GTTF: Avoid assertion failures caused by clock drift....

GTTF: Avoid assertion failures caused by clock drift.

Instead, we add a test dedicated to verifying the QPC
time drift. Otherwise the assertion is being flakily hit
in unrelated tests, which crashes the entire test binary.

With a dedicated test, we still have necessary monitoring,
but it's in one place.

TEST=base_unittests, net_unittests (stop crashing)
BUG=none

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=54978

[cpu_(ooo_6.6-7.5), 2010-08-03 02:11:06]

are you missing the .cc file?

On 2010/08/02 23:54:53, Paweł Hajdan Jr. wrote:
>

[Paweł Hajdan Jr., 2010-08-03 16:47:39]

On 2010/08/03 02:11:06, cpu wrote:
> are you missing the .cc file?

Oh, yes. Uploaded now. It was stuck in some other stale gcl changelist, eh.
Please take another look.

[brettw, 2010-08-03 18:17:35]

http://codereview.chromium.org/3082014/diff/6001/7002
File base/time_win.cc (right):

http://codereview.chromium.org/3082014/diff/6001/7002#newcode345
base/time_win.cc:345: return abs((UnreliableNow() - ReliableNow()) - skew_);
Is this right? It seems like the sign of UnreliableNow() - ReliableNow() could
be either positive or negative. So I don't really know what it means when you
subtract skew_ from that. Do you need another abs in there?

[cpu_(ooo_6.6-7.5), 2010-08-04 17:44:41]

http://codereview.chromium.org/3082014/diff/6001/7002
File base/time_win.cc (right):

http://codereview.chromium.org/3082014/diff/6001/7002#newcode345
base/time_win.cc:345: return abs((UnreliableNow() - ReliableNow()) - skew_);
On 2010/08/03 18:17:35, brettw wrote:
> Is this right? It seems like the sign of UnreliableNow() - ReliableNow() could
> be either positive or negative. So I don't really know what it means when you

Seems ok to me. You *could* remove the outer abs() and then just say that the
drift can be positive or negative.

Let me elaborate with a table

First case one clock faster than the other in a widening case

time   U       R      Drift
0      10      11     (computed skew = -1)
10     20      22     -2 - (-1) = -1     
 
Second case is the same but inverted

time   U       R      Drift
0      11      10     (computed skew = 1)
10     22      20     2 - 1 =  1     

If so then you put the abs() in the test itself.

> subtract skew_ from that. Do you need another abs in there?

[cpu_(ooo_6.6-7.5), 2010-08-04 17:49:02]

lgtm btw

On 2010/08/04 17:44:41, cpu wrote:
> http://codereview.chromium.org/3082014/diff/6001/7002
> File base/time_win.cc (right):
> 
> http://codereview.chromium.org/3082014/diff/6001/7002#newcode345
> base/time_win.cc:345: return abs((UnreliableNow() - ReliableNow()) - skew_);
> On 2010/08/03 18:17:35, brettw wrote:
> > Is this right? It seems like the sign of UnreliableNow() - ReliableNow()
could
> > be either positive or negative. So I don't really know what it means when
you
> 
> Seems ok to me. You *could* remove the outer abs() and then just say that the
> drift can be positive or negative.
> 
> Let me elaborate with a table
> 
> First case one clock faster than the other in a widening case
> 
> time   U       R      Drift
> 0      10      11     (computed skew = -1)
> 10     20      22     -2 - (-1) = -1     
>  
> Second case is the same but inverted
> 
> time   U       R      Drift
> 0      11      10     (computed skew = 1)
> 10     22      20     2 - 1 =  1     
> 
> If so then you put the abs() in the test itself.
> 
> > subtract skew_ from that. Do you need another abs in there?

[brettw, 2010-08-04 22:14:44]

LGTM