Include getters and setters callbacks invocations in CPU profiler log.

Include getters and setters callbacks invocations in CPU profiler log.

Logging getters and setters from DOM API is extremely useful for web
developers as setting (and getting!) several properties can cause
page relayouts which take significant time.

Committed: http://code.google.com/p/v8/source/detail?r=3363

[Søren Thygesen Gjesse, 2009-11-25 10:09:35]

We should find a way to avoid calls to logger for each getter/setter callback.
Don't you have the same problem of using the same callback function for several
functions as well? What about interceptors?

http://codereview.chromium.org/434074/diff/1/3
File src/log.h (right):

http://codereview.chromium.org/434074/diff/1/3#newcode211
src/log.h:211: static void GetCallbackEvent(String* name, Address entry_point);
How about GetCallbackEvent -> GetterCallbackEvent?

http://codereview.chromium.org/434074/diff/1/3#newcode212
src/log.h:212: static void SetCallbackEvent(String* name, Address entry_point);
and SetCallbackEvent -> SetterCallbackEvent?

http://codereview.chromium.org/434074/diff/1/4
File src/stub-cache.cc (right):

http://codereview.chromium.org/434074/diff/1/4#newcode755
src/stub-cache.cc:755: LOG(GetCallbackEvent(*name, getter_address));
Even though this is caught by the log compression I don't like having a call to
the logger in this code-path - it is a quite hot piece of code in some
situations.

Would it be possible to add the name the VMState instead? It might require
additional handling of VMState chain during GC, but it has a much less overhead
- and predictable overhaea as well. However as the name is in a handle on the
stack here sticking the raw pointer into the VMState should be safe...

[mnaganov (inactive), 2009-11-25 15:11:17]

http://codereview.chromium.org/434074/diff/1/3
File src/log.h (right):

http://codereview.chromium.org/434074/diff/1/3#newcode211
src/log.h:211: static void GetCallbackEvent(String* name, Address entry_point);
On 2009/11/25 10:09:35, Søren Gjesse wrote:
> How about GetCallbackEvent -> GetterCallbackEvent?

Done.

http://codereview.chromium.org/434074/diff/1/3#newcode212
src/log.h:212: static void SetCallbackEvent(String* name, Address entry_point);
On 2009/11/25 10:09:35, Søren Gjesse wrote:
> and SetCallbackEvent -> SetterCallbackEvent?

Done.

http://codereview.chromium.org/434074/diff/1/4
File src/stub-cache.cc (right):

http://codereview.chromium.org/434074/diff/1/4#newcode755
src/stub-cache.cc:755: LOG(GetCallbackEvent(*name, getter_address));
On 2009/11/25 10:09:35, Søren Gjesse wrote:
> Even though this is caught by the log compression I don't like having a call
to
> the logger in this code-path - it is a quite hot piece of code in some
> situations.
> 
> Would it be possible to add the name the VMState instead? It might require
> additional handling of VMState chain during GC, but it has a much less
overhead
> - and predictable overhaea as well. However as the name is in a handle on the
> stack here sticking the raw pointer into the VMState should be safe...

OK. Let's assume that all getters and setters are used only for one property (so
name argument is unused by them), which is true in case of Chromium. This
eliminates additional complexity. The only extra step is to log getter / setter
entry points by traversing AccessorInfo objects.

[Søren Thygesen Gjesse, 2009-11-25 16:09:16]

LGTM