bug#44294: No widen by xref-find-definitions

[Dmitry Gutov]

On 29.10.2020 23:18, Juri Linkov wrote:
> reopen 44294
> stop
>
> > > -         (user-error "Rerun etags: `%s' not found in %s"
> > > -                      pat buffer-file-name)))
> > > +         (if (or (= (point-min) 1) (not widen-automatically))
> > > +              (user-error "Rerun etags: `%s' not found in %s"
> > > +                          pat buffer-file-name)
> > > +            ;; Rerun after removing narrowing
> > > +            (widen)
> > > +            (etags-goto-tag-location tag-info))))
> >
> > By the way... have you tried to use the same method here as in elisp-mode?
> > Meaning, widen unconditionally inside 'save-restriction'.
> >
> > There should be no reason for backends to do it differently. And this way,
> > you don't have to always search twice for a missing tag when inside
> > a narrowing.
>
> It should not widen unnecessarily when the found position is within the
> narrowed region.

Hence the use of save-restriction in elisp-mode which you added.

etags can use the exact same approach.

>  In this regard, xref--goto-char does the right thing:

Yes.

>    (defun xref--goto-char (pos)
>      (cond
>       ((and (<= (point-min) pos) (<= pos (point-max))))
>       (widen-automatically (widen))
>       (t (user-error "Position is outside accessible part of buffer")))
>      (goto-char pos))
>
> It widens only when position is outside accessible part of buffer
> (and widen-automatically is non-nil).

And this code runs whether the backend is elisp, etags, or whatever else.