setitimer with Cygwin and Hurd

setitimer with Cygwin and Hurd

[Mike Gran]

Hi-
In trying to get Guile master to build on a few
different VMs, I've run into a problem with setitimer and
getitimer for ITIMER_PROF and ITIMER_VIRTUAL.
On Hurd, calls to setitimer and getitimer for the profiling
and virtual timers will always fail with ENOSYS.
On Cygwin, calls to setitimer and getitimer for the
profiling and virtual timers will always fail with EINVAL.
I figure the way forward is to patch the test suite
so that any tests that require the profiling or virtual
itimers is 'unsupported.  But you'd end up with quite a
bit of special casing in a half-dozen of the test scripts.
Anyone have a better suggestion?
-Mike Gran

Re: setitimer with Cygwin and Hurd

[Andy Wingo]

On Mon 06 Mar 2017 06:46, Mike Gran <spk121@yahoo.com> writes:

> On Hurd, calls to setitimer and getitimer for the profiling
> and virtual timers will always fail with ENOSYS.
> On Cygwin, calls to setitimer and getitimer for the
> profiling and virtual timers will always fail with EINVAL.

Is it possible to avoid defining these interfaces if they will never
work?

Is that even a good idea? :)  Maybe not.  Would be better to have
(defined? 'setitimer) or so.  If not, a setitimer feature I guess --
(provided? 'setitimer).

Andy

Re: setitimer with Cygwin and Hurd

[Mike Gran]

>> On Hurd, calls to setitimer and getitimer for the profiling
>> and virtual timers will always fail with ENOSYS.
>> On Cygwin, calls to setitimer and getitimer for the
>> profiling and virtual timers will always fail with EINVAL.

>Is it possible to avoid defining these interfaces if they will never
>work?

>Is that even a good idea? :)  Maybe not.  Would be better to have
>(defined? 'setitimer) or so.  If not, a setitimer feature I guess --
>(provided? 'setitimer).
This is what I came up with
branch wip-itimer-checks
commit f5b362586d7622c408d4402c7cb496c45ffb56e8
What do you think?
-Mike

Re: setitimer with Cygwin and Hurd

[Andy Wingo]

On Tue 07 Mar 2017 15:31, Mike Gran <spk121@yahoo.com> writes:

>>> On Hurd, calls to setitimer and getitimer for the profiling
>>> and virtual timers will always fail with ENOSYS.
>>> On Cygwin, calls to setitimer and getitimer for the
>>> profiling and virtual timers will always fail with EINVAL.
>
>>Is it possible to avoid defining these interfaces if they will never
>>work?
>
>>Is that even a good idea? :)  Maybe not.  Would be better to have
>>(defined? 'setitimer) or so.  If not, a setitimer feature I guess --
>>(provided? 'setitimer).
> This is what I came up with
> branch wip-itimer-checks
> commit f5b362586d7622c408d4402c7cb496c45ffb56e8
> What do you think?

Looks great.  Thank you, please push!

Andy