Re: [PATCH v2 2/2] merge with Scheme regexp; fix bugs

From: Scott L. Burson <scott@sympoiesis.com>
Date: Jan 14, 2026, 06:18 UTC
Message-ID: <CAF5LJ4DrKkJpCfOkkEsYvDH7qF1Bx-v75GryxUbr6UgmJq05cw@mail.gmail.com>
In-Reply-To: <7c642644-09a5-4a50-931b-a630d459932d@kdbg.org>

On Tue, Dec 2, 2025 at 2:27 AM Johannes Sixt <j6t@kdbg.org> wrote:

Show 23 quoted lines

>
> Am 27.11.25 um 17:09 schrieb Johannes Sixt:
> > Am 27.11.25 um 03:38 schrieb Scott L. Burson via GitGitGadget:
> >>       /*
> >> -      * R7RS valid identifiers include any sequence enclosed
> >> -      * within vertical lines having no backslashes
> >> +      * The union of R7RS and Common Lisp symbol syntax: allows arbitrary
> >> +      * strings between vertical bars, including escaped backslashes and
> >> +      * vertical bars.
> >>        */
> >> -     "\\|([^\\\\]*)\\|"
> >> +     "\\|([^\\\\]|\\\\\\\\|\\\\\\|)*\\|"
> >
> > Without the C quoting we have
> >
> >       \|([^\\]|\\\\|\\\|)*\|
> >
> > So, this is everthing from | up to the next |, except that \| does not
> > stop scanning and \\ is also considered so that \\| is not regarded as \
> > followed by \|. Good.
>
> Actually, no. Regular expressions don't choose the first match if a
> different alternative gives a longer match in total.

Ah, good catch.

I noticed another bug. At least in Common Lisp, and I expect also in Scheme, while backslash and vertical bar are the only characters that must be escaped to be included, in fact any character _may_ be escaped. (This came to my attention when Emacs Paredit escaped a double-quote for me, between vertical bars, unnecessarily. Of course, in a string, double-quote would need to be escaped.)

So the correct regexp, with both of these bugs fixed, is

    "\\|([^|\\\\]|\\\\.)*\\|"

Or, without the C quoting:

    \|([^|\\]|\\.)*\|

For example, for

Show 14 quoted lines

> the change
>
> -  (let ((|one two| |three four|)))
> +  (let ((|1 two| |three four|)))
>
> we get to see the word diff
>
>   (let (([-|one two| |three four|-]{+|1 two| |three four|+})))
>
> but the desired result is
>
>   (let (([-|one two|-]{+|1 two|+} |three four|)))
>
> I think it helps to include | in the bracket expression.

Done.

> It may be worth its own patch that also adds a test in t/t4034/scheme/.

I updated the existing test to check for both of these bugs (and verified that it did so by reintroducing them). It's all in this one patch.

The branch now contains two commits, yours and mine. Do I just do /submit at this point, or do I need to submit them separately?