Bug 174646 – Implement Unicode RegExp support in the YARR JIT

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Michael Saboff]

Currently unicode regular expressions are handled in the Yarr interpreter.  We should add unicode support to the Yarr JIT.

<rdar://problem/33183180>

[Michael Saboff]

Created attachment 315867 [details]
Work in Progress

This passes tests on both Mac and iOS 64 bit, but there is some more tuning that can be done.

[Michael Saboff]

Created attachment 315930 [details]
Updated Work in Progress with build fixes for 32 bit platforms

[Build Bot]

Comment on attachment 315930 [details]
Updated Work in Progress with build fixes for 32 bit platforms

Attachment 315930 [details] did not pass ios-sim-ews (ios-simulator-wk2):
Output: http://webkit-queues.webkit.org/results/4158569

New failing tests:
js/regexp-unicode.html
imported/w3c/IndexedDB-private-browsing/idbfactory_open.html

[Build Bot]

Created attachment 316059 [details]
Archive of layout-test-results from ews125 for ios-simulator-wk2

The attached test failures were seen while running run-webkit-tests on the ios-sim-ews.
Bot: ews125  Port: ios-simulator-wk2  Platform: Mac OS X 10.12.5

[Michael Saboff]

Created attachment 318771 [details]
Patch

[Michael Saboff]

Created attachment 318789 [details]
Updated Patch

Added check for quantifier overflow of character terms.

[Filip Pizlo]

Comment on attachment 318789 [details]
Updated Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=318789&action=review

LGTM

> Source/JavaScriptCore/ChangeLog:38
> +        function, getEffectiveAddress64(), for ARM64.  It just calls x86Lea64() on X86-64.

Nice.

[JF Bastien]

Comment on attachment 318789 [details]
Updated Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=318789&action=review

lgtm, not an expert in this though....

> Source/JavaScriptCore/yarr/YarrInterpreter.cpp:2067
>  COMPILE_ASSERT(sizeof(Interpreter<UChar>::BackTrackInfoParentheses) == (YarrStackSpaceForBackTrackInfoParentheses * sizeof(uintptr_t)), CheckYarrStackSpaceForBackTrackInfoParentheses);

Move to static_assert?

> Source/JavaScriptCore/yarr/YarrJIT.cpp:2925
> +        , m_canonicalMode(m_pattern.unicode() ? CanonicalMode::Unicode : CanonicalMode::UCS2)

This is a bit weird because Unicode normalization has a concept of "canonical" which doesn't match this: http://unicode.org/faq/normalization.html

[Michael Saboff]

(In reply to JF Bastien from comment #8)
> Comment on attachment 318789 [details]
> Updated Patch
> 
> View in context:
> https://bugs.webkit.org/attachment.cgi?id=318789&action=review
> 
> lgtm, not an expert in this though....
> 
> > Source/JavaScriptCore/yarr/YarrInterpreter.cpp:2067
> >  COMPILE_ASSERT(sizeof(Interpreter<UChar>::BackTrackInfoParentheses) == (YarrStackSpaceForBackTrackInfoParentheses * sizeof(uintptr_t)), CheckYarrStackSpaceForBackTrackInfoParentheses);
> 
> Move to static_assert?

COMPILE_ASSERT resolves to static_assert on appropriate platforms.

> > Source/JavaScriptCore/yarr/YarrJIT.cpp:2925
> > +        , m_canonicalMode(m_pattern.unicode() ? CanonicalMode::Unicode : CanonicalMode::UCS2)
> 
> This is a bit weird because Unicode normalization has a concept of
> "canonical" which doesn't match this:
> http://unicode.org/faq/normalization.html

This is part of our implementation Canonicalize() for case folding as specified in the standard at https://tc39.github.io/ecma262/#sec-runtime-semantics-canonicalize-ch.  We use CanonicalMode::Unicode to signify what that section has as "Unicode is true".  UCS2 is legacy canonicalization.

[Michael Saboff]

Committed r221052: <http://trac.webkit.org/changeset/221052>