Re: unwind-protect within while-no-input

[Daniel Mendler via "Emacs development discussions." <emacs-devel@gnu.org>]

Dmitry Gutov <dmitry@gutov.dev> writes:

> Hi Daniel,
>
> It's good to see you in this thread.
>
> On 12/05/2024 01:08, Daniel Mendler wrote:
>
>>>> Is cape-wrap-noninterruptible from the cape package helpful?
>>> It should help, but it would be odd if any non-trivial completion function would
>>> need to use cape to function inside corfu-mode reliably.
>> cape-wrap-noninterruptible and similarly cape-wrap-silent and
>> cape-wrap-purify are meant as stop gap measures, used to wrap non-robust
>> Capfs, until they are improved. There are multiple Capfs in the wild
>> which regularly print messages or throw unexpected errors. For example,
>> pcomplete-completions-at-point even modified the buffer until Emacs 29,
>> which could be mitigated with cape-wrap-purify.

> What's your recommendation regarding the necessary cleanup inside the
> while-no-input block?

This is a general question, which applies to all situations where
asynchronous exceptions (or user interrupts) occur. One basically has to
protect all critical sections with (let ((throw-on-input nil)
(inhibit-quit t)) ...), which locally disables the exceptions.

Global state manipulations which should not be interrupted, should be
guarded. Most of the Emacs code isn't written in such an exception-safe
way. That's usually okay, as long as the state can still be recovered
somehow after an interrupt by the user. For example if buffer-local
state gets invalid, killing and reopening a buffer will solve the
problem.

However inside Capfs or other code which is very often interrupted,
greater care is necessary. Inside such a critical section, the cleanup
code could be registered somewhere, such that it doesn't get lost. The
cleanup code itself should likely be guarded too.

Daniel