Fix for bug 6767 - Limit stack trace collection for stack overflow exceptions.

Fix for bug 6767 - Limit stack trace collection for stack overflow exceptions.

Use pre allocated stack object for collecting strack trace in the case of OOM
and stack overflow exceptions.
Committed: https://code.google.com/p/dart/source/detail?r=19074

[hausner, 2013-02-25 23:09:08]

LGTM w/comments

https://chromiumcodereview.appspot.com/12320103/diff/1009/runtime/vm/exceptio...
File runtime/vm/exceptions.cc (right):

https://chromiumcodereview.appspot.com/12320103/diff/1009/runtime/vm/exceptio...
runtime/vm/exceptions.cc:105: for (intptr_t i = start; i <
Stacktrace::kPreallocatedStackdepth; i++) {
You could implement the top N slots as a circular buffer, but it's probably not
worth the complexity.

https://chromiumcodereview.appspot.com/12320103/diff/1009/runtime/vm/raw_obje...
File runtime/vm/raw_object_snapshot.cc (right):

https://chromiumcodereview.appspot.com/12320103/diff/1009/runtime/vm/raw_obje...
runtime/vm/raw_object_snapshot.cc:2227: ASSERT(reader != NULL);
How about simply (pre-)allocating a new stack trace object instead of writing
the pre-allocated one to the snapshot and reading it back in?

[siva, 2013-02-26 18:49:48]

Thanks.

https://chromiumcodereview.appspot.com/12320103/diff/1009/runtime/vm/exceptio...
File runtime/vm/exceptions.cc (right):

https://chromiumcodereview.appspot.com/12320103/diff/1009/runtime/vm/exceptio...
runtime/vm/exceptions.cc:105: for (intptr_t i = start; i <
Stacktrace::kPreallocatedStackdepth; i++) {
True. Will consider that as an optimization for another CL.

On 2013/02/25 23:09:08, hausner wrote:
> You could implement the top N slots as a circular buffer, but it's probably
not
> worth the complexity.

https://chromiumcodereview.appspot.com/12320103/diff/1009/runtime/vm/raw_obje...
File runtime/vm/raw_object_snapshot.cc (right):

https://chromiumcodereview.appspot.com/12320103/diff/1009/runtime/vm/raw_obje...
runtime/vm/raw_object_snapshot.cc:2227: ASSERT(reader != NULL);
I tried doing what you suggested, it makes the code too hacky as it now requires
special handling for the object store during snapshot writes (the object store
is currently dumped as an object during writes and would require special
handling of the preallocated_stack_trace field to not write it out).

On 2013/02/25 23:09:08, hausner wrote:
> How about simply (pre-)allocating a new stack trace object instead of writing
> the pre-allocated one to the snapshot and reading it back in?

[siva, 2013-02-26 19:07:50]

Committed patchset #3 manually as r19074 (presubmit successful).