Frank Lichtenheld <[EMAIL PROTECTED]> writes: > | [CC] src/main/event.o > | cc1: warnings being treated as errors > | /build/buildd/elinks-0.12~20080127/src/main/event.c: In function > 'unregister_event_hook': > | /build/buildd/elinks-0.12~20080127/src/util/math.h:36: error: assuming > signed overflow does not occur when assuming that (X + c) < X is always false
The code in unregister_event_hook() is:
| for (i = 0; i < event->count; i++) {
| if (event->handlers[i].callback != callback)
| continue;
|
| move_event_handler(event, i, i + 1);
which calls:
| static inline void
| move_event_handler(struct event *event, int to, int from)
| {
| int d = int_max(to, from);
which is defined in src/util/math.h as:
| static inline int
| int_max(register int x, register int y)
| {
| if (x > y) return x;
| return y;
| }
So GCC warns about the i > i + 1 comparison. I think the i + 1
is very unlikely to overflow in practice, although it is possible
because event->count is unsigned int and i is signed int. Also,
the overflow would need a 64-bit machine because otherwise
register_event_hook() would run out of memory first.
However, there are other places in ELinks where signed overflows
are possible and are expected to wrap around. The one such place
I know about is parse_bencoding_integer(). You should therefore
compile ELinks with -fno-strict-overflow or -fwrapv. I will add
one of those to the configure script when I have time.
pgpeRuUXlEgGC.pgp
Description: PGP signature
