The perils of relying on output streams in C (2011) [pdf]

[cafebabe1991]

https://www.gnu.org/ghm/2011/paris/slides/jim-meyering-goodbye-world.pdf

[ploxiln]

djb's old daemontools suite had an interesting solution to the "how to log failure when you can't write to a log" problem: a persistent child process with some space reserved in its argv, where it can stash some error messages which you can see with "ps" even when disks have failed.

http://cr.yp.to/daemontools/readproctitle.html

[_smxa → ploxiln]

I still use this technique.

[listprocess]

for the corresponding text and video see-- https://www.gnu.org/ghm/2011/paris/#sec-2-1 and https://audio-video.gnu.org/video/ghm2011/Jim_Meyering-Goodb...

[coliveira]

printf and friends was not supposed to be robust, was supposed to be easy to use. C already provides the means to write robust libraries, if that is necessary.

[chris_wot → coliveira]

Can you point me to how to do this?

[coldnose → chris_wot]

https://linux.die.net/man/2/write

[quietbritishjim → coldnose]

write() is part of Unix, not C.

[renox]

The "loosing the errno" issue seems to me easily avoidable: just save the current value if ferror returns an error.