On Sun, Jan 8, 2017 at 4:29 PM, Dominique d'Humières <domi...@lps.ens.fr> wrote:
>> r244027 reverts r244011. Sorry for the breakage. It seems to affect
>> all i686 as well in addition to power, maybe all 32-bit hosts.
>
> For the record, I see the following failures with an instrumented r244026 (as
> in pr78672)
[snip]
I was finally able to get a 32-bit i686 compiler going (my attempts to
do this on a x86_64-pc-linux-gnu host failed, in the end I resorted to
running 32-bit builds/tests on a i686 container). At least on i686,
the patch below on top of the big charlen->size_t patch fixes the
failures:
diff --git a/gcc/fortran/resolve.c b/gcc/fortran/resolve.c
index be63038..82319ed 100644
--- a/gcc/fortran/resolve.c
+++ b/gcc/fortran/resolve.c
@@ -4726,7 +4726,7 @@ gfc_resolve_substring_charlen (gfc_expr *e)
/* Length = (end - start + 1). */
e->ts.u.cl->length = gfc_subtract (end, start);
e->ts.u.cl->length = gfc_add (e->ts.u.cl->length,
- gfc_get_int_expr (gfc_default_integer_kind,
+ gfc_get_int_expr (gfc_charlen_int_kind,
NULL, 1));
/* F2008, 6.4.1: Both the starting point and the ending point shall
@@ -11420,9 +11420,10 @@ resolve_charlen (gfc_charlen *cl)
/* F2008, 4.4.3.2: If the character length parameter value evaluates to
a negative value, the length of character entities declared is zero. */
- if (cl->length && mpz_sgn (cl->length->value.integer) < 0)
+ if (cl->length && cl->length->expr_type == EXPR_CONSTANT
+ && mpz_sgn (cl->length->value.integer) < 0)
gfc_replace_expr (cl->length,
- gfc_get_int_expr (gfc_default_integer_kind, NULL, 0));
+ gfc_get_int_expr (gfc_charlen_int_kind, NULL, 0));
/* Check that the character length is not too large. */
k = gfc_validate_kind (BT_INTEGER, gfc_charlen_int_kind, false);
So what happened was that without the EXPR_CONSTANT check, I was
accessing uninitialized memory (for some reason probably due to memory
layout or such, this didn't cause failures on x86_64-pc-linux-gnu).
Also, I found a couple of additional places where gfc_charlen_int_kind
should be used instead of gfc_default_integer_kind which is included
in the patch above, although AFAICT they have nothing to do with the
testcase failures.
Unless there are objections, I'll commit the fixed patch in a few days.
--
Janne Blomqvist
