Re: git: cd0727ec709b - main - libc: Add <stdio.h> C23 feature test macro

On 8/15/26 11:37, Faraz Vahedi wrote:
> Hello Dimitry,
> 
> Thanks for the report.
> 
>>> The test used by this package seems defective.
>>> I don't think we should try to support such defective code.
>>> Perhaps the gmp source can be patched to check for
>>> __STDC_VERSION_STDIO_H__?
>>
>> Of course it could, and the patch is pretty simple. But I do not agree
>> that the code is "defective". Before C23, there was no standard way of
>> checking whether you had FILE declared.
> 
> The approach is neither portable nor standard, and therefore I believe we
> should not contaminate our libc with an unnecessary compatibility macro.
> If gmp requires FILE, including <stdio.h> would be sufficient, and if for any
> reason it prefers not to include it and instead relies on some indication,
> that issue should be addressed therein.  We can patch the port and, in
> parallel, submit a fix upstream.
> 
> I honestly expected that we might encounter a case like this, but I would
> appreciate it if we could avoid defining an unnecessary macro solely to
> accommodate an approach that is not standards-compliant.
> 
>> That said, I have applied such a patch locally, and it seems to work.
>> But obviously a maintainer of a port would say "this has worked for 20
>> years, why change it now" ? :)
> 
> I agree that it is a reasonable argument for a port maintainer, but it doesn’t
> really establish that the approach is correct or portable.  It only establishes
> that the particular set of libc implementations the port has encountered
> happened to expose the macro it was looking for.
> 
> If the port is relying on an implementation-defined include guard, then a libc
> changing that guard is not, strictly speaking, breaking a supported interface.
> Consequently, the appropriate fix is for the port to fix it rather than for libc to
> preserve it indefinitely as a compatibility interface that is neither established
> nor widely recognised.
> 
> Cheers,
> Faraz

I might put it another way: if there is now a standard compliant way to
do the detection for some range of C standards (even a singleton), gmp
upstream should likely be updated to allow that way to work for it for
that range. That would involve continuing the historical-examples based
tests technique otherwise.

It might be that __STDC_VERSION_STDIO_H__ should be tested by gmp for
even when compiled for a C predating C23. (Not likely to have
inappropriate matching activity.) But FreeBSD might not be the only
context for the older _STDIO_H_ reference in gmp, so gmp likely would
not remove that usage.

I view all that as separate from if the include/stdio.h (or the like)
for FreeBSD should have _STDIO_H_ removed from being in the
historically-accessible programming interface. I do not know if the
principle of least astonishment should win here.

-- 
===
Mark Millard
marklmi at yahoo.com