uu timeout is less precise than GNU timeout

[sylvestre]

uu timeout sleeps for 100ms intervals at

coreutils/src/uucore/src/lib/features/process.rs

Line 163 in 2319a99

thread::sleep(Duration::from_millis(100));

instead of using timer_create or setitimer,
which are more precise.

timeout (GNU coreutils) 9.7
Packaged by Debian (9.7-3)                                                                                                                                                                                            Copyright (C) 2025 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <https://gnu.org/licenses/gpl.html>.
This is free software: you are free to change and redistribute it.                                                                                                                                                    There is NO WARRANTY, to the extent permitted by law.

Written by Padraig Brady.                                                                                                                                                                                             root@b082d601e194:/tmp# time timeout 0.01 sleep 1
real    0m0.014s
user    0m0.000s
sys     0m0.004s

root@03081849be59:/tmp# timeout --version
timeout (uutils coreutils) 0.6.0
root@03081849be59:/tmp# time timeout 0.01 sleep 1
                                                                                                                                                                                                                      real    0m0.117s
user    0m0.002s
sys     0m0.014s                                                                                                                                                                                                      ```

[oech3]

also #9099

[Alonely0]

Timeout (specifically its backend found in uucore::process) requires a rewrite I've attempted 10 times since last October or November. GNU timeout has many undocumented edge cases due to its various (IMO questionable) design decisions, and we don't even cover half. This is notwithstanding how our actual implementation is a giant FIXME. Maybe I'll get to it again eventually, probably after I'm done with ls and exams.

Editing the message to make clear that, despite the initial tone, that I deeply respect everyone who contributed to uu timeout, GNU timeout, etc. I'm just frustrated with uu timeout to the point of absolute exhaustion, so I apologize if it came through the wrong way.

[MrTroika]

Timeout (specifically its backend found in uucore::process) requires a rewrite I've attempted 10 times since last October or November. GNU timeout is just absurdly full of undocumented edge cases due to its various cuestionable design decisions, and we don't even cover half. This is notwithstanding how our actual implementation is so subpar I don't even have words. Maybe I'll get to it again eventually, probably after I'm done with ls.

Editing the message to make clear that, despite the tone, that I deeply respect everyone who contributed to uu timeout, GNU timeout, etc. I'm just frustrated with uu timeout to the point of absolute exhaustion, so I apologize if it came through the wrong way.

Just maybe long standing code was the way it was for good reason, and your addition, which by your own admission, is both broken and flammable, should be reverted.

[J-eremy]

Theres not really an "undocumented" thing in open source if you know what your looking at, that term is more reserved for closed source things. In fact you can probably git blame every line and hopefully see exactly why it was added/changed.

I think this is more of an eyes bigger than stomach situation. With this mindset every core util will be full of undocumented edge cases, as that is exactly what most of the utilities were based on, edge cases.

If you think about it, this whole "memory safe" thing is a giant undocumented edge case.

I wish you luck, but your trying to reinvent not one, but many really old wheels. Things sometimes are the way they are because they have to be and progress for the sake of progress is never a good thing.

[Alonely0]

Just maybe long standing code was the way it was for good reason, and your addition, which by your own admission, is both broken and flammable, should be reverted.

I am having some trouble understanding what you are trying to say. It's been a long day, so forgive me if I'm misunderstanding something. We have not merged any code of mine to uucore::process, which is the part of timeout that I said needed rewriting, so idk what you'd revert; I may or may not have contributions to uu timeout, but if so, they are orthogonal and fixed specific bugs. The longstanding code we are talking about is a quick-and-dirty solution marked as a FIXME (not verbatim) here, and is the way it is because it's hard to do in general, but especially in Rust because of SIGPIPE shenanigans.

Theres not really an "undocumented" thing in open source if you know what your looking at, that term is more reserved for closed source things. In fact you can probably git blame every line and hopefully see exactly why it was added/changed.

Undocumented means undocumented. You don't get to say it is documented because the source is available with a specific license and some git history; you can make the same argument with closed-source software given enough motivation and tools like strace and gdb. Undocumented means it's not in their manpage or the GNU website wiki, period.

Anyway, GNU timeout's various signal handling (iirc SIGPIPE and some other things) was undocumented, among other things, at the time of writing of this comment (to the best of my ability). We are bug-for-bug; not all bugs are documented, and GNU timeout's behavior when messing with signals stems from their architectural decisions, mostly masking/blocking a bazillion signals and their "completion" function being the signal handler of the others, iirc from my strace reverse-engineering. Take it with a pinch of salt, tho.

If you think about it, this whole "memory safe" thing is a giant undocumented edge case.

It. is. not. It is so not undocumented it is ISO 26262:2018.

I wish you luck, but your trying to reinvent not one, but many really old wheels. Things sometimes are the way they are because they have to be and progress for the sake of progress is never a good thing.

Thank you, I wish you luck too. I hope you can see my point of view, and I really appreciate discussion with someone with fresh eyes (literally and metaphorically).

---

This conversation has gotten off-topic, so unless you wanna discuss specific technical details of {GNU, uu} timeout related to this issue, we can continue it in the uu discord or another communication channel.

[ai-mannamalai]

kind of tragicomic;

[metux]

GNU timeout has many undocumented edge cases due to its various (IMO questionable) design decisions,

Can you please give us a bit more insight which ones exactly ?
thanks.

[Alonely0]

Fixed by #13237

@sylvestre mind closing?