Of the other benchmarks, closure-datetime spends about half of its
time in DST calculations. And judging from the results of JSC, they do
cache DST intensively.

I'll upload a new version of my change later today. It will cache less
aggressively, then it does now.

On Mon, May 17, 2010 at 11:29,  <[email protected]> wrote:
> I don't think the date-parse benchmark is a good reason to change the DST
> calculation this drastically.  It is an artificial benchmark, not used in
> published results.  The principle of relying on the OS to tell us what is in
> DST
> and what is not is important, I think, and there are dangers to abandoning
> it.
> Do other, important benchmarks, also spend most of their time in DST?
>
> The original problem was to find a way to make V8 notice changes in the time
> zone.  We can do this on context creation, on a timer, or some other way.
>  Right
> now, I think we do it on missing the DST cache, and also on fetching the
> string
> for the time zone.
>
>
>
> http://codereview.chromium.org/1758013/diff/4001/5003
> File src/platform-posix.cc (right):
>
> http://codereview.chromium.org/1758013/diff/4001/5003#newcode111
> src/platform-posix.cc:111: time_t tv = static_cast<time_t>(time /
> msPerSecond);
> Eliminating the floor call changes the answer if time is negative.  Are
> we not calling this function with negative values?  If so, comment.
>
> http://codereview.chromium.org/1758013/diff/4001/5006
> File src/runtime.cc (right):
>
> http://codereview.chromium.org/1758013/diff/4001/5006#newcode7280
> src/runtime.cc:7280: static int EquivalentTime(double t) {
> This doesn't work.  There are also special rules for years divisible by
> 1000, and there is the calendar shift to the Gregorian calendar.  Also,
> Windows now can return different DST values for different equivalent
> years in the range 1970 - present, and for compatibility, we report
> those different DST values.  Only for years outside the range
> 1970-2037(?) do we use the equivalent year.
>
> http://codereview.chromium.org/1758013/show
>

-- 
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev

Reply via email to