Re: puzzling Gnome-ERROR

[eeyem u washington edu]

On Fri, Dec 24, 1999 at 04:47:22AM -0500, jgotts@old.tqstats.com wrote:
> This is the exact error you will get from several applications if you try to
> use the stock GNOME on a libc 5.x system.  (Apparently there is a bug
> somewhere in autoconf/automake/libtool or the GNOME macros).

> In message <yb7ln6lty6f.fsf@icon.labs.redhat.com>, Havoc Pennington writes:
> >eeyem@u.washington.edu writes: 
> >> Problem:
> >> Gnome-ERROR **: Could not create per-user Gnome directory
> >> </home/disco/.gnome> - aborting

(In case you've forgotten the original problem:
The above error would happen only if the directory was already there.
If he rm'd his .gnome directory, the program would run the first time,
but fail on consecutive runs.)

> >Yuck, this is your basic shitty error messages. Pet peeve of mine. I'm
> >fixing this one in gnome-libs 2.0. The message shouldn't use g_error()
> >anyway since a coredump is useless.
> >
> >For now, ask the user to try running your program with "strace" - it
> >will show the error that caused mkdir to fail. That should give a
> >clue.

Uwe Koloska(koloska@rcs.urz.tu-dresden.de) wrote me the following:
    I have had the same problem and by debugging it find out that mkdir()
    answers with 1 what is in contradiction with the manpage (0 on success and
    -1 on error) after I upgraded my system to glibc2 this problem disappears.
    I think it is someway related with threads ...

    a quick workaround if you compile the apps for yourself: just before
    gnome_init() call "gnome_do_not_create_directories = 1;"

It looks like Havoc has already fixed this (or tried to) in 1.36:
    if (mkdir(gnome_user_dir, 0700) < 0) { 
	if (errno != EEXIST) {
            /* error message */

but, If you examine the old code:
    if( 
        mkdir(gnome_user_dir, 0700) /* This is per-user info
                                       - no need for others to see it */
        && errno != EEXIST)
It suggests that even if mkdir() was returning a 1 to indicate failure,
errno should've been EEXIST, and the error message shouldn't have been printed.

This leads me to believe that errno wasn't being set properly,
which is what jgotts@old.tqstats.com was suggesting at the top of this message:
the gnome libs needed to be compiled with -D_REENTRANT.

Either way, Havoc's code works around the problem, I guess.

My conclusion: threads and GNOME are a lot of trouble.  I'll be using fork()
in all future projects.

-- 
Evan Martin - eeyem@u.washington.edu - http://students.washington.edu/eeyem