Re: pointer addition and arrays

[Marc Nieper-Wißkirchen]

Am Fr., 15. Dez. 2023 um 02:00 Uhr schrieb Paul Eggert <eggert@cs.ucla.edu>:

On 12/3/23 05:15, Andreas F. Borchert wrote:
> You will have
> then to possibly live with annoying warnings and odd behaviours.

The C standard says that when P is a null pointer, P==P must be true
whereas P<=P has undefined behavior. However, in practice any
implementation where P==P succeeds but P<=P fails is a pedantic
implementation that is merely enforcing what are arguably bugs in the
standard. Let's not waste time worrying about implementations like that.

I would agree if implementations did this because of pedantry, just because.  But the problem is if further down in the compiler pipeline, the P <= P _expression_ produces some intermediate code that becomes equal to code the optimizer should rightfully optimize away (or replace by the equivalent of __builtin_unreachable).

Gnulib should lead by example and follow the standard (on systems that implement the standard.  I like Example 2 on this page [1] very much.  It shows that well-justified optimizations can easily lead to consequences of UB that make the compiler look evil (or pedantic).  We cannot circumvent the principle of explosion ([2]).

Thanks,

Marc

--

[1] https://mohitmv.github.io/blog/Shocking-Undefined-Behaviour-In-Action/

[2] https://en.wikipedia.org/wiki/Principle_of_explosion