Re: 'while do done' hangs interactive zsh

[Mikael Magnusson <mikachu@xxxxxxxxx>]

On 5/16/21, Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx> wrote:
> On Sun, May 16, 2021 at 7:28 AM Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx>
> wrote:
>>
>> > Even more, if the user enters "while do; done" in an interactive zsh
>> > instance, the busy loop is not interruptible by ^C, ^\ or ^Z; the shell
>> > has to be killed via some external means.
>>
>> I can NOT reproduce this in the latest revision from git, but I can
>> reproduce it in zsh-5.8.
>
> Specifically, the shell is interruptible by ^C.  I haven't attempted
> to bisect where this was introduced; there are no obvious deltas to
> signals.c, loop.c, or parse.c to explain it.

I just updated to latest git and I still can't ^C the "while do;
done". I do have some local patches on top but nothing that
intentionally messes with anything related to this afaik...

> It is impossible to interrupt the parser with SIGINT.  zhandler is
> entered with signal queueing enabled; when signals are allowed again,
> zhandler is called again and we pass through this code:
> 675        if (list_pipe || chline || simple_pline) {
> 676            breaks = loops;
> 677            errflag |= ERRFLAG_INT;
> 678            inerrflush();
> 679            check_cursh_sig(SIGINT);
> 680        }
> 681        lastval = 128 + SIGINT;
> Because we're still in the parser, none of (list_pipe || chline ||
> simple_pline) is true, so we never set breaks or errflag, only
> lastval.  I'm not immediately sure what to do about that; perhaps just
> move the errflag setting outside that test?
>
> The attached makes both "while do" and "do done" into parse errors; I
> didn't want to try messing with the tokenizer, so better solutions are
> welcome.

I use "do done" in production code so you can't remove that. I could
imagine someone using "while do" too, but I don't so I won't complain
about that part :). It is faster to not run the builtin "true" than it
is to run it, and it has been valid syntax all these years.

> I mention tokenizing because in "for do" the string "do" is not a
> token; "while do;" could conceivably mean to execute the command "do"
> as the test condition.


-- 
Mikael Magnusson