Prof Brian D Ripley writes:
On Wed, 13 Feb 2002, Jan de Leeuw wrote:
If I configure in R-patched and R-devel, with the same options to
configure, then R-patched says
checking for ATL_xerbla in -latlas... yes
checking for cblas_dgemm in -lcblas... yes
checking for dgemm_ in -lf77blas... yes
which R-devel says
checking for ATL_xerbla in -latlas... yes
checking for cblas_dgemm in -lcblas... yes
checking for dgemm in -lf77blas... no
This is on the same computer and using the same libraries. Observe
in R-patched we check for dgemm_ and in R-devel for dgemm. Is that
what is intended ?
Confirmed here. It seems related to the change to autoconf 2.5x,
but it is not working as intended. The configure code has
if test "${r_cv_prog_f77_append_underscore}" = yes \
|| test -n "${F2C}"; then
dgemm_func=dgemm_
else
dgemm_func=dgemm
fi
...
AC_CHECK_LIB(f77blas, $dgemm_func,
BLAS_LIBS="-lf77blas $BLAS_LIBS", , $BLAS_LIBS $FLIBS)
and r_cv_prog_f77_append_underscore is being set correctly, but only
later in the configure. I think that comes from following the
autoconf recommended order of tests (libraries, headers, ...), but we
need an exception here.