Re: some accept-process-output races fixed; Tramp FIXMEs

[Michael Albinus]

Paul Eggert <address@hidden> writes:

> 0.1 is not merely suboptimal (i.e., slower than it should be), it's
> also incorrect (as it can sometimes lose data due to races). Even if
> we must use 0.1 in older Emacs to work around more-serious bugs there,
> we should use nil in Emacs master.

I've adapted all calls of accept-process-output to use a timeout of
either nil or 0. Running Tramp's test suite does not show relevant
performance differences for most of the backends, except smb. Here we
have a performance boost, (411.009675 sec before, 54.109701 sec after
the change).

The change works on all Emacsen 24-27, tested on GNU/Linux, with the
different backends. So I've taken the risk to apply it w/o compatibility
code. Let's see what people report.

I plan to test it also on OpenBSD and MS Windows, once I find respective
test machines. But this will happen for Emacs 27 only, I believe.

Best regards, Michael.