bug#36967: 27.0.50; Duplicate lines in xref output

[Dmitry Gutov <dgutov@yandex.ru>]

On 06.12.2020 22:31, Juri Linkov wrote:
>> I'm _slightly_ worried about extra garbage if we do seq-group-by twice
>> (with an extra list for every line, even those that don't need it), but
>> that's what benchmarking is for (can do that later).
> 
> I'm worried about extra garbage too, so this patch doesn't cause extra lists.

Thanks! Seems to work well. Especially with this addition:

diff --git a/lisp/progmodes/xref.el b/lisp/progmodes/xref.el
index 9f5fc57142..9ad956d496 100644
--- a/lisp/progmodes/xref.el
+++ b/lisp/progmodes/xref.el
@@ -595,7 +595,8 @@ xref-prev-line

  (defun xref--item-at-point ()
    (save-excursion
-    (back-to-indentation)
+    (when (eolp)
+      (forward-char -1))
      (get-text-property (point) 'xref-item)))

  (defun xref-goto-xref (&optional quit)

> However, there is a problem: when xref--insert-xrefs is called by
> 'M-.' that uses etags, then it signals an error:

Does it happen when two symbols do indeed get defined on the same line? 
Sounds like a rare situation.

>    cl-no-applicable-method xref-file-location-column
> 
> And I don't know how to fix etags to add columns.

Using xref-file-location-column there kind of breaks the abstraction. I 
was thinking more along the lines of text-prop search for the column 
inside the summary string:

(setq column
       (with-temp-buffer
         (insert summary)
         (goto-char (point-min))
         (let ((match
                (text-property-search-forward
                 'face 'xref-match
                 (lambda (target value)
                   (or (eq target value)
                       (and (listp value)
                            (member target value)))))))
           (and match
                (1- (prop-match-beginning match))))))

That's quite a bit longer, though. And it still will work only with some 
kinds of xrefs: I'm not sure I know how to add columns to etags either. 
Not to mention third-party xref backends.

The rest won't get deduplication, but that's probably all right.