Fix 13058 - Thread priority handling doesn't work in Solaris #871
Conversation
Conflicts: posix.mak Merged upstream
| enum PC_KY_NULL = 0; // kv chain terminator | ||
| enum PC_KY_CLNAME = 1; // get the class name of a process or LWP | ||
|
|
||
| struct pc_vaparm_t { |
There was a problem hiding this comment.
} on a separate line please ;-)
|
LGTM except for a few code style nitpicks. I don't know if this is the 'correct' way to solve this issue for solaris but I doubt there's anyone in the core contributor team who has enough solaris experience to double-check this. But the code is well-documented and no reason not to trust @jasonbking that this solution is correct and it certainly won't do any harm. So if these minor style issues are fixed this should be ready to get merged. |
|
Should all be fixed (plus noticed a very minor typo in one of the exception messages). I can point to the relevant Illumos code, though it'd be a bit involved as both posix functions get about 2-3 functions deep before you get to the common function they all call (which is where the bug is), though it also shows they just call priocntl(2) to implement their interfaces. |
Fix 13058 - Thread priority handling doesn't work in Solaris
This fixes (or rather works around a long standing Solaris bug) dealing with adjusting a thread's priority. The workaround is to merely utilize the underlying interfaces that the posix functions call.
Utilizing the underlying interface also allows us to obtain the effective maximum priority instead of the absolute maximum. On Solaris systems, typically priorities range from [-60, 60], however non-privileged processes have a cap at 0, effectively limiting them to [-60, 0] (though an admin can change this on the fly). This includes code that (at startup) determines the effective range and uses that as that seems likely to be the best fit to the existing thread.d API.
With the workaround, the tests pass, without it, it fails in adjusting the thread priority back to it's original value.
https://issues.dlang.org/show_bug.cgi?id=13058