Speed up GetRandomBytes by only entering signal-blocking scope once.

Speed up GetRandomBytes by only entering signal-blocking scope once.

BUG=
R=sgjesse@google.com

Committed: https://code.google.com/p/dart/source/detail?r=34356

[Søren Gjesse, 2014-03-25 08:21:23]

LGTM, with comments

https://codereview.chromium.org/209333014/diff/1/runtime/bin/crypto_macos.cc
File runtime/bin/crypto_macos.cc (right):

https://codereview.chromium.org/209333014/diff/1/runtime/bin/crypto_macos.cc#...
runtime/bin/crypto_macos.cc:20: ThreadSignalBlocker signal_blocker(SIGPROF);
Maybe there are other places where we can do the same?

https://codereview.chromium.org/209333014/diff/1/runtime/platform/signal_bloc...
File runtime/platform/signal_blocker.h (right):

https://codereview.chromium.org/209333014/diff/1/runtime/platform/signal_bloc...
runtime/platform/signal_blocker.h:84: #define
TEMP_FAILURE_RETRY_NO_SIGNAL_BLOCKER(expression)                       \
It would be great if this macro could check that a ThreadSignalBlocker is in
scope.

The simple way is to just make it take a ThreadSignalBlocket reference, that it
just asserted to be not NULL in the macro.

ThreadSignalBlocker sb(...)

TEMP_FAILURE_RETRY_NO_SIGNAL_BLOCKER(&sb, ...)

It should brobably be called something like

#define TEMP_FAILURE_RETRY_WITH_SIGNAL_BLOCKER(sb, expresstion)

[Anders Johnsen, 2014-03-25 13:06:03]

https://codereview.chromium.org/209333014/diff/1/runtime/bin/crypto_macos.cc
File runtime/bin/crypto_macos.cc (right):

https://codereview.chromium.org/209333014/diff/1/runtime/bin/crypto_macos.cc#...
runtime/bin/crypto_macos.cc:20: ThreadSignalBlocker signal_blocker(SIGPROF);
On 2014/03/25 08:21:23, Søren Gjesse wrote:
> Maybe there are other places where we can do the same?

Found another use-case in eventhandler :)

https://codereview.chromium.org/209333014/diff/1/runtime/platform/signal_bloc...
File runtime/platform/signal_blocker.h (right):

https://codereview.chromium.org/209333014/diff/1/runtime/platform/signal_bloc...
runtime/platform/signal_blocker.h:84: #define
TEMP_FAILURE_RETRY_NO_SIGNAL_BLOCKER(expression)                       \
On 2014/03/25 08:21:23, Søren Gjesse wrote:
> It would be great if this macro could check that a ThreadSignalBlocker is in
> scope.
> 
> The simple way is to just make it take a ThreadSignalBlocket reference, that
it
> just asserted to be not NULL in the macro.
> 
> ThreadSignalBlocker sb(...)
> 
> TEMP_FAILURE_RETRY_NO_SIGNAL_BLOCKER(&sb, ...)
> 
> It should brobably be called something like
> 
> #define TEMP_FAILURE_RETRY_WITH_SIGNAL_BLOCKER(sb, expresstion)

I've changed it to validate in debug mode, that it's blocking SIGPROF calls.

[Anders Johnsen, 2014-03-25 13:07:01]

Committed patchset #2 manually as r34356 (presubmit successful).