to remove records.
  • Update sequence and FrozenArray checking/conversion to get the iterator method and pass it along, using the new ConditionalSequenceConverter helper which forwards to the new sequence converters that accept the iterator method.

LayoutTests:

  • js/dom/sequence-in-union-iterator-access-expected.txt: Added.
  • js/dom/sequence-in-union-iterator-access.html: Added.

Add test case showing that @@iterator is only accessed once when converting a sequence
as part of a union.

7:11 AM Changeset in webkit [217528] by Yusuke Suzuki
  • 2 edits in trunk/Source/WTF

[JSC][Linux][FreeBSD] Use faster Interpreter::getOpcodeID()
​https://bugs.webkit.org/show_bug.cgi?id=172686

Reviewed by Mark Lam.

As of r217526, JSC gets faster Interpreter::getOpcodeID() by
embedding OpcodeID value just before the LLInt machine code
handler pointer. By doing so, we can retrieve OpcodeID from
the LLInt machine code handler by dereferencing the code
pointer. *((int*)ptr - 1).

This patch allows Linux and FreeBSD environments to use this
optimization.

  • wtf/Platform.h:
4:33 AM Changeset in webkit [217527] by Yusuke Suzuki
  • 5 edits in trunk/Source/JavaScriptCore

Unreviewed, build fix for Windows
​https://bugs.webkit.org/show_bug.cgi?id=172413

Optimized jsDynamicCast for JSMap and JSSet will be handled in [1].

[1]: ​https://bugs.webkit.org/show_bug.cgi?id=172685

  • runtime/JSMap.h:

(JSC::isJSMap):
(JSC::jsDynamicCast): Deleted.
(JSC::>): Deleted.

  • runtime/JSSet.h:

(JSC::isJSSet):
(JSC::jsDynamicCast): Deleted.
(JSC::>): Deleted.

  • runtime/MapConstructor.cpp:

(JSC::constructMap):

  • runtime/SetConstructor.cpp:

(JSC::constructSet):

1:12 AM Changeset in webkit [217526] by mark.lam@apple.com
  • 9 edits in trunk/Source

Implement a faster Interpreter::getOpcodeID().
​https://bugs.webkit.org/show_bug.cgi?id=172669

Reviewed by Saam Barati.

Source/JavaScriptCore:

We can implement Interpreter::getOpcodeID() without a hash table lookup by always
embedding the OpcodeID in the 32-bit word just before the start of the LLInt
handler code that executes each opcode. getOpcodeID() can therefore just read
the 32-bits before the opcode address to get its OpcodeID.

This is currently only enabled for CPU(X86), CPU(X86_64), CPU(ARM64),
CPU(ARM_THUMB2), and only for OS(DARWIN). It'll probably just work for linux as
well, but I'll let the Linux folks turn that on after they have verified that it
works on linux too.

I'll also take this opportunity to clean up how we initialize the opcodeIDTable:

  1. we only need to initialize it once per process, not once per VM / interpreter instance.
  2. we can initialize it in the Interpreter constructor instead of requiring a separate call to an initialize() function.

On debug builds, the Interpreter constructor will also verify that getOpcodeID()
is working correctly for each opcode when USE(LLINT_EMBEDDED_OPCODE_ID).

  • bytecode/BytecodeList.json:
  • generate-bytecode-files:
  • interpreter/Interpreter.cpp:

(JSC::Interpreter::Interpreter):
(JSC::Interpreter::opcodeIDTable):
(JSC::Interpreter::initialize): Deleted.

  • interpreter/Interpreter.h:

(JSC::Interpreter::getOpcode):
(JSC::Interpreter::getOpcodeID):

  • llint/LowLevelInterpreter.cpp:
  • runtime/VM.cpp:

(JSC::VM::VM):

Source/WTF:

Added the USE(LLINT_EMBEDDED_OPCODE_ID) configuration.

  • wtf/Platform.h:
Note: See TracTimeline for information about the timeline view.