Re: [F03] EXTENDS_TYPE_OF, CLASS IS and libgfortran

[Janus Weil <janus at gcc dot gnu dot org>]

2009/11/3 Steve Kargl <sgk@troutmask.apl.washington.edu>:
> On Tue, Nov 03, 2009 at 08:52:01PM +0100, Thomas Koenig wrote:
>> On Mon, 2009-11-02 at 23:45 +0100, Janus Weil wrote:
>> > However, since my experience with libgfortran is practically zero, I
>> > will need some help with this. What I would do is probably to add a
>> > new file 'extends_type_of.c' to libgfortran/intrinsics/, whose content
>> > could look something like this:
>>
>> [...]
>>
>>
>> > Now, first technical question: How do I correctly tell the libgfortran
>> > Makefile to pick up this file, and add the function to the library?
>>
>> You add the file to Makefile.am, probably to the gfor_helper_src
>> variable. ?Then you build everything.
>>
>> This regenerates Makefile.in by running autoconf automatically. ?Make
>> sure (beforehand :-) that you have the correct version of autoconf, and
>> make sure to check Makefile.in against the SVN version to catch any
>> unpleasant surprises that may suddenly raise their ugly head.
>>
>> Declare your function if you want it to be callable from the front end,
>> and use the export_proto macro on the declaration. ?You can do that in
>> the file itself, unless you want to call it from somewhere else; then it
>> should go into libgfortran.h (and you should use iexport_proto).
>>
>> Add your new global symbol to gfortran.map.
>>
>
> This is only the library side of the chnages.
>
> If these are truly intrinsic procedures, you may need to add
> code to intrinsic.c, check.c, simplify.c, iresolve.c, trans-intrinsic.c
> and gfortran.h.
>
> gfortran.h ?--> see gfc_isym_id.
> intrinsic.c --> add intrinsic procedure to the list of intrinsics with
> ? ? ? ? ? ? ? ?the right standard.
> check.c ? ? --> check that the arguments are valid.
> simplify.c ?--> if the arguments are constants, then simplify the expression.
> iresolve.c ?--> resolves the called procedure to the right library routine.
> trans-intrinsic.c --> use the gfc_isym_id to determine the translation.
> ? ? ? ? ? ? ? ? ? ? ?Note, if the procedures can be in-lined, you do it
> ? ? ? ? ? ? ? ? ? ? ?here.

Yes, most of this has been done already for EXTENDS_TYPE_OF, including
a function 'gfc_conv_extends_type_of' in trans-intrinsic.c, which so
far just outputs an error message ("... not implemented ...").

And as Tobias noted in PR41580, we may need to add simplifiers later.

Still, thanks for the reminder, Steve :)

Cheers,
Janus