bug#64391: buffer narrowing slowdown regression in emacs 29

[Gregory Heytings]

> > On a second thought, I believe its better to not replace LABEL with 
> > TAG, because that would mean changing that word in many places, 
> > including places in which such a change would make the text less 
> > understandable, e.g. the docstring of narrow-to-region:
>
> LABEL implies a string or a symbol, whereas :label can accept "any 
> object that is not nil".  The description of 'catch' uses TAG.

I know, and I agree that in principle TAG would be better.  However all 
the code and documentation was written with the word "label", and although 
using a non-symbol is possible, LABEL is intended to be a symbol, not an 
arbitrary Lisp object.  So perhaps the best thing to do would be this:

diff --git a/lisp/subr.el b/lisp/subr.el
index 0b397b7bebf..c2110cb4bb2 100644
--- a/lisp/subr.el
+++ b/lisp/subr.el
@@ -3972,7 +3972,8 @@ with-restriction

 \(fn START END [:label LABEL] BODY)"
   (declare (indent 2) (debug t))
-  (if (eq (car rest) :label)
+  (if (and (eq (car rest) :label)
+           (symbolp (cadr rest)))
       `(internal--with-restriction ,start ,end (lambda () ,@(cddr rest))
                                  ,(cadr rest))
     `(internal--with-restriction ,start ,end (lambda () ,@rest))))

> > --- a/doc/lispref/positions.texi
> > +++ b/doc/lispref/positions.texi
> > @@ -1169,9 +1169,10 @@ Narrowing
> >
> >   @cindex labeled narrowing
> >   @cindex labeled restriction
> > -When the optional argument @var{label}, a symbol, is present, the
> > -narrowing is @dfn{labeled}.  A labeled narrowing differs from a
> > -non-labeled one in several ways:
> > +When the optional argument @var{label}, which is evaluated to get the
> > +label to use and must not be @code{nil},
>
> What "must not be nil": the label or the result of its evaluation?

The result of the evaluation of the label argument.  I don't know how to 
make this clearer.