Fix one off error.

Fix one off error.

Proper condition to start eviction is when next possible index is equal
to cache length.

Committed: http://code.google.com/p/v8/source/detail?r=4460

[antonm, 2010-04-20 19:37:51]

Lasse,

may you have a look?

If it's fine, I'd bring it to trunk.

[Lasse Reichstein, 2010-04-21 07:26:35]

LGTM.
Am I correct that the bug means that you write one entry (two elements) past the
end of the FixedArray?

http://codereview.chromium.org/1709001/diff/1/2
File src/runtime.cc (right):

http://codereview.chromium.org/1709001/diff/1/2#newcode10094
src/runtime.cc:10094: if (size < cache->length()) {
Is there something we could pre-fill the cache with to remove this case (and
also remove the "size" handling entirely)? It would need to have a key value
that is never an input to the function (e.g., a newly created object that isn't
used anywhere else).

http://codereview.chromium.org/1709001/diff/1/2#newcode10098
src/runtime.cc:10098: int target_index = finger_index + 2;
This constant, 2, should have a name.
Perhaps add  JSFunctionResultCache::kEntrySize = 2  and use it here.

[antonm, 2010-04-21 11:02:12]

Lasse, thanks lot for review!

I'm submitting it with one change in 5 mins (to run the tests) as I want it on
trunk soon.

And you are quite right: my mistake made the code overwrite values past the
fixed array, so poor maps :(

http://codereview.chromium.org/1709001/diff/1/2
File src/runtime.cc (right):

http://codereview.chromium.org/1709001/diff/1/2#newcode10094
src/runtime.cc:10094: if (size < cache->length()) {
On 2010/04/21 07:26:35, Lasse Reichstein wrote:
> Is there something we could pre-fill the cache with to remove this case (and
> also remove the "size" handling entirely)? It would need to have a key value
> that is never an input to the function (e.g., a newly created object that
isn't
> used anywhere else).
> 

Caches are allocated filled with holes, so it's already the case.  However,
could you explain what's the benefit of omitting this check?

And as we iterate backwards, that seems difficult to remove size completely. 

Probably I miss some elegant solution.

http://codereview.chromium.org/1709001/diff/1/2#newcode10098
src/runtime.cc:10098: int target_index = finger_index + 2;
On 2010/04/21 07:26:35, Lasse Reichstein wrote:
> This constant, 2, should have a name.
> Perhaps add  JSFunctionResultCache::kEntrySize = 2  and use it here.

Sure.