A quick fix for CORE-2257: Internal consistency check while altering …

…dependent procedures.

To be reviewed some day, along with the whole ALTER mechanism.

src/jrd/met.epp

@@ -3009,12 +3009,14 @@ jrd_prc* MET_procedure(thread_db* tdbb, int id, bool noscan, USHORT flags)
 	// or if the cached procedure instance is in use and thus cannot be re-scanned.
 	// In the latter case, the metadata cache slot is updated with the new instance
 	// and the old one will be automagically destroyed when all the interested
-	// requests are released (see CMP_decrement_prc_use_count).
+	// requests are released (see CMP_decrement_prc_use_count). But we must forcibly
+	// scan the procedure in place if it's being altered, as the callers expect the
+	// procedure instance to persist in the cache.
 	//
 	// Otherwise, if the cached procedure instance exists but is not used
 	// by any compiled statements, then it can be safely re-scanned in place.
 
-	if (!procedure || procedure->prc_use_count)
+	if (!procedure || (!(procedure->prc_flags & PRC_being_altered) && procedure->prc_use_count))
 	{
 		procedure = FB_NEW(*dbb->dbb_permanent) jrd_prc(*dbb->dbb_permanent);
 	}