Flush number string cache on GC (bug 1605). Also start with a small

Flush number string cache on GC (bug 1605).  Also start with a small
number string cache and only grow it if needed, which will be useful
for saving boot time memory use.
Committed: https://code.google.com/p/v8/source/detail?r=10502

[Michael Starzinger, 2012-01-25 13:00:54]

LGTM. A few comments to address, but mainly nits.

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.cc
File src/heap.cc (right):

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.cc#newcode899
src/heap.cc:899: FlushNumberStringCache();
If it's intentional that the new heuristic for flushing is "always flush", then
I am fine with that. Please document that in the issue tracker.

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.cc#newcode2630
src/heap.cc:2630: // max_semispace_size_ == 512 KB => number_string_cache_size =
32.
This comment no longer is on sync with the computation below.

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.cc#newcode2633
src/heap.cc:2633: number_string_cache_size = Max(kInitialNumberStringCacheSize *
2,
The constant is multiplied with 2 twice. I think we can drop this
multiplication.

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.cc#newcode2647
src/heap.cc:2647: if (maybe_obj->ToObject(&new_cache)) {
Can we add a comment that this is a best effort allocation and we ignore
failures here?

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.h
File src/heap.h (right):

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.h#newcode1804
src/heap.h:1804: // Initializes the number to string cache based on the max
semispace size.
This comment line should either be moved up two lines or be dropped completely.

[Erik Corry, 2012-01-25 14:37:26]

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.cc
File src/heap.cc (right):

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.cc#newcode899
src/heap.cc:899: FlushNumberStringCache();
On 2012/01/25 13:00:56, Michael Starzinger wrote:
> If it's intentional that the new heuristic for flushing is "always flush",
then
> I am fine with that. Please document that in the issue tracker.

Done.

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.cc#newcode2630
src/heap.cc:2630: // max_semispace_size_ == 512 KB => number_string_cache_size =
32.
On 2012/01/25 13:00:56, Michael Starzinger wrote:
> This comment no longer is on sync with the computation below.

Done.

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.cc#newcode2633
src/heap.cc:2633: number_string_cache_size = Max(kInitialNumberStringCacheSize *
2,
On 2012/01/25 13:00:56, Michael Starzinger wrote:
> The constant is multiplied with 2 twice. I think we can drop this
> multiplication.

The first multiplication is there to ensure that the full size is always bigger
than the initial size.  The second is there because there is a string and a
number per entry so the length has to be twice the size (in entries).  Comments
have been updated.

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.cc#newcode2647
src/heap.cc:2647: if (maybe_obj->ToObject(&new_cache)) {
On 2012/01/25 13:00:56, Michael Starzinger wrote:
> Can we add a comment that this is a best effort allocation and we ignore
> failures here?

Done.

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.h
File src/heap.h (right):

https://chromiumcodereview.appspot.com/9235029/diff/1/src/heap.h#newcode1804
src/heap.h:1804: // Initializes the number to string cache based on the max
semispace size.
On 2012/01/25 13:00:56, Michael Starzinger wrote:
> This comment line should either be moved up two lines or be dropped
completely.

Done.