pkg: Eliminate flaky analysis test failure.

pkg: Eliminate flaky analysis test failure.

Running in debug mode (or any slower configuration) makes this test fail since
it waits for a max of 100 event loop iterations for the analysis results.

Instead wait without limit until results are ready, with checking every few milliseconds.


BUG=
R=scheglov@google.com

Committed: https://github.com/dart-lang/sdk/commit/c4d0ff8a4c580821ffb63ca3a65d292c874b2914

[Florian Schneider, 2015-10-01 16:42:17]

fschneider@google.com changed reviewers:
+ scheglov@google.com

[scheglov, 2015-10-01 16:59:41]

Couple questions.

1. Will this change to cause tests to get stuck infinitely, if we run them
outside of test bots?

2. Why does slow configuration require more cycles? I thought a cycle is a
single operation, not just some time passed.

3. I think in some cases computing this information takes less than 10 ms, so
maybe we could reduce it to 1 ms.

[Florian Schneider, 2015-10-01 17:09:12]

On 2015/10/01 16:59:41, scheglov wrote:
> Couple questions.
> 
> 1. Will this change to cause tests to get stuck infinitely, if we run them
> outside of test bots?
> 
Only if the analysis result never arrives. This would a bug in the analysis,
wouldn't it?

I can also imagine putting a generous timeout value in (e.g. 30s).


> 2. Why does slow configuration require more cycles? I thought a cycle is a
> single operation, not just some time passed.
>

I'm not familiar with the analysis server. The VM takes considerably more time
to compile and run the analyzer in DEBUG mode. We don't run this configuration
on the bots, but we use it locally to test the VM.

It seems to return to the event loop often enough to make this test fail. Is is
run in a different isolate?
 
> 3. I think in some cases computing this information takes less than 10 ms, so
> maybe we could reduce it to 1 ms.

I guess 1ms would work fine.

[scheglov, 2015-10-01 17:19:06]

With some timeout LGTM.

On 2015/10/01 17:09:12, Florian Schneider wrote:
> On 2015/10/01 16:59:41, scheglov wrote:
> > Couple questions.
> > 
> > 1. Will this change to cause tests to get stuck infinitely, if we run them
> > outside of test bots?
> > 
> Only if the analysis result never arrives. This would a bug in the analysis,
> wouldn't it?
> 
> I can also imagine putting a generous timeout value in (e.g. 30s).
> 
> 
> > 2. Why does slow configuration require more cycles? I thought a cycle is a
> > single operation, not just some time passed.
> >
> 
> I'm not familiar with the analysis server. The VM takes considerably more time
> to compile and run the analyzer in DEBUG mode. We don't run this configuration
> on the bots, but we use it locally to test the VM.
> 
> It seems to return to the event loop often enough to make this test fail. Is
is
> run in a different isolate?
>  
> > 3. I think in some cases computing this information takes less than 10 ms,
so
> > maybe we could reduce it to 1 ms.
> 
> I guess 1ms would work fine.

[Florian Schneider, 2015-10-01 17:27:36]

Done.

[scheglov, 2015-10-01 17:31:14]

LGTM
Thanks.

[Florian Schneider, 2015-10-01 17:35:02]

Committed patchset #2 (id:20001) manually as
c4d0ff8a4c580821ffb63ca3a65d292c874b2914 (presubmit successful).