On 11/27/2015 08:42, Baptiste Daroussin wrote:
> On Fri, Nov 27, 2015 at 02:24:05PM +0100, Tijl Coosemans wrote:
>> On Fri, 27 Nov 2015 13:59:54 +0100 Baptiste Daroussin <b...@freebsd.org>
>> wrote:
>>> On Tue, Nov 10, 2015 at 07:23:40PM +0100, Rainer Hurling wrote:
>>>> I think I found the problem.
>>>>
>>>> In my initial mail of this thread, I reported, that after upgrading
>>>> Freebsd 11.0-CURRENT to r290538 (including locale and localedef updates)
>>>> I am not able to build lang/gccXX any more. All I get are errors like
>>>> that in usr/ports/lang/gccXX/work/build/gcc:
>>>>
>>>> ----
>>>> In file included from .././../gcc-4.8.5/gcc/genflags.c:26:
>>>> In file included from ./tm.h:16:
>>>> ./options.h:4293:3: error: redefinition of enumerator 'OPT_C'
>>>> OPT_C = 129, /* -C */
>>>> ^
>>>> ----
>>>>
>>>> After more than 20 of them the build stops with
>>>> fatal error: too many errors emitted, stopping now [-ferror-limit=]
>>>> 20 errors generated.
>>>>
>>>>
>>>> This is with locale for Germany:
>>>> LANG=de_DE.UTF-8
>>>> LC_CTYPE="de_DE.UTF-8"
>>>> LC_COLLATE="de_DE.UTF-8"
>>>> LC_TIME="de_DE.UTF-8"
>>>> LC_NUMERIC="de_DE.UTF-8"
>>>> LC_MONETARY="de_DE.UTF-8"
>>>> LC_MESSAGES="de_DE.UTF-8"
>>>> LC_ALL=
>>>>
>>>>
>>>> If I use 'LC_COLLATE="C"' for the build, the build works fine again:
>>>>
>>>> cd /usr/ports/lang/gcc48
>>>> env LC_COLLATE="C" make
>>>> ...
>>>>
>>>>
>>>> So it seems, that something with the new 'locale' code in base of HEAD
>>>> is not working as expected here? (At least for other locales than US?)
>>>>
>>>> I added bapt@, because he is the author introducing the new code into HEAD.
>>>>
>>>> Hope, my explanations are clear enough to get the problem. Please feel
>>>> free to ask for more information, if needed.
>>>
>>> Your explanations are good, sorry for the delay for my reply I will look
>>> into it.
>>
>> I ran into this with lang/sdcc* which includes cpp from gcc. At some
>> point the build runs an awk script (opt-gather.awk) that collects command
>> line options defined in *.opt files, sorts them and then puts them into
>> a C array. The problem seems to be that FreeBSD awk takes collation into
>> account when sorting strings while GNU awk doesn't. POSIX says that
>> FreeBSD is correct, so I was thinking of adding something like this to
>> bsd.port.mk:
>>
>> USE_LOCALE?= C
>> LANG= ${USE_LOCALE}
>> LC_ALL= ${USE_LOCALE}
>> .export LANG LC_ALL
>>
>> This gives a consistent locale environment for port builds. Some ports
>> already set LANG or LC_ALL. That would have to be reviewed and I haven't
>> had time for that yet.
>
> yeah we should imho to that. The same issue appears with GNU tr which also
> do not take in account collation (IIRC) while ours do.
FWIW, I ran into this building devel/arm-none-eabi-gcc too. Setting
LC_ALL and LANG fixed it for me. Also, Mk/bsd.ruby.mk got something
similar for gem builds a while ago, though it uses en_US.UTF-8 for
LC_ALL and LC_CTYPE=UTF-8. As I recall, this was due to difference in
generated docs based on language. We may want to remove the bits from
Mk/bsd.ruby.mk if we add it globally. Also, I wonder what impacts this
will have beyond fixing builds, such as docs, plist differences or baked
in defaults, etc.
Steve
_______________________________________________
freebsd-ports@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscr...@freebsd.org"
