From: Theo Buehler <tb@theobuehler.org>
Subject: Re: mg: "support" for exuberant/universal ctags tags files
To: Omar Polo <op@omarpolo.com>
Cc: tech@openbsd.org
Date: Fri, 14 Jun 2024 09:35:59 +0200

On Tue, Jun 04, 2024 at 08:59:56AM +0200, Omar Polo wrote:
> 36-week bump :)
> 
> I noticed I still had this in my tree.  Actually I am in doubt whether
> to actually keep the first hunk or not, I could as well drop it, but the
> others are nice for compatibility.
> 
> While it could be interesting to support uctags annotations, I don't
> have plans to do so.
> 
> Thoughts?

I think you should just commit this.

ok tb

> 
> On 2023/09/27 18:23:52 +0200, Omar Polo <op@omarpolo.com> wrote:
> > TL;DR "support" because this is not about supporting the fancy new
> > stuff, but just not breaking on a universal-ctags generated tag file.
> > I'd just like to tell mg enough of the file format to ignore the
> > extensions over plain tag files.
> > 
> > The diff is mostly from troglobit' mg:
> > https://github.com/troglobit/mg/commit/64cd07ecf5c521d50966eca8ba04026e7c9f3be8
> > 
> > The documentation about the universal-ctags extension over tags files
> > is documented in utags(5) after you install the universal-ctags
> > package.  Tags file entries follows this scheme:
> > 
> > 	{tagname}<tab>{tagfile}<tab>{tagaddress}
> > 
> > where tagaddress is an ex command (could be anything, I haven't
> > checked how vi(1) actually implements it, but mg assumes it's a search
> > and just does that.  Works most of the times).
> > 
> > To store extra metadata about the tag itself, the address is put a
> > comment in the tagaddress.  For example, here's how the main() entry
> > would look like.  Not the type information after ;"
> > 
> > % uctags *.c
> > % g ^main tags
> > tags:526: main  main.c  /^main(int argc, char **argv)$/;"       f       typeref:typename:int
> > 
> > Similarly, universal-ctags introduced support for some meta-tag using
> > some dummy entries under the !_TAG_ "namespace"
> > (e.g. !_TAG_FILE_FORMAT, !_TAG_FILE_SORTED, ...)
> > 
> > So, the idea is to just ignore tags starting with !_TAG_ and ignore
> > comments.  As simple as that :-)
> 
> Index: tags.c
> ===================================================================
> RCS file: /home/cvs/src/usr.bin/mg/tags.c,v
> diff -u -p -r1.27 tags.c
> --- tags.c	29 Mar 2023 19:09:04 -0000	1.27
> +++ tags.c	27 Sep 2023 16:21:42 -0000
> @@ -281,6 +281,10 @@ loadtags(const char *fn)
>  	}
>  	while ((l = fparseln(fd, NULL, NULL, "\\\\\0",
>  	    FPARSELN_UNESCCONT | FPARSELN_UNESCREST)) != NULL) {
> +		if (!strncmp(l, "!_TAG_", 6)) {
> +			free(l);
> +			continue;
> +		}
>  		if (addctag(l) == FALSE) {
>  			fclose(fd);
>  			return (FALSE);
> @@ -340,7 +344,7 @@ int
>  addctag(char *s)
>  {
>  	struct ctag *t = NULL;
> -	char *l;
> +	char *l, *c;
>  
>  	if ((t = malloc(sizeof(struct ctag))) == NULL) {
>  		dobeep();
> @@ -357,6 +361,15 @@ addctag(char *s)
>  	*l++ = '\0';
>  	if (*l == '\0')
>  		goto cleanup;
> +
> +	/*
> +	 * Newer universal ctags format abuse vi comments in the
> +	 * pattern to store extra metadata.  Since we don't support it
> +	 * remove it so the pattern is not mangled.
> +	 */
> +	if ((c = strstr(l, ";\"")) != NULL)
> +		*c = '\0';
> +
>  	t->pat = strip(l, strlen(l));
>  	if (RB_INSERT(tagtree, &tags, t) != NULL) {
>  		free(t);
>