Re: GCC 2.95.2 and GLIBC 2.1.93, broken streambuf.h

[Ben Collins <bcollins at debian dot org>]

Appoliges, I meant for this to go to gcc-bugs, not patches.

On Fri, Sep 01, 2000 at 11:14:01PM -0400, Ben Collins wrote:
> I saw this posted about 2 weeks ago to the libc-alpha list, and Ulrich's
> answer was basically that gcc was at fault. I never saw it come up on the
> gcc lists, and a web search shows nothing in the gcc-bugs archives.
> 
> Basically installing glibc 2.1.9[123] built with gcc-2.95.2 will not allow
> C++ applications to compile. Here's the error:
> 
> /usr/lib/gcc-lib/powerpc-linux/2.95.2/../../../../include/g++-3/streambuf.h: In method `struct streampos streambuf::pubseekoff(long long int, ios::seek_dir, int = 3)':
> /usr/lib/gcc-lib/powerpc-linux/2.95.2/../../../../include/g++-3/streambuf.h:362: conversion from `__off64_t' to non-scalar type `streampos' requested
> /usr/lib/gcc-lib/powerpc-linux/2.95.2/../../../../include/g++-3/streambuf.h: In method `struct streampos streambuf::pubseekpos(_G_fpos64_t, int = 3)':
> /usr/lib/gcc-lib/powerpc-linux/2.95.2/../../../../include/g++-3/streambuf.h:364: `struct streampos' used where a `long long int' was expected
> /usr/lib/gcc-lib/powerpc-linux/2.95.2/../../../../include/g++-3/streambuf.h:364: warning: control reaches
> end of non-void function `streambuf::pubseekpos(_G_fpos64_t, int)'
> 
> This occurs on i386 and sparc too, so there's nothing special about the
> architecture. Now, when I try to recompile gcc 2.95.2 with glibc 2.1.93
> installed, I get this:
> 
>   /usr/src/gcc/gcc-2.95.2/build/gcc/xgcc -B/usr/src/gcc/gcc-2.95.2/build/gcc/ -B/usr/powerpc-linux/bin/ -c -g -O2 -fvtable-thunks -D_GNU_SOURCE -fno-implicit-templates -I. -I/usr/src/gcc/gcc-2.95.2/src/libio -nostdinc++ -D_IO_MTSAFE_IO -fPIC /usr/src/gcc/gcc-2.95.2/src/libio/filebuf.cc -o pic/filebuf.o
> /usr/src/gcc/gcc-2.95.2/src/libio/filebuf.cc: In method `class filebuf * filebuf::open(const char *, int,
> int = 436)':
> /usr/src/gcc/gcc-2.95.2/src/libio/filebuf.cc:119: no match for `streampos == int'
> 
> 
> Now, I'm not a C++ person. I can see the problem, but the resolution is
> beyond my know how. Is there a fix for this? This is in regard to moving
> Debian's current (6 archs) unstable to glibc 2.2pre. We want to do this in
> phases for several reasons, not the least of which is the extreme overhead
> in recompiling things that need to be (soname change for libstdc++ etc..),
> but also to actually not break upgradability from previous releases of
> Debian, and to ensure we can easily differentiate problems caused by the
> gcc and glibc upgrades. This is why we aren't going to just drop current
> gcc CVS and glibc 2.1.93 in at the same time.
> 
> Help appreciated,
>   Ben
> 
> -- 
>  -----------=======-=-======-=========-----------=====------------=-=------
> /  Ben Collins  --  ...on that fantastic voyage...  --  Debian GNU/Linux   \
> `  bcollins@debian.org  --  bcollins@openldap.org  --  bcollins@linux.com  '
>  `---=========------=======-------------=-=-----=-===-======-------=--=---'
> 

-- 
 -----------=======-=-======-=========-----------=====------------=-=------
/  Ben Collins  --  ...on that fantastic voyage...  --  Debian GNU/Linux   \
`  bcollins@debian.org  --  bcollins@openldap.org  --  bcollins@linux.com  '
 `---=========------=======-------------=-=-----=-===-======-------=--=---'