100% CPU usage when sxiv is compiled with AUTORELOAD=nop and run in background in a subshell

[iamleot]

Running v24 compiled with AUTORELOAD=nop in background in a subshell leads
to 100% CPU usage due a lot of select(2) calls, e.g.:

$ ( sxiv 16x16.png & )

Possible interesting lines from ktruss(1):

$ ktruss -id sh -c '( sxiv 16x16.png & )'
[...]
 10081      1 sxiv     __gettimeofday50(0x7f7fff79fce0, 0) = 0
 10081      1 sxiv     recvmsg(0x3, 0x7f7fff79fba0, 0) Err#35 EAGAIN
 10081      1 sxiv     __select50(0x4, 0x7f7fff79fd30, 0, 0, 0x7f7fff79fd20) = 1
 10081      1 sxiv     __gettimeofday50(0x7f7fff79fce0, 0) = 0
 10081      1 sxiv     recvmsg(0x3, 0x7f7fff79fba0, 0) Err#35 EAGAIN
 10081      1 sxiv     __select50(0x4, 0x7f7fff79fd30, 0, 0, 0x7f7fff79fd20) = 1
 10081      1 sxiv     __gettimeofday50(0x7f7fff79fce0, 0) = 0
 10081      1 sxiv     recvmsg(0x3, 0x7f7fff79fba0, 0) Err#35 EAGAIN
 10081      1 sxiv     __select50(0x4, 0x7f7fff79fd30, 0, 0, 0x7f7fff79fd20) = 1
 10081      1 sxiv     __gettimeofday50(0x7f7fff79fce0, 0) = 0
 10081      1 sxiv     recvmsg(0x3, 0x7f7fff79fba0, 0) Err#35 EAGAIN
 10081      1 sxiv     __select50(0x4, 0x7f7fff79fd30, 0, 0, 0x7f7fff79fd20) = 1
 10081      1 sxiv     __gettimeofday50(0x7f7fff79fce0, 0) = 0
 10081      1 sxiv     recvmsg(0x3, 0x7f7fff79fba0, 0) Err#35 EAGAIN
 10081      1 sxiv     __select50(0x4, 0x7f7fff79fd30, 0, 0, 0x7f7fff79fd20) = 1
 10081      1 sxiv     __gettimeofday50(0x7f7fff79fce0, 0) = 0
[...]

This only happens when sxiv is run in the background and in a subshell, there
aren't any problem running sxiv via:

$ sxiv 16x16.png
$ sxiv 16x16.png &
$ ( sxiv 16x16.png )

Unfortunately I have not figured out why that happens but the following patch
(tested with v24, but should cleanly apply also on the master branch, if
not please let me know and I'll adjust it!):

--- autoreload_nop.c.orig	2017-10-27 15:25:56.000000000 +0000
+++ autoreload_nop.c
@@ -20,7 +20,7 @@
 
 void arl_init(arl_t *arl)
 {
-	(void) arl;
+	arl->fd = -1;
 }
 
 void arl_cleanup(arl_t *arl)

Please note that I have only noticed that on NetBSD, not sure if also other
platforms are affected.

Do not hesitate to contact me if more information, further debugging, etc. is
needed!

Thanks for the attention!

[sdx23]

Thank you for your report. This bug is probably my fault, sorry for that.

Please note that I have only noticed that on NetBSD, not sure if also other platforms are affected.

Reproduced the behaviour as described with debian jessie, on master. Indeed, check_timeouts() and arl_handle() will be called repeatedly if sxiv is called non-interactive, i.e. the same will happen for echo "" | ./sxiv timage.png.

The patch suggested is exactly what is needed. Without, arl->fd is 0, thus resembling stdin. In both suggested non-interactive modes there is data available to be read from stdin and thus FD_ISSET true. Same happens when you start sxiv normally and then press e.g. enter in the terminal window. In the originally reported setting stdin is actually /dev/null (see ls -alh /proc/self/fd/0) which provides EOF to be read.

[xyb3rt]

Thanks for reporting and fixing this bug.

[iamleot]

Hello Bert, Bert Münnich writes:

Thanks for reporting and fixing this bug.

Thank you very much for the quick merge and for sxiv! Thanks again to @sdx23 for helping in the analysis of the problem! Keep up the good work!