dmcgee@galway ~/projects/pacman-maint (maint)
$ ./src/pacman/pacman -Sy --dbpath /tmp/foo/bar
error: problem setting dbpath '/tmp/foo/bar' (could not find or read directory)

So the library throws an error at least when setting a DB path like this; however we shouldn't ever segfault later on if you continue unhindered. Did you get a backtrace? This could happen in a lot of places in theory so I'm not completely sure where we should be checking it for sanity.

Okay, I had to write a dedicated C program to reproduce this because we wrote a patch to workaround the problem in the package manager. But I can confirm it is reproducible.

The program I wrote was a very simple three-step affair:

#include <alpm.h>

int main ( int argc, char* argv[] )
{
alpm_initialize();
alpm_option_set_dbpath("/non/existant/path");
alpm_trans_init ( 0, NULL, NULL, NULL );
return 0;
}

Basically, init ALPM, set the db-path to something that doesn't exist, and simply try to place a lock like you would when, say, updating sync databases. In the package manager the segfault happened amidst a cast to a transaction flag defined in ALPM while the whole thing was being tested for success through an if(). In other words, the offending line was a PURE call to ALPM and nothing else.

So, that C program above was so I wouldn't regress my code but also to confirm it could happen in a clean C (Not C++, and also so we can be sure it's not in my project.) environment. Now, the stuff you'd like to see, Dan:

(gdb) run
The program being debugged has been started already.
Start it from the beginning? (y or n) y
Starting program: /home/yaro/projects/tmp/segfault
[Thread debugging using libthread_db enabled]

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff78d1541 in strlen () from /lib/libc.so.6
(gdb) backtrace
#0 0x00007ffff78d1541 in strlen () from /lib/libc.so.6
#1 0x00007ffff78d1296 in strdup () from /lib/libc.so.6
#2 0x00007ffff7bd22a5 in alpm_trans_init () from /usr/lib/libalpm.so.6
#3 0x00000000004006eb in main (argc=1, argv=0x7fffffffdd88) at main.c:7

I do have ALPM (And thus, pacman, too.) built with debug symbols, so I hope that's usable.

Is there any further thing you need?

No, this looks absolutely perfect, thank you for such a helpful follow-up. We'll get this fixed in the maint branch more than likely.