The "callargs" variable is modified within PG_TRY and then referenced
within PG_CATCH, which is exactly the coding pattern we've now found
to be unsafe. Marking "callargs" volatile would be problematic because
it is passed by reference to some Tcl functions, so fix the problem
by not modifying it within PG_TRY. We can just postpone the free()
till we exit the PG_TRY construct, as is already done elsewhere in this
same file.
Also, fix failure to free(callargs) when exiting on too-many-arguments
error. This is only a minor memory leak, but a leak nonetheless.
In passing, remove some unnecessary "volatile" markings in the same
function. Those doubtless are there because gcc 2.95.3 whinged about
them, but we now know that its algorithm for complaining is many bricks
shy of a load.
This is certainly a live bug with compilers that optimize similarly
to current gcc, so back-patch to all active branches.
int j;
Tcl_HashEntry *hashent;
pltcl_query_desc *qdesc;
- const char *volatile nulls = NULL;
- CONST84 char *volatile arrayname = NULL;
- CONST84 char *volatile loop_body = NULL;
+ const char *nulls = NULL;
+ CONST84 char *arrayname = NULL;
+ CONST84 char *loop_body = NULL;
int count = 0;
int callnargs;
CONST84 char **callargs = NULL;
if (i != argc)
{
Tcl_SetResult(interp, usage, TCL_STATIC);
+ if (callargs)
+ ckfree((char *) callargs);
return TCL_ERROR;
}
}
}
- if (callargs)
- ckfree((char *) callargs);
- callargs = NULL;
-
/************************************************************
* Execute the plan
************************************************************/
}
PG_END_TRY();
+ if (callargs)
+ ckfree((char *) callargs);
+
return my_rc;
}