Re: [PATCH] Fix blocking read timeouts at a small probability

[Koichi Murase]

2021年2月10日(水) 23:40 Chet Ramey <chet.ramey@case.edu>:
> On 2/10/21 10:21 AM, Chet Ramey wrote:
> > Thanks for the analysis and patch. I'll take a look. At first glance, I
> > would like to find a simpler way to do it.

Thank you. Yes, if there is a simpler but still robust and reliable
way to do it, that would be nice. Anyway, I want to see an authentic
fix but not a partial fix (that looks to work in "most" cases). With a
partial fix, there could be some special use cases where the failure
rate becomes non-negligible, such as the test case for this time
against the previous partial fix.

> I'd also like to know how this is going to work with read -e and readline,
> since those don't seem to know anything about the timer.

OK, that is a good point. I completely forgot about `read -e'. I took
a look at `rl_getc (stream)' (lib/readline/input.c). If one wants to
switch to this approach, the timeout needs to be specified for the
pselect(2) calls in `rl_getc ()'. In that case, I think readline could
provide a proper public interface to accept the setting of the
timeout. Because readline might be linked with different versions of
Bash, the handling of SIGALRM in readline can be left as is, but the
new interface for the timeout can be just added?

--
Koichi