Re: AC_FC_MODULE_FLAG

[Jim Edwards]

Okay, here is another method that seems to work,  I need to make the
subdirectory read-only then
try to compile another module in the main directory.   With the -M flag it
fails because it can't write
the .mod file in the conftest.dir

Index: fortran.m4
===================================================================
--- fortran.m4  (revision 327)
+++ fortran.m4  (working copy)
@@ -593,7 +593,7 @@
 # IBM: -Idir (-qmoddir=dir for writing)
 # Intel: -Idir -I dir (-mod dir for writing)
 # Absoft: -pdir
-# Lahey: -mod dir
+# Lahey: -Idir (-Mdir or -mod dir for writing)
 # Cray: -module dir, -p dir (-J dir for writing)
 #       -e m is needed to enable writing .mod files at all
 # Compaq: -Idir
@@ -615,17 +615,23 @@
       write(*,'(a)') 'gotcha!'
       end subroutine
       end module]],
-  [cd ..
+  # For Lahey -M will also write module and object files to that directory
+  # make it read-only so that lahey fails over to -I
+  [chmod -w .
+   cd ..
    ac_fc_module_flag_FCFLAGS_save=$FCFLAGS
    # Flag ordering is significant for gfortran and Sun.
    for ac_flag in -M -I '-I ' '-M ' -p '-mod ' '-module ' '-Am -I'; do
      # Add the flag twice to prevent matching an output flag.
      FCFLAGS="$ac_fc_module_flag_FCFLAGS_save ${ac_flag}conftest.dir
${ac_flag}conftest.dir"
      AC_COMPILE_IFELSE([[
-      program main
+      module conftest_main
       use conftest_module
+      contains
+      subroutine conftest
       call conftest_routine
-      end program]],
+      end subroutine
+      end module]],
        [ac_cv_fc_module_flag="$ac_flag"])
      if test "$ac_cv_fc_module_flag" != unknown; then
        break
@@ -633,6 +639,7 @@
    done
    FCFLAGS=$ac_fc_module_flag_FCFLAGS_save
 ])
+chmod +w conftest.dir
 rm -rf conftest.dir
 AC_LANG_POP([Fortran])
 ])






On Wed, Jul 6, 2011 at 2:15 PM, Ralf Wildenhues <address@hidden>wrote:

> Hi Jim,
>
> thanks for the bug report!
>
> * Jim Edwards wrote on Wed, Jul 06, 2011 at 08:00:13PM CEST:
> > When using the lahey fortran compiler the flag -M is used as the path to
> > modules, but its also the path to where new modules should be written.
> > So if you are referencing modules in a read-only directory using the -M
> flag
> > will cause your program to fail.   I think that the solution is to
> > test for -I first (that's also the most common result, so it would make
> > sense to test first I think)
>
> Well, with Solaris Fortran, -I is the flag to search for both header
> includes as well as modules, but it also has -M for module search path
> only.  I'm kind of wary that if some command-line gets hardcoded -M
> flags from elsewhere, they could then end up overriding the ones
> computed by configure, although they come later in the command-line.
>
> Can we fix the test somehow so that -M is not chosen with Lahey?  Adding
> the flag twice worked for some compiler at least.  Could also test
> whether the compiler creates a module somewhere.  Care to experiment a
> bit to find something that works with your compiler?  (I don't have
> Lahey to test.)
>
> > --- fortran.m4    (revision 327)
> > +++ fortran.m4    (working copy)
> > @@ -618,7 +618,7 @@
> >    [cd ..
> >     ac_fc_module_flag_FCFLAGS_save=$FCFLAGS
> >     # Flag ordering is significant for gfortran and Sun.
> > -   for ac_flag in -M -I '-I ' '-M ' -p '-mod ' '-module ' '-Am -I'; do
> > +   for ac_flag in -I '-mod ' -I '-I ' '-M ' -p '-module ' '-Am -I'; do
> >       # Add the flag twice to prevent matching an output flag.
> >       FCFLAGS="$ac_fc_module_flag_FCFLAGS_save ${ac_flag}conftest.dir
> > ${ac_flag}conftest.dir"
> >       AC_COMPILE_IFELSE([[
>
> Thanks,
> Ralf
>