a bug that breaks the performance testing mode of the libstdc++ test suite [_two_ dashes before "rpath" where one is the correct # to have]

Jonathan Wakely jwakely.gcc@gmail.com
Tue Mar 19 11:19:00 GMT 2013


On 18 March 2013 21:58, Abe wrote:
> Dear all,
>
> I recently found what seems to be a bug that breaks the performance
> testing mode of the libstdc++ test suite; I wrote a short, simple fix
> for it; since I don`t have write access to the source tree [or, AFAIK,
> direct submission access], here it is:

Patches for libstdc++ should be sent to the gcc-patches list as well.

> --- check_performance   2013-03-14 16:36:02.000000000 -0500
> +++ check_performance___orig    2010-01-06 16:55:52.000000000 -0600
> @@ -28,8 +28,8 @@
>  FLAGS=`$flags_script --cxxflags`
>  THREAD_FLAG='-pthread'
>  COMPILER=`$flags_script --build-cxx`
> -SH_FLAG="-Wl,-rpath -Wl,$BUILD_DIR/../../gcc \
> -         -Wl,-rpath -Wl,$BUILD_DIR/src/.libs"
> +SH_FLAG="-Wl,--rpath -Wl,$BUILD_DIR/../../gcc \
> +         -Wl,--rpath -Wl,$BUILD_DIR/src/.libs"
>  ST_FLAG="-static"
>  LINK=$SH_FLAG
>  CXX="$COMPILER $INCLUDES $FLAGS -std=gnu++0x $CXXFLAGS $LINK"
>
>
>
> The diff is against this file, which was still broken in "trunk" as of
> the last time I checked [the server seems to be down ATM]:
>
>   http://gcc.gnu.org/svn/gcc/trunk/libstdc++-v3/scripts/check_performance
>
>
>
> I`ve checked that this _does_, in fact, solve the problem.  IMO this
> tiny bug fix should be considered for inclusion in 4.7.3 if there will
> be such a release,  as well as mainlining it to trunk before 4.8 is
> released.

4.8 has already branched from trunk.

In any case, the patch looks wrong, the ld manual confirms that -rpath
is the correct option.

Since you haven't said what platform you're using I can't say much
more, but on GNU/Linux the current option is the right one.



More information about the Libstdc++ mailing list