Implemented function that returns code age.

Implemented function that returns code age.

This may be used to display debugging information.

BUG=
R=danno@chromium.org, mstarzinger@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=15083

[danno, 2013-06-11 13:39:58]

I think this is a great idea, but worry that since there are no callers of
GetAge and no tests for it, so it runs the risk of bit-rotting. Can you at least
add this to the debug output of the Code object and/or anywhere else that makes
sense?

[Michael Starzinger, 2013-06-11 17:30:07]

https://codereview.chromium.org/16509005/diff/6001/src/mark-compact.cc
File src/mark-compact.cc (right):

https://codereview.chromium.org/16509005/diff/6001/src/mark-compact.cc#newcod...
src/mark-compact.cc:1008: PrintF("[code-flushing clears: %s - age: %d]\n",
*name, age);
Won't this age be constant since we at a certain age and the code-flusher should
be active all the time in normal execution (i.e. without attached debugger)
anyways?

[Hannes Payer (out of office), 2013-06-12 08:56:32]

https://codereview.chromium.org/16509005/diff/6001/src/mark-compact.cc
File src/mark-compact.cc (right):

https://codereview.chromium.org/16509005/diff/6001/src/mark-compact.cc#newcod...
src/mark-compact.cc:1008: PrintF("[code-flushing clears: %s - age: %d]\n",
*name, age);
On 2013/06/11 17:30:07, Michael Starzinger wrote:
> Won't this age be constant since we at a certain age and the code-flusher
should
> be active all the time in normal execution (i.e. without attached debugger)
> anyways?

True, an assert is more appropriate.

[danno, 2013-06-12 09:08:13]

lgtm if you address comments

https://codereview.chromium.org/16509005/diff/11001/src/mark-compact.cc
File src/mark-compact.cc (right):

https://codereview.chromium.org/16509005/diff/11001/src/mark-compact.cc#newco...
src/mark-compact.cc:1047: ASSERT(!FLAG_age_code || code->GetAge() >=
Code::kSexagenarianCodeAge);
Perhaps add a constant for Code::kOldCodeAge that is a copy of
Code::kSexagenarianCodeAge so that all usages are consistent and named?

[Michael Starzinger, 2013-06-12 09:09:32]

https://codereview.chromium.org/16509005/diff/11001/src/mark-compact.cc
File src/mark-compact.cc (right):

https://codereview.chromium.org/16509005/diff/11001/src/mark-compact.cc#newco...
src/mark-compact.cc:1005: ASSERT(!FLAG_age_code || code->GetAge() >=
Code::kSexagenarianCodeAge);
I don't think this assert is entirely correct, if there are two closures
enqueued as flushing candidates then the first one will already have cleared the
SharedFunctionInfo, the second one already has cleared code. Sure, the
LazyRecompileStub should always be marked, but still, this assert seems mighty
weird and confuses more than it helps. Asserts should clarify the code an make
it safer, not make it harder to understand.

How about we move this assert outside the if, and rephrase is slightly to a more
general form ...

ASSERT(!FLAG_arg_code || !candidate->is_compiled() || code->GetAge() >= 3);

https://codereview.chromium.org/16509005/diff/11001/src/mark-compact.cc#newco...
src/mark-compact.cc:1047: ASSERT(!FLAG_age_code || code->GetAge() >=
Code::kSexagenarianCodeAge);
Likewise.

[Michael Starzinger, 2013-06-12 09:12:17]

https://codereview.chromium.org/16509005/diff/11001/src/mark-compact.cc
File src/mark-compact.cc (right):

https://codereview.chromium.org/16509005/diff/11001/src/mark-compact.cc#newco...
src/mark-compact.cc:1005: ASSERT(!FLAG_age_code || code->GetAge() >=
Code::kSexagenarianCodeAge);
On 2013/06/12 09:09:32, Michael Starzinger wrote:
> I don't think this assert is entirely correct, if there are two closures
> enqueued as flushing candidates then the first one will already have cleared
the
> SharedFunctionInfo, the second one already has cleared code. Sure, the
> LazyRecompileStub should always be marked, but still, this assert seems mighty
> weird and confuses more than it helps. Asserts should clarify the code an make
> it safer, not make it harder to understand.
> 
> How about we move this assert outside the if, and rephrase is slightly to a
more
> general form ...
> 
> ASSERT(!FLAG_arg_code || !candidate->is_compiled() || code->GetAge() >= 3);

Sorry, should have been ...

ASSERT(!FLAG_age_code || !candidate->is_compiled() ||
candidate->code()->GetAge() >= 3);

[Hannes Payer (out of office), 2013-06-12 10:59:11]

Since the Assert does not hold, I added the GetAge() call back to the tracing
output. The printed aged is going to be 0 or old age.

[Michael Starzinger, 2013-06-12 11:04:04]

LGTM from my end, with one nit.

https://codereview.chromium.org/16509005/diff/15002/src/mark-compact.cc
File src/mark-compact.cc (right):

https://codereview.chromium.org/16509005/diff/15002/src/mark-compact.cc#newco...
src/mark-compact.cc:1007: PrintF("[code-flushing clears: %s - age: %d]\n",
*name,
nit: Should fit into one line.

[Hannes Payer (out of office), 2013-06-12 12:39:21]

Committed patchset #10 manually as r15083 (presubmit successful).

[Hannes Payer (out of office), 2013-06-12 12:46:16]

https://codereview.chromium.org/16509005/diff/15002/src/mark-compact.cc
File src/mark-compact.cc (right):

https://codereview.chromium.org/16509005/diff/15002/src/mark-compact.cc#newco...
src/mark-compact.cc:1007: PrintF("[code-flushing clears: %s - age: %d]\n",
*name,
On 2013/06/12 11:04:04, Michael Starzinger wrote:
> nit: Should fit into one line.

Done.