gtags and symbolic links

[Petr Uzel-2]

Hi list,

I have some questions and maybe some suggestions about global treatment of
symbolic links.

I know there was some discussion about this [1] some time ago, but in my
opinion some issues still remain.

First of all, I'm thinking about implementing the nofollow option for gtags, as
proposed on [2]. My question is, how exactly should this behave: should that
only omit symbolic links that point to a directory or should that ignore
symbolic links completely (e.g. ./foo.c pointing to ./bar.c) ?

When thinking about it and looking into the code, I don't quite understand the
following part of find_read_traverse()  (libutil/find.c):

522  for (;;) {
523     while (curp->p < curp->end) {
524       char type = *(curp->p);
525       const char *unit = curp->p + 1;
526
527       curp->p += strlen(curp->p) + 1;
528
529       /*
530        * Skip files described in the skip list.
531        */
532       /* makepath() returns unsafe module local area. */
533       strlimcpy(path, makepath(dir, unit, NULL), sizeof(path));
534       if (type == 'd')
535          strcat(path, "/");
536       if (skipthisfile(path))
537          continue;
538       if (type == 'f') {
539         /*
540          * Skip the following:
541          * o directory
542          * o file which does not exist
543          * o dead symbolic link
544          */
545          if (!test("f", path)) {
546             if (test("d", path))
547               warning("'%s' is a directory. (Ignored)", path);
548             else
549               warning("'%s' not found. (Ignored)", path);
550             continue;
551          }
552          /*
553           * GLOBAL cannot treat path which includes blanks.
554           * It will be improved in the future.
555           */
556          if (locatestring(path, " ", MATCH_FIRST)) {
557             warning("'%s' ignored, because it includes blank.", &path[2]);
558             continue;
559          }
<SNIP>

To be specific, what's the purpose of lines 539-551? If I'm not mistaken, the
test at line 545 can never succeed because it is the opposite of the test at
line 538.

Next, is there any benefit of determining the type of a file in getdirs()
function (I mean prefixing the filename with f,d) instead of determining the
type directly in the find_read_traverse() function?

And finally: what actually made me start thinking about this is fact that gtags
can not treat recursive links, like in [3]. This actually makes it difficult to
use gtags on directories created by quilt setup, because quilt creates exactly
this type of directory structure (for a good reason). This could be worked
around by the nofollow option, but IMHO the recursive link detection should be
done also in default mode - what do you think?


Thanks in advance,



[1] http://lists.gnu.org/archive/html/bug-global/2008-01/msg00000.html
[2] http://www.gnu.org/software/global/plans.html
[3] http://lists.gnu.org/archive/html/bug-global/2008-01/msg00001.html



--
Best regards / s pozdravem

Petr Uzel, Packages maintainer
---------------------------------------------------------------------
SUSE LINUX, s.r.o.                          e-mail: puzel@...
Lihovarská 1060/12                          tel: [phone]
190 00 Prague 9                             fax: [phone]
Czech Republic                              http://www.suse.cz


_______________________________________________
Bug-global mailing list
Bug-global@...
http://lists.gnu.org/mailman/listinfo/bug-global

[Shigio YAMAGUCHI-2]

Hi,
> First of all, I'm thinking about implementing the nofollow option for gtags, as
> proposed on [2]. My question is, how exactly should this behave: should that
> only omit symbolic links that point to a directory or should that ignore
> symbolic links completely (e.g. ./foo.c pointing to ./bar.c) ?

I have not thought about details yet.
Is there something proposal?

> When thinking about it and looking into the code, I don't quite understand the
> following part of find_read_traverse()  (libutil/find.c):
...
> To be specific, what's the purpose of lines 539-551? If I'm not mistaken, the
> test at line 545 can never succeed because it is the opposite of the test at
> line 538.

It might be so. But at present, this code is harmless.

> Next, is there any benefit of determining the type of a file in getdirs()
> function (I mean prefixing the filename with f,d) instead of determining =
> the
> type directly in the find_read_traverse() function?

This is due to a historical reason.
In some systems like BSD, dirent structure has a member named 'd_type' which
describes file type. Determining type by it, we can avoid to calling stat(2)
for each files.  But another systems like GNU/Linux doesn't has it.
I wanted to hide the difference in a basic function, getdirs().

By the way, GLOBAL doesn't use the 'd_type' currently due to the following bug:
[http://lists.gnu.org/archive/html/bug-global/2003-08/msg00002.html]

> And finally: what actually made me start thinking about this is fact that gtags
> can not treat recursive links, like in [3]. This actually makes it difficult to
> use gtags on directories created by quilt setup, because quilt creates exactly
> this type of directory structure (for a good reason). This could be worked
> around by the nofollow option, but IMHO the recursive link detection should be
> done also in default mode - what do you think?

What is quilt setup?
What should GLOBAL do?
--
Shigio YAMAGUCHI <shigio@...>
PGP fingerprint: D1CB 0B89 B346 4AB6 5663  C4B6 3CA5 BBB3 57BE DDA3


_______________________________________________
Bug-global mailing list
Bug-global@...
http://lists.gnu.org/mailman/listinfo/bug-global

[Petr Uzel-2]

Hi again,

On Thu, May 14, 2009 at 07:44:50PM +0900, Shigio YAMAGUCHI wrote:
> Hi,
> > First of all, I'm thinking about implementing the nofollow option for gtags, as
> > proposed on [2]. My question is, how exactly should this behave: should that
> > only omit symbolic links that point to a directory or should that ignore
> > symbolic links completely (e.g. ./foo.c pointing to ./bar.c) ?
>
> I have not thought about details yet.
> Is there something proposal?

My proposal is to introduce new 'nofollow' option for gtags that would
make gtags not to descend to symbolic links pointing to directories.

>
> > When thinking about it and looking into the code, I don't quite understand the
> > following part of find_read_traverse()  (libutil/find.c):
> ...
> > To be specific, what's the purpose of lines 539-551? If I'm not mistaken, the
> > test at line 545 can never succeed because it is the opposite of the test at
> > line 538.
>
> It might be so. But at present, this code is harmless.

OK, that's what I've thought.

>
> > Next, is there any benefit of determining the type of a file in getdirs()
> > function (I mean prefixing the filename with f,d) instead of determining =
> > the
> > type directly in the find_read_traverse() function?
>
> This is due to a historical reason.
> In some systems like BSD, dirent structure has a member named 'd_type' which
> describes file type. Determining type by it, we can avoid to calling stat(2)
> for each files.  But another systems like GNU/Linux doesn't has it.
> I wanted to hide the difference in a basic function, getdirs().
>
> By the way, GLOBAL doesn't use the 'd_type' currently due to the following bug:
> [http://lists.gnu.org/archive/html/bug-global/2003-08/msg00002.html]

Thanks for clarifiction. Manpage for linux readdir states that the
d_type isn't supported by all filesystems, so it IMHO means that you
can't rely on that (IOW it's not a bug but a feature). Is the
performance gain of using d_type on BSD's worth the trouble with
maintaining getdirs() ?

>
> > And finally: what actually made me start thinking about this is fact that gtags
> > can not treat recursive links, like in [3]. This actually makes it difficult to
> > use gtags on directories created by quilt setup, because quilt creates exactly
> > this type of directory structure (for a good reason). This could be worked
> > around by the nofollow option, but IMHO the recursive link detection should be
> > done also in default mode - what do you think?
>
> What is quilt setup?

http://linux.die.net/man/1/quilt

quilt setup is a 'quilt subcommand' intended to help maintaining
RPM's. Given a specfile and a tarball with sources, it extracts the
tarball, cds to this directory and creates symbolic link pointing to
'..' -> so running gtags in the directory with sources fails.

> What should GLOBAL do?

It should detect and ignore recursive links (like find,ctags,cscope,.... do).


Thanks,

--
Best regards / s pozdravem

Petr Uzel, Packages maintainer
---------------------------------------------------------------------
SUSE LINUX, s.r.o.                          e-mail: puzel@...
Lihovarská 1060/12                          tel: [phone]
190 00 Prague 9                             fax: [phone]
Czech Republic                              http://www.suse.cz


_______________________________________________
Bug-global mailing list
Bug-global@...
http://lists.gnu.org/mailman/listinfo/bug-global

[Shigio YAMAGUCHI-2]

> My proposal is to introduce new 'nofollow' option for gtags that would
> make gtags not to descend to symbolic links pointing to directories.
...
> It should detect and ignore recursive links (like find,ctags,cscope,.... do).

What is the 'recursive links'?
(I understand that it is a symbolic link to a directory.)

Does it mean a symbolic link which points to '.' or one of the parent directories?

For example, making a symbolic link '/home/user/src/lib/recursive',

        Current directory: /home/user/src/lib
        parent directories: /, /home, /home/user, /home/user/src

Otherwise, the "ignoring recursive links" means simply not to trace same
directory twice?

--
Shigio YAMAGUCHI <shigio@...>
PGP fingerprint: D1CB 0B89 B346 4AB6 5663  C4B6 3CA5 BBB3 57BE DDA3


_______________________________________________
Bug-global mailing list
Bug-global@...
http://lists.gnu.org/mailman/listinfo/bug-global

[Petr Uzel-2]

Hi Shigio,

On Fri, May 15, 2009 at 07:43:05AM +0900, Shigio YAMAGUCHI wrote:
> > My proposal is to introduce new 'nofollow' option for gtags that would
> > make gtags not to descend to symbolic links pointing to directories.
> ...
> > It should detect and ignore recursive links (like find,ctags,cscope,.... do).
>
> What is the 'recursive links'?
> (I understand that it is a symbolic link to a directory.)
>
> Does it mean a symbolic link which points to '.' or one of the parent directories?

By recursive link, I mean a link (pointing to a directory) which creates a loop in
directory structure, like this:

ls -lR dir/
dir/:
total 0
lrwxrwxrwx 1 puzel users 2 May 15 09:32 link -> ..
-rw-r--r-- 1 puzel users 0 May 15 09:32 main.c

Try running 'strace gtags' in 'dir' to see what happens.

>
> For example, making a symbolic link '/home/user/src/lib/recursive',
>
> Current directory: /home/user/src/lib
> parent directories: /, /home, /home/user, /home/user/src

Sorry, I'm afraid I don't understand this example.

> Otherwise, the "ignoring recursive links" means simply not to trace same
> directory twice?

With proposed nofollow option, which I've describe in previous post,
the recursive links should not make any troubles at all.

When following symbolic links to directories (now default), gtags
should detect the loop somehow. If I'm not mistaken, ctags does this
by checking that a directory we are going into isn't parent of the
current directory.


--
Best regards / s pozdravem

Petr Uzel, Packages maintainer
---------------------------------------------------------------------
SUSE LINUX, s.r.o.                          e-mail: puzel@...
Lihovarská 1060/12                          tel: [phone]
190 00 Prague 9                             fax: [phone]
Czech Republic                              http://www.suse.cz


_______________________________________________
Bug-global mailing list
Bug-global@...
http://lists.gnu.org/mailman/listinfo/bug-global

[Shigio YAMAGUCHI-2]

Hi Petr
> By recursive link, I mean a link (pointing to a directory) which creates =
> a loop in
> directory structure, like this:
> ls -lR dir/
> dir/:
> total 0
> lrwxrwxrwx 1 puzel users 2 May 15 09:32 link -> ..
> -rw-r--r-- 1 puzel users 0 May 15 09:32 main.c

I have roughly understood it.

> When following symbolic links to directories (now default), gtags
> should detect the loop somehow.

I agree with you.
I will put this into the TODO list.
I think that this function should be active unless the --nofollow option
is specified.
--
Shigio YAMAGUCHI <shigio@...>
PGP fingerprint: D1CB 0B89 B346 4AB6 5663  C4B6 3CA5 BBB3 57BE DDA3


_______________________________________________
Bug-global mailing list
Bug-global@...
http://lists.gnu.org/mailman/listinfo/bug-global