When asserting that an event object is not signaled,...

When asserting that an event object is not signaled,
pass important values (the return value of WaitForSingleObject
and the error code) as function arguments so that they are
available in crash dumps.

R=eroman
BUG=51950, 52008
TEST=none

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=57247

[wtc, 2010-08-24 01:02:56]

eroman: will this work as intended?

I added the CHECK log message to prevent the compiler from
optimizing away the function arguments rv and err.

[eroman, 2010-08-24 02:37:47]

LGTM.

> eroman: will this work as intended?

I believe that function will get inlined by whole-program-optimization.

I recommend doing something like this:

CHECK(false);  // Crash.

// This log is unreachable since we have already crashed, but it should prevent
the compiler from optimizing away the |rv| and |err| variables so they appear
nicely on the stack.

LOG(INFO) << "rv=" << rv << ", err=" << err;

[wtc, 2010-08-24 17:07:09]

eroman:  Thank you for the suggestion.  I don't understand why

    CHECK(failed);
    LOG(INFO) << rv << err;

is better than

    CHECK(failed) << rv << err;

at preventing the compiler from optimizing away the |rv| and |err|
variables.

[eroman, 2010-08-24 18:46:14]

We just discussed this in person, for the record the issue is with how our
logging works. Conceptually code runs in reverse order to how it is written.

So this code:

  CHECK(false) << foo << bar;

Will actually run like:
(1) Make foo and bar into a string, emit it to the log.
(2) Crash

Therefore "foo" and "bar" are not live by the time the crash occurs (rather they
are hidden away in some heap string which we can't access).

By contrast, if we Crash first, and THEN log the values, the compiler will have
to keep the variables at the time of the crash.

[wtc, 2010-08-24 18:57:13]

Please review Patch Set 4.  Thanks.

[eroman, 2010-08-24 19:01:44]

LGTM