bug#73771: 30.0.91; etags generates broken TAGS file for multi-line regex match

[Eli Zaretskii <eliz@gnu.org>]

> Cc: 73771@debbugs.gnu.org, morgan@ice9.digital
> Date: Sun, 13 Oct 2024 15:58:44 +0300
> From: Eli Zaretskii <eliz@gnu.org>
> 
> > From: Francesco Potortì <pot@gnu.org>
> > Date: Sun, 13 Oct 2024 12:48:40 +0200
> > Cc: 73771@debbugs.gnu.org,
> > 	Morgan Willcock <morgan@ice9.digital>
> > 
> > >This is because etags always records one extra character with the
> > >regexp match, which is harmless, unless that extra character is a
> > >newline.  The patch below fixes it:
> > >
> > >diff --git a/lib-src/etags.c b/lib-src/etags.c
> > >index a822a82..848d8ea 100644
> > >--- a/lib-src/etags.c
> > >+++ b/lib-src/etags.c
> > >@@ -7420,7 +7420,7 @@ regex_tag_multiline (void)
> > > 
> > > 	      /* Force explicit tag name, if a name is there. */
> > > 	      pfnote (name, true, buffer + linecharno,
> > >-		      charno - linecharno + 1, lineno, linecharno);
> > >+		      charno - linecharno, lineno, linecharno);
> > > 
> > > 	      if (debug)
> > > 		fprintf (stderr, "%s on %s:%"PRIdMAX": %s\n",
> > >
> > >Francesco, why does the code add one more character there?
> > 
> > While I can't remember the reason, I am sure it was not done by chance.
> > 
> > Sure, I may have been wrong at that time, or maybe that reason is now obsolete.  To check, one should run the etags regression test and check the results.
> 
> I already did.  The only differences I see are in that 1 extra
> character, which is now no longer written to TAGS in the tests which
> use multi-line regexps.
> 
> Note that the change is only in this single call to pfnote, from the
> code that is part of generating tags from multi-line regexp matches.
> 
> > Off the top of my head, it may have to do with implicitly named tags that are initial substring of a different tag, but this may be a fake reconstruction of my mind,
> 
> The tags we produce from regexp matches are norally explicitly named,
> AFAICT:
> 
> 	      /* Match occurred.  Construct a tag. */
> 	      while (charno < rp->regs.end[0])
> 		if (buffer[charno++] == '\n')
> 		  lineno++, linecharno = charno;
> 	      name = rp->name;
> 	      if (name[0] == '\0')
> 		name = NULL;
> 	      else /* make a named tag */
> 		name = substitute (buffer, rp->name, &rp->regs);
> 
> 	      /* Force explicit tag name, if a name is there. */
> 	      pfnote (name, true, buffer + linecharno,
> 		      charno - linecharno, lineno, linecharno);
> 
> But if you say that we should add that offset of 1 when 'name' is
> NULL, I'm okay with doing that.

I've decided to install the change on the master branch.  Let's see
what, if anything, it will break.

I'm closing the bug with this message.