Paul TBBle Hampson wrote:
I've just submitted to the relevant Debian bug a patch against 1.1.7's rlm_sqlippool which indirects all the functions called from rlm_sql via lt_dlsym'd pointers. This prevents the symbol (in)visibility issue we are having here due to lack of support for RTLD_GLOBAL.
I think it's worth taking a look at re-arranging the code. There should be an SQL back-end module that does nothing but connect to the DB. Then, a series of SQL "worker" modules can call that module for everything else..
Further, both the 1.1.7 and 2.0.0 rlm_sqlippool modules output "database query error" when something goes wrong, and I think it'd be really handy to have it output the sql_error() result as well, since I just personally learnt the hard way that 1142 is MySQL for "permission error".
Yes. Also, sql_user_name, etc.
By the by, I wonder if sqlcounter has the same linking issue? It doesn't have a Debian bug open, but that might as easily mean no one is using the module to have noticed the issue.
Yes. Alan DeKok.