Re: git: a8b70cf26030 - main - netpfil: Use accessor functions and named constants for all tcphdr flags

From: <tuexen_at_fh-muenster.de>
Date: Fri, 29 Dec 2023 21:46:49 UTC
> On Dec 29, 2023, at 20:43, Antoine Brodin <antoine@freebsd.org> wrote:
> 
> On Fri, Dec 29, 2023 at 7:33 PM Warner Losh <imp@bsdimp.com> wrote:
>> On Fri, Dec 29, 2023, 4:39 AM Dimitry Andric <dim@freebsd.org> wrote:
>>> 
>>> On 29 Dec 2023, at 11:17, Dimitry Andric <dim@FreeBSD.org> wrote:
>>>> 
>>>> On 29 Dec 2023, at 08:35, Antoine Brodin <antoine@FreeBSD.org> wrote:
>>>>> 
>>>>> On Thu, Dec 28, 2023 at 10:37 PM Gleb Smirnoff <glebius@freebsd.org> wrote:
>>>>>> 
>>>>>> Antoine,
>>>>>> 
>>>>>> On Thu, Dec 28, 2023 at 08:48:36PM +0000, Antoine Brodin wrote:
>>>>>> A> >     netpfil: Use accessor functions and named constants for all tcphdr flags
>>>>>> A> >
>>>>>> ...
>>>>>> A> This breaks some ports:
>>>>>> A> /usr/include/netinet/tcp.h:82:8: error: unknown type name 'inline'
>>>>>> 
>>>>>> Definitely some ports that use some strange compilers :)
>>>>>> 
>>>>>> Can you please give at least port names?
>>>>> 
>>>>> Some examples:
>>>>> 
>>>>> https://pkg-status.freebsd.org/gohan04/data/mainamd64PR275986-default-foo/2023-12-28_20h35m41s/logs/errors/nspr-4.35.log
>>>>> https://pkg-status.freebsd.org/gohan04/data/mainamd64PR275986-default-foo/2023-12-28_20h35m41s/logs/errors/opusfile-0.12_1.log
>>>> 
>>>> The culprit is likely the "-ansi" option. Just get rid of that, it is not really useful:
>>>> 
>>>> cc -o prmapopt.o -c -fvisibility=hidden -O2 -pipe -fstack-protector-strong -fno-strict-aliasing -ansi -Wall -fPIC -UDEBUG -DPACKAGE_NAME=\"\" -DPACKAGE_TARNAME=\"\" -DPACKAGE_VERSION=\"\" -DPACKAGE_STRING=\"\" -DPACKAGE_BUGREPORT=\"\" -DPACKAGE_URL=\"\" -DNDEBUG=1 -DHAVE_VISIBILITY_HIDDEN_ATTRIBUTE=1 -DHAVE_VISIBILITY_PRAGMA=1 -DXP_UNIX=1 -DFREEBSD=1 -DHAVE_BSD_FLOCK=1 -DHAVE_SOCKLEN_T=1 -DHAVE_POINTER_LOCALTIME_R=1 -DHAVE_DLADDR=1 -DHAVE_LCHOWN=1 -DHAVE_SETPRIORITY=1 -DHAVE_STRERROR=1 -DHAVE_SYSCALL=1 -DHAVE_SECURE_GETENV=1 -D_REENTRANT=1 -D_THREAD_SAFE=1 -DFORCE_PR_LOG -D_PR_PTHREADS -UHAVE_CVAR_BUILT_ON_SEM -D_NSPR_BUILD_ -I../../../dist/include/nspr -I../../../pr/include -I../../../pr/include/private prmapopt.c
>>>> In file included from prmapopt.c:46:
>>>> In file included from prmapopt.c:46:
>>>> /usr/include/netinet/tcp.h:82:8: error: unknown type name 'inline'
>>>>  82 | static inline uint16_t
>>>>     |        ^
>>> 
>>> Hm, I may have spoken too soon here. If this port has always compiled successfully with -ansi, then indeed it may not be that handy to directly use the 'inline' keyword in system headers. I think most other system headers use '__inline', which is supported by both gcc and clang, even in ANSI mode. And if the compiler somehow does not support __inline, sys/cdefs.h can make sure all those keywords get removed.
>> 
>> 
>> Yes. You have to use __inline in system headers.
> 
> Yes I confirm that using __inline instead of inline fixes both nspr
> and opusfile.
Thanks for the feedback.

A patch is under discussion: https://reviews.freebsd.org/D43231

Best regards
Michael
> 
> Antoine