complete out the os.sched_getaffinity error set #6886
Conversation
komuw
commented
Oct 30, 2020
komuw
commented
- complete out the os.sched_getaffinity error set
- add doc string for sched_getaffinity
- add test for os.sched_getaffinity
| @@ -601,3 +601,7 @@ test "getrlimit and setrlimit" { | |||
| const cpuLimit = try os.getrlimit(.CPU); | |||
| try os.setrlimit(.CPU, cpuLimit); | |||
| } | |||
|
|
|||
| test "os.sched_getaffinity" { | |||
| testing.expectError(error.NoThread, os.sched_getaffinity(std.math.maxInt(i32))); | |||
There was a problem hiding this comment.
My assumption here is that we will never have as many threads as maxInt(i32) when running tests.
But there is a good probability that I'm wrong. Thoughts?
There was a problem hiding this comment.
On windows and Osx, this is failing with
error: use of undeclared identifier 'cpu_set_t'
pub fn sched_getaffinity(pid: pid_t) SchedGetAffinityError!cpu_set_t {
Should I restrict this test to linux only?
There was a problem hiding this comment.
On windows and Osx, this is failing with
Welcome to the world of cross-platform non-posix abstractions (and yet another opportunity to plug #6600):
- NetBSD has
pthread_setaffinity_np/pthread_getaffinity_np - FreeBSD has
cpuset_getaffinity/cpuset_setaffinity - macOS has no direct equivalent,
thread_policy_get/thread_policy_setlet you get/set the affinity tag for a given thread but that's it - OpenBSD has no such API (cc @semarie)
My assumption here is that we will never have as many threads as maxInt(i32) when running tests.
That's a pretty safe assumption (why are PIDs signed???), spawning that many threads requires an ungodly amount of memory heh
There was a problem hiding this comment.
I have skipped the test for non-linux OSes. I have also opened #6907 to implement sched_getaffinity for other OSes. I'll attempt to do that in a follow up PR, if no one else will have done so already.
|
* OpenBSD has no such API (cc @semarie)
hep. I confirm that it doesn't exist on OpenBSD.
Depending your intent for `os.sched_getaffinity()` interface, such cases could be:
- a `@compileError`, but it means any use of it will break compilation on such platforms
- a nop (returning a constant on `os.sched_getaffinity`, and
silencious ignore value setting on `os.sched_setaffinity`).
the linux man page documents it as "setting the CPU affinity mask can
be used to obtain performance benefits". so I prefer a functional
program but with less performance instead of an unbuildable program.
|
Vexu
added
the
standard library
|
The code is correct in master branch; those error codes are not reachable, except by programmer error. |
