[oi-dev] "Fixed Enlightenment" or "Beware of epoll"

[Vincent Torri]

On Wed, Apr 12, 2017 at 8:49 AM, Alexander Pyhalov <alp at rsu.ru> wrote:
> On 12.04.2017 08:32, Aurélien Larcher wrote:
>>
>> Hi,
>> yesterday I figured out why EFL stopped working suddenly several (too
>> many)
>> months ago: <sys/epoll.h> was added and one source started using silently
>> epoll in the main loop.
>>
>> But epoll in illumos does not implement the same behaviour as in Linux:
>> https://us-east.manta.joyent.com/smartosman/public/man5/epoll.5.html
>> and in this particular case EFL relies on a peculiarity of Linux epoll: on
>> illumos this produces a deadlock.
>>
>> The "fix" was to avoid using automatically epoll if <sys/epoll.h> is
>> present on the system.
>
>
> I still don't understand why we expose epoll for public if it's just a part
> of internal linux emulation level.
> I'd better hide linux-compatibility interfaces behind some guard...
>
> As for issue with enlightenment, can we commit some fix to avoid using it to
> enlightenment library for now (without updating it)
> to avoid shipping package which is known to be broken?

I can provide a fix in the build system of the EFL for the next
release, but indeed, it would be better to hide the linux emulation
layer. I can do that today

Vincent Torri

> ---
> System Administrator of Southern Federal University Computer Center
>
>
>
> _______________________________________________
> oi-dev mailing list
> oi-dev at openindiana.org
> https://openindiana.org/mailman/listinfo/oi-dev