On 11/8/26 6:58 PM, Jerry D wrote:
On 11/8/26 2:07 AM, Mikael Morin wrote:
Le 08/11/2026 à 10:15, Thomas Koenig a écrit :
Hi,
debugging a regression on a patch, I just stumbled across a
symbol with addr.flavor = FL_PARAMETER and attr.dummy = 1, which
IMHO makes less than zero sense.
This apparently comes from, in coarray.cc,
#define ADD_ARG(name, nsym, stype, skind, sintent) \
gfc_get_symbol (name, sub_ns,
&nsym); \
nsym->ts.type =
stype; \
nsym->ts.kind =
skind; \
nsym->attr.flavor =
FL_PARAMETER; \
nsym->attr.dummy =
1; \
nsym->attr.intent =
sintent; \
nsym->declared_at = expr-
>where; \
gfc_set_sym_referenced
(nsym); \
*argptr = gfc_get_formal_arglist
(); \
(*argptr)->sym =
nsym; \
However, I'm not into coarray code, so there may be some hidden meaning
to it after all.
Comments?.
This could have been written by a person knowing less the fortran world
than the C or middle-end world, where "parameter" means procedure argument.
Best regards
Thomas
Nice catch Thomas!
It needs to change to FL_VARIABLE in four places. This wiggles on one test case
so it is adjusted.
See attached trivial patch.
I will commit this tomorrow.
Regression tested on x86_64 -m32 and -m64
Cheers,
Jerry
Committed:
commit 122212d2fd6bc9547bb5280ae2f0bef20990f749 (HEAD -> master, origin/master,
origin/HEAD)
Author: Jerry DeLisle <[email protected]>
Date: Sun Nov 8 18:46:13 2026 -0800
fortran: Make coarray accessor dummy arguments variables
--- snip ---