Emit comment with instruction+reason before deopt calls.

Emit comment with instruction+reason before deopt calls.

Note that we still need to migrate from sometimes emitting those
comments by hand to passing a reason explicitly, but this can be done
incrementally in separate CLs.

R=jarin@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=24061

[Sven Panne, 2014-09-18 14:19:12]

svenpanne@chromium.org changed reviewers:
+ jarin@chromium.org

[Jarin, 2014-09-18 22:24:52]

LGTM.

How about wrapping "const char* reason" in a class? I can easily imagine that we
might want to impose more structure on the deopt information later; having a
wrapper would make this simpler. (I understand that ultimately we will want to
thread the deopt reason through generated code, so the wrapper might not be
completely trivial.)

[Sven Panne, 2014-09-19 06:26:30]

On 2014/09/18 22:24:52, jarin wrote:
> How about wrapping "const char* reason" in a class? I can easily imagine that
we
> might want to impose more structure on the deopt information later; having a
> wrapper would make this simpler. (I understand that ultimately we will want to
> thread the deopt reason through generated code, so the wrapper might not be
> completely trivial.)

I thought about something similar, too, but I wasn't sure if it's worth it or if
it would complicate things more. If we do it, I think the mnemonic/reason pair
should be in that class, perhaps even a source position?

My proposal would be: Land this CL as it is and introduce the class in a
separate CL (if it works out).

[Sven Panne, 2014-09-19 06:27:23]

Committed patchset #2 (id:20001) manually as 24061 (presubmit successful).