On Mon, Jun 03, 2019 at 09:21:48PM +0200, Peter Eisentraut wrote:
> After many years of trying, it seems the -fsanitize=undefined checking
> in gcc is now working somewhat reliably. Attached is a patch that fixes
> all errors of the kind
>
> runtime error: null pointer passed as argument N, which is declared to
> never be null
>
> Most of the cases are calls to memcpy(), memcmp(), etc. with a length of
> zero, so one appears to get away with passing a null pointer.
I just saw this proposal. The undefined behavior in question is strictly
academic. These changes do remove the need for new users to discover
-fno-sanitize=nonnull-attribute, but they make the code longer and no clearer.
Given the variety of code this touches, I expect future commits will
reintroduce the complained-of usage patterns, prompting yet more commits to
restore the invariant achieved here. Hence, I'm -0 for this change.