I'm going to try to clean up the stuff valgrind is complaining about. The leakage stuff will take a little time, it looks like, so I'm going for the low-hanging fruit first. As XEmacs runs, valgrind complains bitterly about lots and lots of cases where we essentially memcpy(&x, &x, sizeof(x)). Since memcpy() is only guaranteed to work on nonoverlapping addresses, a conforming memcpy implementation *could* scramble the bytes of x. It turns out these memcpy()s are gcc's doing. For structs x and y, "x = y;" gets turned into a memcpy call, without first checking that &x != &y. That should probably be considered a gcc bug, but on the other hand, there's no need for us to do "x = x;" either. So here's the first case I hit. In event-Xt.c, function emacs_Xt_event_handler, line 1460 says: x_event_copy = &EVENT_MAGIC_X_EVENT (emacs_event); which the preprocessor expands to: x_event_copy = &((emacs_event)->event.magic.underlying.x_event); Next a bunch of stuff happens in a switch statement, followed on line 1496 by: SET_EVENT_MAGIC_X_EVENT (emacs_event, *x_event_copy); which the preprocessor expands to: do { (emacs_event)->event.magic.underlying.x_event = (*x_event_copy); } while (0); There is no code in between that changes the value of the pointers emacs_event or x_event_copy, so this amounts to: x = &y; y = *x; As far as I can see, the SET_EVENT_MAGIC_X_EVENT call is a no-op. I'd like to remove it. Any objections? -- Jerry James http://loganjerry.googlepages.com/ _______________________________________________ XEmacs-Beta mailing list XEmacs-Beta(a)xemacs.org http://calypso.tux.org/cgi-bin/mailman/listinfo/xemacs-beta