RFD: improved diagnostics when reading module files

[Steve Kargl]

On Fri, Apr 06, 2018 at 11:17:58PM +0200, Harald Anlauf wrote:
> On 04/06/18 21:58, Steve Kargl wrote:
> > I would remove the "While reading module files:" and simply
> > start the error message with "Mismatch".  I haven't looked at
> > module.c, but if the module name or derived type is available
> > you might considering added that to the message.   If you
> > try to reference a locus with "at %C", do you get a caret
> > pointing at a "USE XXXX" statement?  This might also be helpful.
> >> +               }
> >> +             /* gcc_assert (comp_name == c->name); */
> > 
> > Delete the above line.
> 
> With
> 
> Index: gcc/fortran/module.c
> ===================================================================
> --- gcc/fortran/module.c        (revision 259189)
> +++ gcc/fortran/module.c        (working copy)
> @@ -5161,7 +5161,12 @@
>               if (p->u.pointer == NULL)
>                 associate_integer_pointer (p, c);
>               mio_pool_string (&comp_name);
> -             gcc_assert (comp_name == c->name);
> +             if (comp_name != c->name)
> +               {
> +                 gfc_error_now ("Mismatch in derived type components of "
> +                                "%qs from %qs at %C: %qs != %qs",
> +                                sym->name, sym->module, c->name,
> comp_name);
> +               }
>               skip_list (1); /* component end.  */
>             }
>           mio_rparen (); /* component list closing.  */
> 
> 
> I get some improved output.  I'm not really overwhelmed, but it is
> definitely better from my point of view.  OTOH, removing the %C
> makes the error message definitely less useful.
> 
> I haven't been able to produce a simple but sufficiently contorted
> testcase.  But for the case I was debugging, where there was some
> outdated module file holding the outdated layout of a derived type,
> I was able to zgrep back to the culprit.
> 
> With the change above, I now get:
> 
> ../../../../analysis/mo_ir_emis.f90:61:6:
> 
>    use mo_dec_matrix,  only: t_vector       ! decomposed vector
>       1
> Error: Mismatch in derived type components of 't_par_grib2' from
> 'mo_grib12_dwd' at (1): 'modnum' != 'shortname'

Yes, I think it is useful.  An assert() gives an ICE, which is
not useful in this situation.  A slight better error message 
might be

Error: Mismatch in components of derived type 't_par_grib2' from
module 'mo_grib12_dwd' at (1) (expecting 'modnum', but got 'shortname')

You may want to let others way.

BTW, do you have commit privileges for the subversion repository?

-- 
Steve