Re: [Gcl-devel] The perils of gcc 5.x

[Camm Maguire]

Greetings, and thanks so much for your report!

git branch Version_2_6_13pre should fix this to my understanding.  If
not, please let me know.

The only obstacle to releasing 2.6.13 at the moment are some acl2 book
failures on machines with 1Gb ram due to oom.

Take care,

Jerry James <address@hidden> writes:

> On Thu, Jun 18, 2015 at 10:08 PM, Jerry James <address@hidden> wrote:
>> Greetings,
>>
>> Fedora is now building packages with gcc 5.x.  The gcl package build
>> for Fedora Rawhide (the upcoming Fedora 23 release) is failing with
>> gcc 5.1.1.  Part of the reason is that gcc now defaults to the C11
>> standard when building C code.  This breaks the gcl build, because the
>> "inline" keyword now defaults to meaning "static inline", where it
>> used to default to meaning "extern inline".  This leads to lots of
>> duplicate symbols AND lots of missing symbols at link time.
>
> I didn't say this quite right.  This page describes the issue much better:
>
> https://gcc.gnu.org/onlinedocs/gcc/Inline.html
>
> Essentially what happened is that gcc 5.x switched to building C code
> with C11 semantics by default, which means the semantics of the inline
> keyword changed.  GCL code is written to assume the C89 semantics, so
> everything broke.  Those building with gcc 5.x will need to add
> -std=c89 or -std=gnu89 to CFLAGS.
>
>> So somewhere the preprocessor is getting invoked and needs the -P flag
>> to suppress those line numbers, I guess.  I'm not sure where to look.
>> Any suggestions would be greatly appreciated.
>
> This fixes the problem:
>
> sed -i 's,$(CC) -E,& -P,' makefile
>
> Regards,

-- 
Camm Maguire                                        address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah