On 22.01.25 19:16, Peter Eisentraut wrote:
> On 06.01.25 15:52, Peter Eisentraut wrote:
>> On 03.01.25 21:51, Dagfinn Ilmari Mannsåker wrote:
>>> Peter Eisentraut <peter(at)eisentraut(dot)org> writes:
>>>
>>>> I suggest we define pg_noreturn as
>>>>
>>>> 1. If C11 is supported, then _Noreturn, else
>>>> 2. If GCC-compatible, then __attribute__((noreturn)), else
>>>
>>> Would it be worth also checking __has_attribute(noreturn)? Or do all
>>> compilers that have __attribute__((noreturn)) claim to be GCC?
>>
>> I don't think that would expand the set of supported compilers in a
>> significant way. We can always add it if we find one, of course.
>
> In fact, as another thought, we could even drop #2. Among the GCC-
> compatible compilers, both GCC and Clang have supported #1 for ages, and
> the only other candidate I could find on the build farm is the Solaris
> compiler, which also supports C11 by default, per its documentation.
>
>>>> 3. If MSVC, then __declspec((noreturn))
Here is an updated patch set that contains the above small change and
fixes some conflicts that have arisen in the meantime.