libio.h troubles on gcc4 Haiku

[scottmc]

As posted on the HaikuPorts dev mailing list by Blub: ​http://lists.ports.haiku-files.org/pipermail/haikuports-devs-ports.haiku-files.org/2009-February/000262.html

It seems that the libio.h header has some flaws, in gnutls (which atm fails installing for me due to failing texi-references, so I'll try with gtk-doc support) and in expat I ran into some problems with _IO_{putc,peekc,getc}_unlocked defined multiple times, I then copied /boot/develop/headers/posix/libio.h into ./lib/, and made those functions static since they were inlined anyway.

If there's some other fix for that let me know please, if not, have a look at those files:

[scottmc]

Blub's suggested patch

[andreasf]

If we define them as static, we can drop the initial definitions.

expat is still happy then.

[titer]

#3537 "gcc4 issues with extern inlines and -std=gnu99" is a duplicate of this issue.

Grepping /boot/develop/headers for extern inlines, it seems libio.h would be the only file that requires patching.

[stippi]

Alternative stdio.h patch that keeps the semantics.

[stippi]

Thanks to the references given in #3537, I came up with the above patch. The difference is that the behaviour is maintained when compiling with -std99 or -gnu99. Could you please apply this and tell me if that solves the issues? Thanks!

[titer]

I do not believe we need to maintain the behavior: none of Haiku's code overrides the _IO_{putc,peekc,getc}_unlocked functions, and no code should ever assume they can be (look at Linux's libio.h, it just #defines them as macros).

I'm for simplicity, so making them static inlines regardless of what we are building (c89/c99/c++) would be my choice :)

[stippi]

Ok, I commited a simplified version of blub's patch in hrev29471. Thanks for the feedback.

[stippi]

For reference: <​http://www.greenend.org.uk/rjk/2003/03/inline.html>

[bonefish]

Reverted hrev29471 in hrev29547 as it breaks building kernel code in debug mode with gcc 2.95.3. Code for the static inline functions is generated and the referenced functions (underflow(),...) don't exist in the kernel.

[stippi]

Ingo, can you please check with hrev29558? I committed my original patch which should only fix the actual problem without changing any semantics.

[bonefish]

Replying to stippi:

Ingo, can you please check with hrev29558? I committed my original patch which should only fix the actual problem without changing any semantics.

gcc 2.95.3 debug builds work fine with this solution.

[stippi]

Thanks. I consider it fixed then.

[andreasf]

Nope, still with us: Building expat with gcc 4.3 continues to complain about multiple definitions of the functions.

The following workaround fixes the expat build for me:

1. Comment out the initial declarations of the three affected functions.
2. After the __INLINE wizardry, simply define __INLINE to static inline.

Could it be that this error is caused by the GCC 4.3 special handling section in libio.h? No special options for c99 or gnu99 are being passed afaict.

[kaliber]

If you want you can extract a correct patch for libio.h from TiltOS' patch ​http://svn.gna.org/viewcvs/pingwinek/trunk/projects/tiltos/tiltos.patch It is based on a newer code from glibc.

[zooey]

libio-patch suggested by kaliber

[zooey]

Extracted kaliber's suggested patch from TiltOS and attached it here, as libio.patch.

The resulting libio.h at least allows to compile rsync on gcc4, which failed before, too. The patch requires testing with gcc2, though.

[scottmc]

patch seems to work on gcc2 as well, at least I was able to build expat cvs head with the patched libio.h in place.

[zooey]

Fixed in hrev30861