Convert render thread idle delays from seconds to milliseconds.

Convert render thread idle delays from seconds to milliseconds.

This CL eliminates double to integer conversions when scheduling the idle timer and allows subsecond delays.

The existing code simply rounds the delay down to an integer when scheduling the idle timer, which means that any subsecond delay schedules a timer with zero delay.

BUG=
TEST=


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=109866

[ulan, 2011-11-11 17:10:03]

Please take a look. This CL eliminates double to integer conversions when
scheduling the idle timer and allows subsecond delays.

http://codereview.chromium.org/8463019/diff/1/content/renderer/render_thread_...
File content/renderer/render_thread_impl.cc (right):

http://codereview.chromium.org/8463019/diff/1/content/renderer/render_thread_...
content/renderer/render_thread_impl.cc:579: 1000000 /
(idle_notification_delay_in_ms_ + 2000));
Since this computation is in integers, it is not exactly equivalent to the old
version in doubles. The error however is negligible: after a year of running the
error will be less than 40 seconds as demonstrated by the following program:

#include <cstdio>
#include <cmath>
#include <algorithm>
using namespace std;

int main() {
  double s = 1.0;
  long long ms = 1000;
  double error_s = 0;
  double total_time_s = s;
  for (int i = 0; i < 100000; ++i) {
    s = s + 1.0 / (s + 2.0);
    ms = ms + 1000000 / (ms + 2000);
    error_s = max(error_s, fabs(s - ms / 1000.0));
    total_time_s += s;
  }
  printf("%.3lf - %.3lf = %.3lf\n", s, ms / 1000.0, s - ms / 1000.0);
  printf("error %.3lf after %.lf days\n", error_s, total_time_s / 3600 / 24);
  return 0;
}

[jam, 2011-11-11 18:10:09]

lgtm

http://codereview.chromium.org/8463019/diff/1/chrome/renderer/extensions/exte...
File chrome/renderer/extensions/extension_dispatcher.cc (right):

http://codereview.chromium.org/8463019/diff/1/chrome/renderer/extensions/exte...
chrome/renderer/extensions/extension_dispatcher.cc:35: static const int64
kInitialExtensionIdleHandlerDelayMs = 5*1000 /* ms */;
nit, the "/* ms */" is redundant since Ms is in the constant's name

[Matt Perry, 2011-11-11 22:09:25]

LGTM, but what's the motivation for this change? Could you add it to the CL
description or link it to a bug?

[ulan, 2011-11-14 10:12:42]

On 2011/11/11 22:09:25, Matt Perry wrote:
> LGTM, but what's the motivation for this change? Could you add it to the CL
> description or link it to a bug?

I updated the description. The motivation is to allow subsecond delays and avoid
potential double-to-integer conversion bugs (like converting 1-epsilon). The
existing code simply rounds the delay down to an integer when scheduling the
idle timer, so any subsecond delays leads to a constantly firing timer.

This CL prepares ground for my other CL, which detects user idleness and invokes
incremental garbage collection.

I'd like to use the commit queue, so I will include the fix to the "redundant
ms" nit in the other CL.

[commit-bot: I haz the power, 2011-11-14 10:12:51]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/ulan@chromium.org/8463019/1

[commit-bot: I haz the power, 2011-11-14 10:24:42]

Try job failure for 8463019-1 (retry) on win_rel for step "compile" (clobber
build).
It's a second try, previously, step "compile" failed.
http://build.chromium.org/p/tryserver.chromium/buildstatus?builder=win_rel&nu...

[commit-bot: I haz the power, 2011-11-14 11:03:12]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/ulan@chromium.org/8463019/8001

[commit-bot: I haz the power, 2011-11-14 12:09:46]

Change committed as 109866