emacs-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH] Really prevent quitting in sensitive contexts


From: Daniel Colascione
Subject: Re: [PATCH] Really prevent quitting in sensitive contexts
Date: Wed, 11 Mar 2020 19:32:18 -0700
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.4.1

On 3/11/20 7:20 PM, Stefan Monnier wrote:
with-local-quit is actually documented to allow this sort of
foreground quitting,

AFAIK it's not just documented to allow that, but it's designed
specifically for that.

Right, but I can imagine it being useful more generally in some code that wants to be atomic against quit, but with exceptions. I don't know of any such code though.

but since the interactive C-g race is unavoidable, we should deprecate
use of with-local-quit for this purpose.

I tend to agree.

This change adds a new inhibit-quit-override flag.

Sounds like an arm's race: then we'll have to add
a with-local-quit-override-override, etc...

I considered making with-local-quit-override inaccessible from Lisp.

+behaves as if 'inhibit-quit' were t.  This change prevents
+with-local-quit from accidentally allowing quits inside process
+filters, redisplay callbacks, and other special contexts.

I'm not really familiar with those accidental cases.
Maybe they can be fixed more directly by not using with-local-quit?

with-local-quit has been there for ten years. Can we just delete it?


Are there bug#NNN where these are discussed?

No. The motiviation is Tramp inflooping after a quit because it uses with-local-quit inside tramp-accept-process-output, which causes tramp-wait-for-regexp to retry infinitely and quickly.

I found *that* behavior trying to figure out why M-x compile over Tramp was hanging not only Emacs, but also the SSH connection more generally. I *think* it has something to do with compilation-filter calling file-truename, which invokes Tramp's handler, which wants to talk over the same SSH connection that's spewing compilation messages, deadlocking something somewhere.

We probably should make compile not filename lookup directly in the process filter.



reply via email to

[Prev in Thread] Current Thread [Next in Thread]