On 19/07/2025 21:54, Segher Boessenkool wrote:
On Sat, Jul 19, 2025 at 12:57:46PM +0200, David Brown wrote:
"IB" and "UB" are a world apart. "IB" lets you write efficient low-level
code tuned to a specific compiler, target or system, at the price of
portability. It is perfectly fine - indeed crucial to most C programs - to
rely on "IB". It is never appropriate to rely on the effects of "UB".
An implementation is free to implement any particular IB as UB, it just
has to document it does that!
Writing "the compiler does not define this behaviour" is not defining or
documenting anything.
It seems to me - with all due respect - that you fundamentally
misunderstand what "implementation-defined behaviour" means.
Can I suggest you read the relevant section of the gcc manual at
<https://gcc.gnu.org/onlinedocs/gcc/C-Implementation.html>, and perhaps
the corresponding Annex J.2 of the C standards (any version should do,
the changes there are minor) ?
Please think about what it would mean if a compiler documented any of
these IB's as UB. Think how "useful" a C compiler would be if it
declared that the size of an "int" is undefined behaviour, or any other
of the many IB's in C.
If that does not convince you, then I am don't know what else to say
without going round in circles. While I have studied the C standards
more than the solid majority of C programmers, I have no grounds for
claiming authority or expecting you to agree just because /I/ said so.
However, there are plenty of GCC developers who have a detailed
understanding of the standards and are maybe even involved in them, who
might be able to help here.
David
