Re: i686-solaris

[Vincent Lefevre]

On 2013-07-12 11:15:15 +0200, Ludovic Courtès wrote:
> Vincent Lefevre <address@hidden> skribis:
> > On 2013-07-12 09:25:59 +0200, Zimmermann Paul wrote:
[...]
> >> configure:15350: gcc -std=gnu99 -o conftest -Wall -Wmissing-prototypes 
> >> -Wpointer-arith -m32 -O2 -pedantic -fomit-frame-pointer -mtune=pentium2 
> >> -march=pentium2 -ffloat-store 
> >> -I/nix/store/yz4r78ssvhz4cc0xj4z9mv98m2xz9q97-gmp-4.3.2/include 
> >> -L/nix/store/yz4r78ssvhz4cc0xj4z9mv98m2xz9q97-gmp-4.3.2/lib conftest.c 
> >> -lgmp  >&5
> >> configure:15350: $? = 0
> >> configure:15350: ./conftest
> >> gmp.h: 4.3.2 / libgmp: 4.3.1
[...]
> > It would be surprising to have a wrong GMP installation. I have the
> > impression that libgmp 4.3.1 comes from a system-wide installation,
> > e.g. /usr/lib,
> 
> Yes, but we provide explicit -I and -L flags pointing to 4.3.2 (see
> above; this comes from mpfr/release.nix.)

I thought they came from --prefix (in which case the problem would not
be specific to Hydra), but there may be something hidden in the logs.

> However, it may be that ./conftest above is picking /usr/lib/libgmp.so
> at run time, presumably because /nix/store/...-gmp-4.3.2/lib is not in
> the loader’s search path.  Nixpkgs’s ld-wrapper normally adds it to the
> RPATH [0], but perhaps this has lower precedence than /usr/lib?

It didn't: there's no trace of -rpath in the above gcc line.

And $LD_RUN_PATH was not predefined either. MPFR's configure does:

saved_LD_RUN_PATH="$LD_RUN_PATH"
LD_RUN_PATH="${LD_RUN_PATH:+$LD_RUN_PATH$PATH_SEPARATOR}$gmp_lib_path"
export LD_RUN_PATH
[...]
    AC_MSG_WARN([Temporary LD_RUN_PATH was "$LD_RUN_PATH".])

but the output was:

  Temporary LD_RUN_PATH was "".

($gmp_lib_path is set from --with-gmp, when it is used.)

-- 
Vincent Lefèvre <address@hidden> - Web: <http://www.vinc17.net/>
100% accessible validated (X)HTML - Blog: <http://www.vinc17.net/blog/>
Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)