Should 'signal' sometimes return?

Should 'signal' sometimes return?

[Paul Eggert]

Bug#24047 proposes changing Elisp's 'signal' function to not return. 
Although this has been the documented behavior for decades, apparently 
there has been a back door in 'signal' for nearly as long, which causes 
'signal' to return sometimes under control of the Elisp debugger. The 
proposal is to remove this backdoor. If this proposed change to Emacs 
would cause problems for you, please send email to 
24047@debbugs.gnu.org. Thanks.

http://bugs.gnu.org/24047

Re: Should 'signal' sometimes return?

[Stefan Monnier]

> Bug#24047 proposes changing Elisp's 'signal' function to not
> return. Although this has been the documented behavior for decades,
> apparently there has been a back door in 'signal' for nearly as long, which
> causes 'signal' to return sometimes under control of the Elisp debugger. The
> proposal is to remove this backdoor. If this proposed change to Emacs would
> cause problems for you, please send email to 24047@debbugs.gnu.org. Thanks.

There's indeed one back-door, which is that Fsignal can return from
a `quit' signal.  This is part of the support for debug-on-quit and
I wouldn't want to lose this.

I agree that the current situation is unsatisfactory.


        Stefan

Re: Should 'signal' sometimes return?

[Paul Eggert]

On 07/21/2016 05:43 PM, Stefan Monnier wrote:
> There's indeed one back-door, which is that Fsignal can return from
> a `quit' signal.  This is part of the support for debug-on-quit and
> I wouldn't want to lose this.
>
> I agree that the current situation is unsatisfactory.

As I understand it, the debug-on-quit support is intended to catch the 
user typing C-g, as opposed to altering code that calls 'signal' so that 
it returns. If so, how about keeping the current approach for C-g, and 
merely changing 'signal' itself so that it does not return? I wrote a 
new patch to do that, and will send it to Bug#24047.

Re: Should 'signal' sometimes return?

[Davis Herring]

> As I understand it, the debug-on-quit support is intended to catch the
> user typing C-g, as opposed to altering code that calls 'signal' so that
> it returns. If so, how about keeping the current approach for C-g, and
> merely changing 'signal' itself so that it does not return? I wrote a
> new patch to do that, and will send it to Bug#24047.

Just a thought: it could be useful to retain the (signal 'quit nil) 
behavior as an analog to (edebug), although I would understand keeping 
just the latter.

Davis

-- 
This product is sold by volume, not by mass.  If it appears too dense or 
too sparse, it is because mass-energy conversion has occurred during 
shipping.