Use stub where pthread_setaffinity_np() isn't available. define cpu_set_t to cpuset_t on Free/NetB…

Use stub where pthread_setaffinity_np() isn't available. define cpu_set_t to cpuset_t  on Free/NetBSD. Patch originally by Jan Beich

[reed1, 2013-08-06 19:35:26]

works for me, but want another set of eyes

[bungeman-skia, 2013-08-06 19:38:31]

This file is called _linux for a reason. This code should go in
SkThreadUtils_pthread_freebsd.cpp so that the build can select the code, instead
of relying on a number of ifdefs.

[bungeman-skia, 2013-08-06 20:44:12]

Just wanted to point out that this is already something of a mess. Skia itself
(core) doesn't want the threading; I believe only the tools do. This was written
just to test reference counting, but is now being used by other tools. This code
should, as a result, probably all live in tools in the long term. It's also not
written all that well, in particular starting the thread in the constructor is
terrible.

We've been trying as hard as possible to move away from using any platform
ifdefs in source files, and trying to move platform selection into the build.
Currently we have a bunch in the configuration headers, but those pre-date any
real build system, and it would be nice to have them documented by the build
instead as well.

For this change it would at least be nice to have this code in a header named
something like SkThreadUtils_posix_makeBsdLookLikeLinux.h or something, since
that's effectively what it's doing. Better would be an implementation _bsd.cpp
which uses sys/cpuset.h and such directly, not unlike the _mach.cpp.

Also, we don't currently have a bsd builder of any sort. As a result it's really
easy for this sort of thing to break and just get removed later because we're
not testing it. Should we set up a builder for a bsd?

https://codereview.chromium.org/22259005/diff/1/src/utils/SkThreadUtils_pthre...
File src/utils/SkThreadUtils_pthread_linux.cpp (right):

https://codereview.chromium.org/22259005/diff/1/src/utils/SkThreadUtils_pthre...
src/utils/SkThreadUtils_pthread_linux.cpp:21: #define cpu_set_t cpuset_t
isn't this what typedef is for?

[wrightg, 2013-08-07 18:48:17]

On 2013/08/06 20:44:12, bungeman1 wrote:
> Also, we don't currently have a bsd builder of any sort. As a result it's
really
> easy for this sort of thing to break and just get removed later because we're
> not testing it. Should we set up a builder for a bsd?

This patch is actually from our external contributors as we (Mozilla) don't have
any BSD builders either. I'm quite happy to disable Skia in Firefox on BSD
platforms (and thus stop getting/submitting BSD-specific patches) if you don't
feel that you don't have the resources to prioritise BSD.

[bungeman-skia, 2013-08-07 19:29:13]

On 2013/08/07 18:48:17, wrightg wrote:
> On 2013/08/06 20:44:12, bungeman1 wrote:
> > Also, we don't currently have a bsd builder of any sort. As a result it's
> really
> > easy for this sort of thing to break and just get removed later because
we're
> > not testing it. Should we set up a builder for a bsd?
> 
> This patch is actually from our external contributors as we (Mozilla) don't
have
> any BSD builders either. I'm quite happy to disable Skia in Firefox on BSD
> platforms (and thus stop getting/submitting BSD-specific patches) if you don't
> feel that you don't have the resources to prioritise BSD.

I believe that SkThreadUtils is only used by tools, not by anything in core. Is
it possible for you to just not build this file? Is there any reason why the
tools are needed for the general build?