[systemd-devel] Bug in source_io_register?

[David Strauss]

On Mon, Oct 14, 2013 at 1:25 PM, David Strauss <david at davidstrauss.net> wrote:
> On Mon, Oct 14, 2013 at 12:33 PM, Lennart Poettering
> <lennart at poettering.net> wrote:
>> COuld you rebase please and try to reproduce
>> the issue?
>
> I'm not seeing the issue anymore after doing that, but I may have
> fixed something on my side, too.

Actually, I am still seeing this if I force send() to go EAGAIN/EWOULDBLOCK:

Oct 15 17:41:30 olympian sabridge.sh[30744]: Recording that the buffer
got 16384 more bytes full.
Oct 15 17:41:30 olympian sabridge.sh[30744]: Buffer now has 16384 bytes full.
Oct 15 17:41:30 olympian sabridge.sh[30744]: send(5, ...)=15488
Oct 15 17:41:30 olympian sabridge.sh[30744]: send(5, ...)=-1
Oct 15 17:41:30 olympian sabridge.sh[30744]: send(5, ...) completed
with 896 bytes still buffered.
Oct 15 17:41:30 olympian sabridge.sh[30744]: Is send watching enabled on fd=5? 0
Oct 15 17:41:30 olympian sabridge.sh[30744]: Error -17 enabling send
for fd=5: File exists

The "Is send watching enabled" check is in my local code, but it just
runs sd_event_source_get_enabled() to check the watcher.

I'll do some research into it. I'm 90% sure the library side has
something wrong.