Fix issue 21795.

Fix issue 21795.
Add regression test.

R=fschneider@google.com

Committed: https://code.google.com/p/dart/source/detail?r=42824

[regis, 2015-01-10 00:43:18]

regis@google.com changed reviewers:
+ fschneider@google.com, vegorov@google.com

[regis, 2015-01-10 00:58:16]

https://codereview.chromium.org/847613002/diff/1/runtime/vm/code_descriptors.h
File runtime/vm/code_descriptors.h (right):

https://codereview.chromium.org/847613002/diff/1/runtime/vm/code_descriptors....
runtime/vm/code_descriptors.h:138: list_[try_index].needs_stacktrace =
needs_stacktrace;
I rushed to catch my shuttle and I realize now (on the shuttle) that this could
overwrite a 'true' value set earlier by SetNeedsStacktrace, or would the correct
value be passed in this case?

I will let you decide of the proper fix, since I cannot debug from the shuttle
and more importantly since you understand this much better than me. Thanks!

[Florian Schneider, 2015-01-12 12:47:49]

Yes, the point is that we can't rely on any specific order in which blocks are
visited in the back-end.

https://codereview.chromium.org/847613002/diff/1/runtime/vm/code_descriptors.h
File runtime/vm/code_descriptors.h (right):

https://codereview.chromium.org/847613002/diff/1/runtime/vm/code_descriptors....
runtime/vm/code_descriptors.h:138: list_[try_index].needs_stacktrace =
needs_stacktrace;
On 2015/01/10 00:58:16, regis wrote:
> I rushed to catch my shuttle and I realize now (on the shuttle) that this
could
> overwrite a 'true' value set earlier by SetNeedsStacktrace, or would the
correct
> value be passed in this case?
> 
> I will let you decide of the proper fix, since I cannot debug from the shuttle
> and more importantly since you understand this much better than me. Thanks!
> 

You're right. The flag may be overridden, so AddHandler needs to check here if
it is already set. Otherwise, the stacktrace of the re-thrown exception will be
wrong. I attached a regression test to the issue.

[regis, 2015-01-12 19:37:08]

Thanks for the regression test!

PTAL

[Florian Schneider, 2015-01-13 12:58:53]

LGTM.

[regis, 2015-01-13 17:16:56]

Committed patchset #2 (id:20001) manually as r42824 (presubmit successful).