makecontext.3: Fix function declarator with empty parentheses.

glibc uses 'void (*f)(void)' for makecontext()'s second parameter.

C11 marked function declarators with empty parentheses as
obsolescent:


>   6.11.6  Function declarators
> 1 The use of function declarators with empty parentheses (not
>   prototype-format parametertype declarators) is an obsolescent
>   feature.


Let's use the correct syntax by explicitly using '(void)'.

.../glibc$ grep_glibc_prototype makecontext
stdlib/ucontext.h:51:
extern void makecontext (ucontext_t *__ucp, void (*__func) (void),
			 int __argc, ...) __THROW;
.../glibc$

Signed-off-by: Alejandro Colomar <alx.manpages@gmail.com>
---
 man3/makecontext.3 | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

On Thu, Mar 4, 2021 at 2:48 PM Alejandro Colomar via Libc-alpha <
libc-alpha@sourceware.org> wrote:

> glibc uses 'void (*f)(void)' for makecontext()'s second parameter.


Did you mean ‘void (*f)()’ ?

C11 marked function declarators with empty parentheses as
> obsolescent:
>
>
> >   6.11.6  Function declarators
> > 1 The use of function declarators with empty parentheses (not
> >   prototype-format parametertype declarators) is an obsolescent
> >   feature.
>
>
> Let's use the correct syntax by explicitly using '(void)


Unfortunately this change would be incorrect. makecontext’s second
parameter really is a pointer to a function that takes any number and type
of arguments, and there is no other way to write that in C than ‘void
(*)()’. Which, yes, means this function cannot be declared in conformant
C11.

This is actually the Austin Group’s primary rationale for deprecating
makecontext and its relatives.

zw

>

* Zack Weinberg:

> This is actually the Austin Group’s primary rationale for deprecating
> makecontext and its relatives.

That's a bit surprising because open and fcntl have a similar problem:
the argument type before the ellipsis cannot be int.

And doesn't a later C standard add a generic function pointer type?

Thanks,
Florian

Hello Zack, Florian, Michael,

On 3/4/21 8:45 PM, Alejandro Colomar wrote:
> glibc uses 'void (*f)(void)' for makecontext()'s second parameter.
>
> C11 marked function declarators with empty parentheses as
> obsolescent:
>
>
>>   6.11.6  Function declarators
>> 1 The use of function declarators with empty parentheses (not
>>   prototype-format parameter type declarators) is an obsolescent
>>   feature.

I quoted C11, but it's also in C99 (same section 6.11.6.1) and in C89
(in section 3.9.4) with the same wording.

>
>
> Let's use the correct syntax by explicitly using '(void)'.
>
> .../glibc$ grep_glibc_prototype makecontext
> stdlib/ucontext.h:51:
> extern void makecontext (ucontext_t *__ucp, void (*__func) (void),
> 			 int __argc, ...) __THROW;
> .../glibc$
>
> Signed-off-by: Alejandro Colomar <alx.manpages@gmail.com>
> ---


On 3/4/21 9:10 PM, Zack Weinberg wrote:
> On Thu, Mar 4, 2021 at 2:48 PM Alejandro Colomar via Libc-alpha
> <libc-alpha@sourceware.org <mailto:libc-alpha@sourceware.org>> wrote:
>
>     glibc uses 'void (*f)(void)' for makecontext()'s second parameter.
>
>
> Did you mean ‘void (*f)()’ ?

I did actually mean 'void (*f)(void)'.  Glibc uses that for the
prototype (as you can see from my commit message (see above)), and as I
confirmed just now, it also uses that type for the definition of the
function:

[
.../glibc$ grep -rn '^makecontext\s*('
stdlib/makecontext.c:22:makecontext (ucontext_t *ucp, void (*func)
(void), int argc, ...)
.../glibc$
]

However, I should have read the manual page (I must admit that I only
read the SYNOPSIS and EXAMPLES sections of the manual page and the glibc
source before writing the patch).  It's clear that the prototype that
was being used in the manual page was more correct (in the sense that it
more accurately represented the actual expected function pointer) than
the glibc prototype (eventhough the glibc prototype is more standards
conforming).

So my patch is wrong.

Florian, should I file a bug in glibc's bugzilla?

>
>     C11 marked function declarators with empty parentheses as
>     obsolescent:
>
>
>     >   6.11.6  Function declarators
>     > 1 The use of function declarators with empty parentheses (not
>     >   prototype-format parametertype declarators) is an obsolescent
>     >   feature.
>
>
>     Let's use the correct syntax by explicitly using '(void)
>
>
> Unfortunately this change would be incorrect. makecontext’s second
> parameter really is a pointer to a function that takes any number and
> type of arguments, and there is no other way to write that in C than
> ‘void (*)()’. Which, yes, means this function cannot be declared in
> conformant C11.

Yes, you're completely right!  Thanks for noticing.


On 3/4/21 10:10 PM, Florian Weimer wrote:
> * Zack Weinberg:
> 
>> This is actually the Austin Group’s primary rationale for deprecating
>> makecontext and its relatives.
> 
> That's a bit surprising because open and fcntl have a similar problem:
> the argument type before the ellipsis cannot be int.
> 
> And doesn't a later C standard add a generic function pointer type?

Ahh, I found it.  It's not in any standard, yet, but we might see it
soon in C2x: <http://www.open-std.org/jtc1/sc22/wg14/www/docs/n2230.htm>

Actually, one of the proposals is to reuse the empty parentheses for
this generic function pointer (thus removing it as a valid function
declaration), so the current prototype used in the manual page would
still be correct.

Another proposal is to add 'funcptr_t' for that, which seems to be
compatible with older standards.

You could even use it in glibc.


Michael, please ignore this patch.


Thank you all,

Alex

* Alejandro Colomar:

> I did actually mean 'void (*f)(void)'.  Glibc uses that for the
> prototype (as you can see from my commit message (see above)), and as I
> confirmed just now, it also uses that type for the definition of the
> function:
>
> [
> .../glibc$ grep -rn '^makecontext\s*('
> stdlib/makecontext.c:22:makecontext (ucontext_t *ucp, void (*func)
> (void), int argc, ...)
> .../glibc$
> ]
>
> However, I should have read the manual page (I must admit that I only
> read the SYNOPSIS and EXAMPLES sections of the manual page and the glibc
> source before writing the patch).  It's clear that the prototype that
> was being used in the manual page was more correct (in the sense that it
> more accurately represented the actual expected function pointer) than
> the glibc prototype (eventhough the glibc prototype is more standards
> conforming).
>
> So my patch is wrong.
>
> Florian, should I file a bug in glibc's bugzilla?

I don't know.  SUSv2 has (void *func) (), which would make this a glibc
bug.  I'm not sure if I have easy access to POSIX.1 from 2001, which I
believe still has this function.

I am not sure if all glibc implementations of makecontext can be used to
call variadic functions.

Thanks,
Florian

On Mär 05 2021, Florian Weimer via Libc-alpha wrote:

> I don't know.  SUSv2 has (void *func) (), which would make this a glibc
> bug.  I'm not sure if I have easy access to POSIX.1 from 2001, which I
> believe still has this function.

https://pubs.opengroup.org/onlinepubs/009695399/functions/makecontext.html

Andreas.

Hi Florian,

On 3/5/21 11:21 AM, Florian Weimer wrote:
> * Alejandro Colomar:
>> Florian, should I file a bug in glibc's bugzilla?
> 
> I don't know.  SUSv2 has (void *func) (), which would make this a glibc
> bug.  I'm not sure if I have easy access to POSIX.1 from 2001, which I
> believe still has this function.
> 
> I am not sure if all glibc implementations of makecontext can be used to
> call variadic functions.

I reported the bug: <https://sourceware.org/bugzilla/show_bug.cgi?id=27523>

Thanks,

Alex