[Bug fortran/57522] New: [F03] ASSOCIATE construct creates array descriptor with incorrect stride for derived type array component

[Bug fortran/57522] New: [F03] ASSOCIATE construct creates array descriptor with incorrect stride for derived type array component

[alan.briolat at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57522

            Bug ID: 57522
           Summary: [F03] ASSOCIATE construct creates array descriptor
                    with incorrect stride for derived type array component
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: major
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: alan.briolat at gmail dot com

Created attachment 30254
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=30254&action=edit
Code demonstrating the bug

The attached code demonstrates a bug (reproducible on 4.6 to 4.9) where using
ASSOCIATE to reference a derived type component in an array appears to have the
wrong stride.  The output is:

           0           1           2           3
           0           4           1           5

when it should be (and is under ifort):

           0           1           2           3
           0           1           2           3

This is probably related to bug 49636, but is a simpler case.  It's also
probably related to http://gcc.gnu.org/wiki/ArrayDescriptorUpdate (which I
found via that bug).  However, if the changes necessary to support this sort of
reference are "future ABI-breaking work", then the current version really
should warn/error rather than generating the wrong stride and silently doing
the wrong thing.

[Bug fortran/57522] [F03] ASSOCIATE construct creates array descriptor with incorrect stride for derived type array component

[alan.briolat at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57522

--- Comment #1 from Alan Briolat <alan.briolat at gmail dot com> ---
Created attachment 30255
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=30255&action=edit
Working "equivalent" code based on pointers

For comparison, this attached code uses pointers to achieve the same thing and
gets the correct result.  What's so different between the array descriptors
used for pointers and the array descriptors use for ASSOCIATE?

[Bug fortran/57522] [F03] ASSOCIATE construct creates array descriptor with incorrect stride for derived type array component

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57522

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2013-07-02
     Ever confirmed|0                           |1

--- Comment #2 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
Still present at revision 200581.

[Bug fortran/57522] [F03] ASSOCIATE construct creates array descriptor with incorrect stride for derived type array component

[pault at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57522

--- Comment #4 from Paul Thomas <pault at gcc dot gnu.org> ---
(In reply to Tobias Burnus from comment #3)
> > print *, a
....
> 
> I have not yet completely understood why, but the call
> gfc_conv_expr_descriptor yields a unit stride.

We have a choice here:
(i) We use the 'span' variable that is used for pointers to some useful effect.
Ideally, this should be accompanied by passing of the 'span' variable to
pointer formal arguments in function calls.  That way, even if kludgy we would
have a full implementation of the F95(!) standard (see for example PR40737 +
others that I cannot find right now); or
(ii) That we attach 'span' to pointer array descriptors; or
(iii) We recognised one the the ASSOCIATE todos, which is the make a temporary
when needed and write the selector result to that temporary. This is also
required for SELECT TYPE. Examples are where the selector is a function call.

(ii)Would cause a diversion from the array descriptor reform but are we really
in a position to implement that?  Both Tobias and I are taken by daytime work
for the foreseeable future.  It certainly would be a clean way to proceed.

(iii) Must be done sometime... soon. It could be used to fix this case using
write-in/write-out but would not fix the F95 problems.

I am prepared to make this my next gfortran job.  I would suggest (i) or (ii)
followed by (iii).

Cheers

Paul

[Bug fortran/57522] [F03] ASSOCIATE construct creates array descriptor with incorrect stride for derived type array component

[pault at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57522

--- Comment #5 from Paul Thomas <pault at gcc dot gnu.org> ---
Author: pault
Date: Sun Feb  9 20:50:21 2014
New Revision: 207646

URL: http://gcc.gnu.org/viewcvs?rev=207646&root=gcc&view=rev
Log:
2014-02-09  Paul Thomas  <pault@gcc.gnu.org>

    PR fortran/57522
    * resolve.c (resolve_assoc_var): Set the subref_array_pointer
    attribute for the 'associate-name' if necessary.
    * trans-stmt.c (trans_associate_var): If the 'associate-name'
    is a subref_array_pointer, assign the element size of the
    associate variable to 'span'.

2014-02-09  Paul Thomas  <pault@gcc.gnu.org>

    PR fortran/57522
    * gfortran.dg/associated_target_5.f03 : New test

Added:
    trunk/gcc/testsuite/gfortran.dg/associated_target_5.f03
Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/resolve.c
    trunk/gcc/fortran/trans-stmt.c
    trunk/gcc/testsuite/ChangeLog

[Bug fortran/57522] [F03] ASSOCIATE construct creates array descriptor with incorrect stride for derived type array component

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57522

--- Comment #6 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
Any reason why tho PR is not closed as fixed?

[Bug fortran/57522] [F03] ASSOCIATE construct creates array descriptor with incorrect stride for derived type array component

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57522

--- Comment #7 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
*** Bug 58339 has been marked as a duplicate of this bug. ***

[Bug fortran/57522] [F03] ASSOCIATE construct creates array descriptor with incorrect stride for derived type array component

[pault at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57522

Paul Thomas <pault at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |FIXED

--- Comment #8 from Paul Thomas <pault at gcc dot gnu.org> ---
No reason whatsoever, except that I forgot!

Thanks for the report.

Paul