Bob Rossi
Sat, 17 May 2008 03:57:12 -0700
On Sat, May 17, 2008 at 01:08:59PM +1000, Bojan Smojver wrote:
> On Fri, 2008-05-16 at 22:43 -0400, Bob Rossi wrote:
>
> > I just upgraded to the 1.3 svn snapshot, to test out the new
> > functionality that was recently added. After I ran my unit tests I
> > noticed that the function apr_dbd_init() was not working properly for
> > me. Then I realized it's because this code,
>
> With 1.3.x and above, we build DBD modules as DSOs. Did they get built
> properly?
Huh, I build with --disable-shared --enable-static. Perhaps it doesn't
work in this configuration? I do see,
$ ls apr-util-1.3.x-2008-05-16-cvs/
bin include lib
$ ls apr-util-1.3.x-2008-05-16-cvs/lib
apr-util-1 aprutil.exp libaprutil-1.a libaprutil-1.la pkgconfig
$ ls apr-util-1.3.x-2008-05-16-cvs/lib/apr-util-1/
apr_dbd_pgsql.a apr_dbd_pgsql.la apr_dbd_sqlite3.a
apr_dbd_sqlite3.la
I'm thinking perhaps this doesn't work with --disable-shared. If I set a
breakpoing in
APU_DECLARE(apr_status_t) apr_dbd_get_driver(apr_pool_t *pool,
const char *name, const apr_dbd_driver_t **driver)
I can see that it does this,
apr_snprintf(path, sizeof path, "%s/apr_dbd_%s.so", APU_DSO_LIBDIR, name);
There is a couple of disturbing things here, for starters, the "path"
variable declared on the stack is hard coded to a certain size,
char path[80];
my installation of apr-util is this long (in this circumstance)
echo $PWD|wc -c
120
I realy think that variable should be dynamic. ie. the length of
APU_DSO_LIBDIR +
1 for '/' +
8 for "apr_dbd_" +
name +
3 for .so +
1 for NULL
since that is the longest possible name, that should cover the rest of
the cases.
So, it seems like there is 2 bugs here perhaps? Doesn't work with
--disable-shared and the path variable is to small on the stack.
Thanks,
Bob Rossi