Last week, evilpie and Ms2ger stepped up to the plate and helped out
with rooting in a big way. With their help, we are now at about 9
remaining jit-test failures (sans --tbpl), down from 700-800 before they
started. Big thanks go out to both of them.
The remaining jit test failures fall into the following causes.
RegExpStatics::pendingInput needs rooting:
basic/testBug731181.js
InlineMap needs rooting:
basic/testComparisons.js
v8-v5/check-earley-boyer.js
Test assertion failure: "got 'mark' expected 'none'":
gc/incremental-state.js
JSObject::preventExtensions needs to be static:
basic/testOverRecursed4.js
jaeger/recompile/bug671943-1.js
ParallelArrayObject::toStringBuffer |elem| is not rooted:
parallelarray/constructor-4.js
ArrayToIdVector trapName needs to be a Handle:
proxy/testDirectProxyGetOwnPropertyNames4.js
proxy/testDirectProxyKeys4.js
I'll file bugs for all of these shortly. Feel free to grab any of them,
just be sure to assign yourself before you start working on it.
Oddly, many of these are only failing on Clang and not GCC, although the
tests that fail on GCC always also fail on Clang. GCC also takes almost
a minute less time to run the rooting analysis: we should probably find
out what is going on here.
Steve has one more issue to fix with Try before we can turn on the root
analysis on inbound. He has a fix for that out for review. Combined with
our success with jit-tests last week, we are quite close. Our primary
goal for this week is to get the analysis turned on for mozilla-inbound.
Subsequently and concurrently we need to:
1) go over the remaining SkipRoots with a fine-toothed comb and find
rooting solutions for any of the remaining SkipRoots that are hiding
actual rooting hazards. We know of at least one in JaegerMonkey that's
going to be trouble.
2) start fixing fuzz bugs and asking for more fuzzer attention.
3) experiment with suppressing GC during IonMonkey compilation.
4) finish the static analysis that Brian is working on with the intent
of making it scalable enough, stable enough, and user-friendly enough to
fully replace our dynamic GCability assertions.
Back in September, Dave and I estimated about 12 weeks to getting the
analysis switched on: it looks like we were right on target.
-Terrence
_______________________________________________
dev-tech-js-engine-internals mailing list
dev-tech-js-engine-internals@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-tech-js-engine-internals
