Fix for crash in histogram BucketIndex

Added CHECKs to make sure histogram caller's don't create
histogram that are larger 1G. Added CHECK's to catch bucket_count_
is not corrupted before accessing ranges_ vector.

BUG=73452
TEST=none. histogram unit tests are enough.


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=75804

[ramant (doing other things), 2011-02-19 21:07:55]

Hi Jim,
  A cut at adding checks that bucket_count_ was not corrupted by memory
thrashers.

thanks much,
raman

[ramant (doing other things), 2011-02-23 02:31:19]

Hi Jim,
  Removed all the checks except the mid+1 check (and the small tweaks for
calculating the mid). Would appreciate your comments.

thanks,
raman

[jar (doing other things), 2011-02-23 18:49:33]

LGTM

I'd really like to see a smaller number, such as 10,000.  You mentioned that
someone is using 64,000, and I'd like to get that fixed, so we can have a lower
threshold.

Once you support sparse histograms, the limit could probably drop down to
something more like 1000 for routine histograms.

[ramant (doing other things), 2011-02-23 22:34:34]

On 2011/02/23 18:49:33, jar wrote:
> LGTM
> 
> I'd really like to see a smaller number, such as 10,000.  You mentioned that
> someone is using 64,000, and I'd like to get that fixed, so we can have a
lower
> threshold.
> 
> Once you support sparse histograms, the limit could probably drop down to
> something more like 1000 for routine histograms.

Hi Jim,
  Changed the max bucket_count to 10,000.
thanks.