| Summary: | [4.6/4.7 Regression] ICE on array assignment statement with allocatable LHS | ||
|---|---|---|---|
| Product: | gcc | Reporter: | Keith Refson <krefson> |
| Component: | fortran | Assignee: | Paul Thomas <pault> |
| Status: | RESOLVED FIXED | ||
| Severity: | normal | CC: | burnus, jakub, jvdelisle2, krefson, pault |
| Priority: | P4 | Keywords: | ice-on-valid-code |
| Version: | 4.6.0 | ||
| Target Milestone: | 4.6.1 | ||
| Host: | Target: | ||
| Build: | Known to work: | ||
| Known to fail: | Last reconfirmed: | 2011-04-08 04:16:46 | |
| Bug Depends on: | |||
| Bug Blocks: | 48456, 48462 | ||
| Attachments: | Fortran 95 source which gives ICE on compilation attempt. | ||
|
Description
Keith Refson
2011-03-30 11:02:10 UTC
(In reply to comment #0) > Disabling the possibility of "allocate-on-assignment" by changing [...] Or simply use: -fno-realloc-lhs With my - admittedly not clean - tree, that gives the following valgrind output: Invalid read of size 8 at 0x56B13A: get_std_lbound (trans-array.c:6825) by 0x573E88: gfc_alloc_allocatable_for_assignment (trans-array.c:7083) by 0x58EBDB: gfc_trans_assignment_1 (trans-expr.c:6186) That's the lines: 6822 else if (expr->expr_type == EXPR_VARIABLE) 6823 { 6824 tmp = TREE_TYPE (expr->symtree->n.sym->backend_decl); 6825 return GFC_TYPE_ARRAY_LBOUND(tmp, dim); The issue is that TYPE_LANG_SPECIFIC (tmp) == tmp->type.lang_specific == NULL and, thus, accessing TYPE_LANG_SPECIFIC(node)->lbound[dim] gives an ICE. The issue is that for "mmv%h0" the proper backend is not: expr->symtree->n.sym->backend_decl but rather expr->ref->u.c->component->backend_decl I am a little suspicious of this: static int count_arglist; +/* Pointer to an array of gfc_expr ** we operate on, plus its size + and counter. */ + +static gfc_expr ***expr_array; +static int expr_size, expr_count; It could be getting walked on or optimized away by the optimizers. Is there another way to do this that is less indirect or relies less on static. Just seems too tricky. I am not looking further on this one. Disregard 2, wrong PR This regression was caused by r167220. Revision 167220 - (view) (download) (annotate) - [select for diffs] Modified Sun Nov 28 13:47:26 2010 UTC (4 months ago) by pault This is obviously mine :-)
This fixes it but I suspect that a few more conditions (no ARRAY_ELEMENT for example) will be needed for it to be correct.
Index: gcc/fortran/trans-array.c
===================================================================
*** gcc/fortran/trans-array.c (revision 171573)
--- gcc/fortran/trans-array.c (working copy)
*************** get_std_lbound (gfc_expr *expr, tree des
*** 6707,6712 ****
--- 6707,6713 ----
tree stride;
tree cond, cond1, cond3, cond4;
tree tmp;
+ gfc_ref *ref;
if (GFC_DESCRIPTOR_TYPE_P (TREE_TYPE (desc)))
{
tmp = gfc_rank_cst[dim];
*************** get_std_lbound (gfc_expr *expr, tree des
*** 6740,6745 ****
--- 6741,6752 ----
else if (expr->expr_type == EXPR_VARIABLE)
{
tmp = TREE_TYPE (expr->symtree->n.sym->backend_decl);
+ for (ref = expr->ref; ref; ref = ref->next)
+ {
+ if (ref->type == REF_COMPONENT
+ && ref->u.c.component->as)
+ tmp = TREE_TYPE (ref->u.c.component->backend_decl);
+ }
return GFC_TYPE_ARRAY_LBOUND(tmp, dim);
}
else if (expr->expr_type == EXPR_FUNCTION)
Paul
Author: pault Date: Tue Apr 12 19:14:49 2011 New Revision: 172339 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=172339 Log: 2011-04-12 Paul Thomas <pault@gcc.gnu.org> PR fortran/48360 PR fortran/48456 * trans-array.c (get_std_lbound): For derived type variables return array valued component lbound. 2011-04-12 Paul Thomas <pault@gcc.gnu.org> PR fortran/48360 PR fortran/48456 * gfortran.dg/realloc_on_assign_6.f03: New test. Added: trunk/gcc/testsuite/gfortran.dg/realloc_on_assign_6.f03 Modified: trunk/gcc/fortran/ChangeLog trunk/gcc/fortran/trans-array.c trunk/gcc/testsuite/ChangeLog Author: pault Date: Wed Apr 13 18:38:17 2011 New Revision: 172390 URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=172390 Log: 2011-04-13 Paul Thomas <pault@gcc.gnu.org> PR fortran/48360 PR fortran/48456 * trans-array.c (get_std_lbound): For derived type variables return array valued component lbound. 2011-04-13 Paul Thomas <pault@gcc.gnu.org> PR fortran/48360 PR fortran/48456 * gfortran.dg/realloc_on_assign_6.f03: New test. Added: branches/gcc-4_6-branch/gcc/testsuite/gfortran.dg/realloc_on_assign_6.f03 Modified: branches/gcc-4_6-branch/gcc/fortran/ChangeLog branches/gcc-4_6-branch/gcc/fortran/trans-array.c branches/gcc-4_6-branch/gcc/testsuite/ChangeLog Fixed on trunk and 4.6. Thanks for the report Paul |