Bug 142826

Summary:	Introduce internal.querySelector[All]NoJIT
Product:	WebKit	Reporter:	Yusuke Suzuki <ysuzuki>
Component:	CSS	Assignee:	Nobody <webkit-unassigned>
Status:	NEW ---
Severity:	Normal	CC:	benjamin
Priority:	P2
Version:	528+ (Nightly build)
Hardware:	Unspecified
OS:	Unspecified
Bug Depends on:	142703
Bug Blocks:

Description Yusuke Suzuki 2015-03-18 08:34:17 PDT

Now, almost all selectors (even if it's very complicated) are compiled by CSS JIT.
To maintain non-JIT SelectorChecker and compare CSS JIT results with expected non-JIT results,
I propose new internal.API, internal.querySelectorNoJIT(root, selector) and internal.querySelectorAllNoJIT(root, selector).

Comment 1 Yusuke Suzuki 2015-03-18 08:34:36 PDT

Benjamin, what do you think of?

Comment 2 Yusuke Suzuki 2015-03-18 08:45:53 PDT

When landing this change, I'll add non-JIT selector matching tests introduced issue 142703.

Comment 3 Benjamin Poulain 2015-03-18 11:04:51 PDT

(In reply to comment #2)
> When landing this change, I'll add non-JIT selector matching tests
> introduced issue 142703.

I agree the non-JIT cases are under-tested, that has been worrying me for a while.

I have been thinking about some solutions:
-Add a new pseudo class :-webkit-no-jit that fails to compile. The problem with this is someone could ship it by accident.
-Change the test running such that the layout tests in fast/selectors are run twice: once with the normal setup, then a second time with all JIT disabled (which we can do at runtime since we rely on the same flags as JavaScriptCore). I think this would have the most value.

I am okay with internal.querySelectorNoJIT(root, selector) and internal.querySelectorAllNoJIT(root, selector). That seems useful.