The Mac Selenium tests are flaky, failing every now and then because of a...

The Mac Selenium tests are flaky, failing every now and then because of a
Python "error 32: Broken Pipe" while sending the .nexe from the Selenium
server to the browser.  This might be caused by the "timeout" for "new
NaclLib()" being too short so the Javascript abandons loading the page before
the .nexe is fully transferred.  As such, increase the "timeout" (it's
actually a loop count) from 100 to 500 (sometimes from 50 to 500) for most of
the uses of "new NaclLib()".


Committed: http://code.google.com/p/nativeclient/source/detail?r=2574

[noelallen_use_chromium, 2010-06-24 17:13:52]

The changes look harmless other than potentially increasing the time it takes to
fail.  Considering some of the timeouts are set at 1000, having some set as low
as 50 does seem dubious, and if it only takes longer in cases where we are
failing, then the possibility of a longer failure case (a few seconds) is worth
it IMO, if it decreases the likelihood of needing to retry and/or debug which is
much longer than seconds.

While we agreed that the check-in would wait for confirmation that it fixes the
broken pipe issue, I think it's actually tangent to wither or not this change
should be made.   Confirmation that the pipe issue is tied to this timeout lets
us close the bug, and give us high confidence that this needs to be done, where
as confirmation that the pipe issue is not directly related to this timeout,
does not invalidate increasing the timeouts in principal. So either way:

LGTM

[Mark Schneckloth, 2010-06-25 18:35:12]

I have been playing with various timeout values on my machine.  No matter how
small I set the timeout (even 1) I cannot induce a failure of a Selenium test. 
So this CL is unlikely to fix the "error 32: Broken pipe" problem.

As pointed out by Noel, increasing the timeout from 100 to 500 (the units are
"number of 100ms loops" so the increase is from 10 to 50 seconds), is not a bad
thing.  It pretty much guarantees that we won't have any flaky errors in the
future where we didn't wait long enough for the .nexe to load.  So I will commit
this and then keep looking for the underlying cause of the "error 32: Broken
pipe" Selenium test error.