------- Comment #1 from fxcoudert at gcc dot gnu dot org 2008-03-25 12:35
-------
Versions <= 4.2.3 fail also for SHAPE:
$ /opt/gcc-4.0.?/bin/gfortran a.f90 -static && ./a.out
UBOUND ZDA1= 0 -5 -1
SHAPE = 0 -5 -1
$ /opt/gcc-4.1.?/bin/gfortran a.f90 -static && ./a.out
UBOUND ZDA1= 0 -5 -1
SHAPE = 0 -5 -1
$ /opt/gcc-4.2.?/bin/gfortran a.f90 -static && ./a.out
UBOUND ZDA1= 0 -5 -1
SHAPE = 0 -5 -1
What's happening is that we simply forget to check for zero-sized sections. The
following code:
call s(1)
contains
subroutine s(n)
real x(10)
print *, ubound(x(5:n))
end subroutine
end program
is translated into:
parm.2.dim[0].lbound = 1;
parm.2.dim[0].ubound = *n + -4;
parm.2.dim[0].stride = 1;
parm.2.data = (void *) &x[4];
parm.2.offset = -5;
D.886 = (parm.2.dim[S.1].ubound - parm.2.dim[S.1].lbound) + 1;
_gfortran_transfer_integer (&dt_parm.0, &D.886, 4);
We're missing something like D.886 = MAX_EXPR (D.886, 0). I remembered taking
care of these problems in the past, in trans-intrinsic.c
(gfc_conv_intrinsic_bound), but apparently I forgot the block of code in the
second arm of the "if (as)" condition. I think the following patch does what is
required (but have not time for testing right now):
Index: trans-intrinsic.c
===================================================================
--- trans-intrinsic.c (revision 132747)
+++ trans-intrinsic.c (working copy)
@@ -930,6 +930,8 @@
size = fold_build2 (MINUS_EXPR, gfc_array_index_type, ubound,
lbound);
se->expr = fold_build2 (PLUS_EXPR, gfc_array_index_type, size,
gfc_index_one_node);
+ se->expr = fold_build2 (MAX_EXPR, gfc_array_index_type, se->expr,
+ gfc_index_zero_node);
}
else
se->expr = gfc_index_one_node;
--
fxcoudert at gcc dot gnu dot org changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |fxcoudert at gcc dot gnu dot
| |org
Known to fail| |4.0.4 4.1.2 4.2.3 4.3.0
| |4.4.0
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35685
