On Tue, 22 May 2007, Pierangelo Masarati wrote:

If that's an issue, I have no problem in explicitly setting that pointer to NULL.

I got some more dbx time on this, and it looks like that's a workaround for something that shouldn't be happening/needs to be fixed anyway. config_filename is always NULL at dlopen() time, just as it should be.

Calling lock(&libradius_mutex) apparently overwrites config_filename, and explicitly initializing to NULL somehow stops this from happening. I can't see anything wrong with radius.c (as in HEAD), though; if anybody can weigh in on this, it'd be much appreciated. I suppose it could still be a compiler or a libthread bug, but I'm still missing a smoking gun here.