712990 - Endian issue with YarrJIT

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Cameron Kaiser [:spectre]]

Attachment: Big endian fix for YarrJIT ignoreCase

http://code.google.com/p/tenfourfox/issues/detail?id=120

Although PowerPC JM+TI is currently stillborne -_-, we are making use of our macroassembler for YARR and Ben Stuhl, one of our TenFourFox contributors, discovered an endianness issue with YARR JIT. It's a small patch and I figured we should push it upstream for SPARC and (future) MIPS consumers. Ben's notes:

I built js/ from 10.4Fx 9 and ran 'make check jstests'. It revealed a few failures, including several; in ecma_3/RegExp/perlstress-001.js, which can be reduced to

'A]'.match(/a]/i) == null /* should be 'A]' */

. I get the same result running the test by hand in the Web Console in 10.4Fx 9.

It turns out to be a minor endian bug in YARR; here's a patch.

Comment 1

[David Mandelin [:dmandelin]]

Did you file a bug with WebKit too?

Comment 2

[Cameron Kaiser [:spectre]]

Not yet, I'll do that too. Thanks for the review!

Comment 3

[David Mandelin [:dmandelin]]

http://hg.mozilla.org/integration/mozilla-inbound/rev/334f198c10b5

Comment 4

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/334f198c10b5