Web lists-archives.com

Re: The timerfd functions slow down emacs




On 2/23/2019 5:11 PM, Corinna Vinschen wrote:
> On Feb 23 21:24, Ken Brown wrote:
>> On 2/23/2019 4:01 PM, Corinna Vinschen wrote:
>>> On Feb 23 20:48, Ken Brown wrote:
>>>> On 2/23/2019 2:15 PM, Corinna Vinschen wrote:
>>>>> Below's the NSSTC I used to test my timerfd implementation (based on
>>>>> another STC to show a problem in POSIX timers).  From what I can tell it
>>>>> works as desired.  If you find a problem, please point it out or send a
>>>>> patch.
>>>>
>>>> Thanks, that saved me a lot of time.  I was in the process of creating a test
>>>> case when your mail arrived.
>>>>
>>>> Your test works fine.  For example:
>>>>
>>>> $ gcc timerfd_test.c -o timerfd_test
>>>>
>>>> $ ./timerfd_test.exe 3 1 10
>>>> 0.000: timer started 52379
>>>> 2.988: 52379 read: 1; total=1
>>>> 3.988: 52379 read: 1; total=2
>>>> 4.989: 52379 read: 1; total=3
>>>> 5.987: 52379 read: 1; total=4
>>>> 7.001: 52379 read: 1; total=5
>>>> 7.987: 52379 read: 1; total=6
>>>> 9.001: 52379 read: 1; total=7
>>>> 9.987: 52379 read: 1; total=8
>>>> 10.987: 52379 read: 1; total=9
>>>> 11.997: 52379 read: 1; total=10
>>>>
>>>> But if I change CLOCK_MONOTONIC by CLOCK_REALTIME in the call to timerfd_create,
>>>> then I get the following:
>>>
>>> Only in timerfd_create?  Not in clock_gettime?  If you don't
>>> do that, you're using a different clock with entirely different
>>> values for the starttime.
>>>
>>> Actually, this testcase started with CLOCK_REALTIME in both calls.
>>> If I revert to that, I get:
>>>
>>> )$ ./timerfd 3 1 10
>>> 0.000: timer started 644
>>> 3.001: 644 read: 1; total=1
>>> 4.001: 644 read: 1; total=2
>>> 5.017: 644 read: 1; total=3
>>> 6.001: 644 read: 1; total=4
>>> 7.001: 644 read: 1; total=5
>>> 8.009: 644 read: 1; total=6
>>> 9.002: 644 read: 1; total=7
>>> 10.001: 644 read: 1; total=8
>>> 11.010: 644 read: 1; total=9
>>> 12.017: 644 read: 1; total=10
>>>
>>> I tested this STC with differnt clocks, but it's important to
>>> use the same cloack for clock_gettime and timerfd_create.
>>
>> Yes, that was careless of me.  I now get the same results as you.  I'll have to
>> go back to the emacs sources and see what else might be different there.
> 
> You're right, nevertheless.  Select on timerfd returns immediately.
> I (hope I) fixed that in git and uploaded new developer snapshots to
> https://cygwin.com/snapshots/
> 
> Please try especially with emacs.

That fixes the emacs problem.

Thanks!

Ken