| Message ID | BD480C51-EFC4-42AD-906D-AAE376D8751E@sandoe-acoustics.co.uk |
|---|---|
| State | New |
| Headers | show |
On Thu, Nov 25, 2010 at 11:18:41AM +0000, IainS wrote: > This will allow fortran libgomp tests to use libquadmath when > required (I'm told none do yet). > .. but it will also ensure that, if the libgfortran.spec includes > substitutions, they are available in the testsuite. > > checked (in conjunction with > http://gcc.gnu.org/ml/fortran/2010-11/msg00348.html) that the static > lib substitution occurs properly for i686-darwin9. > > OK for trunk (with a proper Changelog, of course)? This is wrong. If anything (we don't have any test that needs it), it should be adding -B${blddir}/${lang_spec_path}/ being added, to let the driver find it automatically if it wants to. But forcing -specs when the driver doesn't want it just means we would be testing something that isn't done in the installed compiler. Jakub
On 25 Nov 2010, at 18:11, Jakub Jelinek wrote: > On Thu, Nov 25, 2010 at 11:18:41AM +0000, IainS wrote: >> This will allow fortran libgomp tests to use libquadmath when >> required (I'm told none do yet). >> .. but it will also ensure that, if the libgfortran.spec includes >> substitutions, they are available in the testsuite. >> >> checked (in conjunction with >> http://gcc.gnu.org/ml/fortran/2010-11/msg00348.html) that the static >> lib substitution occurs properly for i686-darwin9. >> >> OK for trunk (with a proper Changelog, of course)? > > This is wrong. If anything (we don't have any test that needs it), well, we can't test -static-libgfortran without it on darwin (assuming that we implement that). > it should be adding -B${blddir}/${lang_spec_path}/ being added, > to let the driver find it automatically if it wants to. > But forcing -specs when the driver doesn't want it just means > we would be testing something that isn't done in the installed > compiler. AFAIU, the installed driver is gfortran - which understands/adds the spec... ... the libgomp testsuite is driven by xgcc - which doesn't. (I believe you would have to do the same if you wished to build a quadmath fortran program from the c/l using 'gcc') Iain
Index: libgomp/testsuite/libgomp.fortran/fortran.exp =================================================================== --- libgomp/testsuite/libgomp.fortran/fortran.exp (revision 167142) +++ libgomp/testsuite/libgomp.fortran/fortran.exp (working copy) @@ -8,6 +8,7 @@ set lang_library_path "../libgfortran/.libs" set lang_link_flags "-lgfortran" set lang_test_file_found 0 set quadmath_library_path "../libquadmath/.libs" +set lang_spec_path "../libgfortran" # Initialize dg. @@ -44,6 +45,8 @@ if { $lang_test_file_found } { lappend ALWAYS_CFLAGS "ldflags=-L${blddir}/${quadmath_library_path}/" # Allow for spec subsitution. lappend ALWAYS_CFLAGS "additional_flags=-B${blddir}/$ {quadmath_library_path}/" + # use the libgfortran spec. + lappend ALWAYS_CFLAGS "additional_flags=-specs=${blddir}/$ {lang_spec_path}/libgfortran.spec" set ld_library_path "$always_ld_library_path:${blddir}/$
This will allow fortran libgomp tests to use libquadmath when required (I'm told none do yet). .. but it will also ensure that, if the libgfortran.spec includes substitutions, they are available in the testsuite. checked (in conjunction with http://gcc.gnu.org/ml/fortran/2010-11/msg00348.html) that the static lib substitution occurs properly for i686-darwin9. OK for trunk (with a proper Changelog, of course)? Iain {lang_library_path}:${blddir}/${quadmath_library_path}" } else { set ld_library_path "$always_ld_library_path" ===