INT_ADD_OVERFLOW broken?

INT_ADD_OVERFLOW broken?

[Daniel Colascione]

[-- Attachment #1: Type: text/plain, Size: 388 bytes --]

When trying to compile this code:

  if (INT_ADD_OVERFLOW ((ptrdiff_t) (bool_header_size - header_size),
                        total_payload_bytes))
    memory_full (SIZE_MAX);

gcc produces this warning:

alloc.c:2051: warning: assuming signed overflow does not occur when
assuming that (X + c) >= X is always true [-Wstrict-overflow]

Is INT_ADD_OVERFLOW actually safe?


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: INT_ADD_OVERFLOW broken?

[Paul Eggert]

Daniel Colascione wrote:

> Is INT_ADD_OVERFLOW actually safe?

It's supposed to be, yes.

As I recall, -Wstrict-overflow and -Wtype-limits issue many false
alarms, for INT_ADD_OVERFLOW as well as for other things, so I don't
recommend their use when building GNU Emacs.  That's why
--enable-gcc-warnings doesn't enable these options.

It's possible that you've found a bug in INT_ADD_OVERFLOW;
to check that it'd be helpful to have a self-contained example
so that the problem can be reproduced and understood.