2017-10-22 17:25 GMT+02:00 Samuel Thibault <samuel.thiba...@gnu.org>:
>>
>> Unfortunately, lwip doesn't provide a lwip_ppoll() function.
>
> Ah, I saw it in lwipv6, I thought it was generally available.
>
> That being said, it'd be better to make the code already use struct
> timespec, converting to coarser resolution only where is has to be done
> (the lwip_poll call).
That's already done. In lwip_io_select_common() the timeout is rounded
just before calling lwip_poll().
It shouldn't be difficult to write a lwip_ppoll() function, I'll try
it some day.
>>
>> 1.- Why inhibit pfinet_{cntl,protid}_portclasses[0] and not
>> pfinet_{cntl,protid}_portclasses[1]. Wouldn't that lead to stop RPCs
>> to /servers/socket/2 but still allow RPCs to /servers/socket/26?
>
> That's probably just a bug, yes.
OK
>> 2.- Why inhibit socketport_class and not addrpot_class?
>
> I guess that's because it's not "too bad" for the addrport RPCs to
> break, and they are usually used very transiently while a socket is
> still active. But for safety it is probably better to inhibit it.
> I guess the cases where there wouldn't be any socket left but some
> addrport left is rare enough that we don't care.
OK
>> 3.- Why use ports_inhibit_class_rpcs() for inhibiting all classes, but
>> then use ports_enable_class() just for socketport_class and
>> ports_resume_class_rpcs() for all other classes?
>
> I have to say I don't know :)
> I don't know if that was really tested actually. From reading the source
> it seems it should really be replaced by ports_resume_class_rpcs.
OK
>> I have kind of the same question about S_startup_dosync()[2]. Why
>> destroy only socketport_class ports and not addrport_class ones?
>
> I guess it's the same answer as above: we can as well destroy them too
> for nicer handling.
OK Again.
I think I've done all your requested changes, but the perl testsuite
indeed found some bugs. So I'll send the new patch when I have them
fixed.
Regards.
