I plan on renumbering the units to microseconds and introducing microsecond profiling to osx only where it work reasonably well. We can then progressively roll out the feature to other backends.

try: -b do -p linux,linux64,macosx64,win32,android -u none -t none

I'll come back to this later.

With this patch we support microsecond profiling in the mac/linux backend. The windows backend just ignores it for now. The profiler does not guarantee the interval is respected. I want to continue improving the UI to display what was requested and how accurate the samples match the request interval.

Comment on attachment 720897 [details] [diff] [review] cleaner patch Review of attachment 720897 [details] [diff] [review]: ----------------------------------------------------------------- ::: tools/profiler/platform-macos.cc @@ +281,5 @@ > } > thread_resume(profiled_thread); > } > > + int intervalMicro_; This is not needed. ::: tools/profiler/platform-win32.cc @@ +72,5 @@ > + { > + interval_ = floor(interval + 0.5); > + if (interval_ < 0) { > + interval_ = 1; > + } Why this check?

(In reply to :Ehsan Akhgari from comment #7) > Comment on attachment 720897 [details] [diff] [review] > cleaner patch > > Review of attachment 720897 [details] [diff] [review]: > ----------------------------------------------------------------- > > ::: tools/profiler/platform-macos.cc > @@ +281,5 @@ > > } > > thread_resume(profiled_thread); > > } > > > > + int intervalMicro_; > > This is not needed. The name change is because by default the units are millisecond weather the type is int or double. In this place the units are explicitly microseconds. The const is just to be consistent with the change in the win32 backend. > > ::: tools/profiler/platform-win32.cc > @@ +72,5 @@ > > + { > > + interval_ = floor(interval + 0.5); > > + if (interval_ < 0) { > > + interval_ = 1; > > + } > > Why this check? Actually this should be <= 0. I wanted to guard against the possibility of rounding down to zero and/or getting a bad value. I wanted to enforce a minimum of 1.

(In reply to comment #8) > (In reply to :Ehsan Akhgari from comment #7) > > Comment on attachment 720897 [details] [diff] [review] > > cleaner patch > > > > Review of attachment 720897 [details] [diff] [review]: > > ----------------------------------------------------------------- > > > > ::: tools/profiler/platform-macos.cc > > @@ +281,5 @@ > > > } > > > thread_resume(profiled_thread); > > > } > > > > > > + int intervalMicro_; > > > > This is not needed. > > The name change is because by default the units are millisecond weather the > type is int or double. In this place the units are explicitly microseconds. The > const is just to be consistent with the change in the win32 backend. OK. > > ::: tools/profiler/platform-win32.cc > > @@ +72,5 @@ > > > + { > > > + interval_ = floor(interval + 0.5); > > > + if (interval_ < 0) { > > > + interval_ = 1; > > > + } > > > > Why this check? > > Actually this should be <= 0. I wanted to guard against the possibility of > rounding down to zero and/or getting a bad value. I wanted to enforce a minimum > of 1. So why don't we have a similar check for other backends?

Ready to land, waiting on bug 779291 to minimize conflicts

try: -b do -p all -u xpcshell -t none

I landed a trivial mingw fixup (missing math.h include): https://hg.mozilla.org/integration/mozilla-inbound/rev/177637fe281f