Re: GTK-Function with pathnamestring???

["Pawel S. Veselov" <Pawel Veselov Sun COM>]

On Wed, Oct 08, 2003 at 12:53:52AM -0400, John K Luebs wrote:
> > argv[0] is always a pointer to a executable. If it's relative, use getcwd(),
> > to append to get full pathname.
> This is not entirely correct. argv[0] is not any different than any of
> the other argv[] elements. It is simply whatever was passed 
> to the exec system call as argv[0], and it can be totally arbitrary.
> argv[0] is not a 100% reliable method of process image path 
> determination, though it is true that it will work in many cases.

Well, this is true. But I haven't really seen any code which doesn't
set argv[0] to executable path. There is also a lot of code out there which
depends on it's validity.

> Furthermore, in the cases that argv[0] is not arbitrary, it will almost
> never be relative. It will either be absolute, or it will be within
> one of the directories in $PATH. The upshot is that all you can do
> with argv[0] is pass it to exec and hope for the best.

I won't say it is so. How many times people invoke binaries using
relative pathname (  ./test,  ./bin/start ), then argv[0] is relative.

[ skipped ]

> The short answer is that there is no portable (amongst POSIX/UNIX like
> systems) way to get a path to your process. On linux, there is a highly 
> non-portable solution, that is to use /proc/<pid>/exe. This is a symlink
> to a path that leads to the inode of the process's image. Almost every
> UNIX like OS has some non portable facility to get this information.

Probably, good idea may be, to have conditional defines for system which
have a solution (Linux, Windows I guess), and use argv[0] for other cases.
That increases the chance.