Thanks Mikael.
The master branch has been updated by Jerry DeLisle <[email protected]>:
https://gcc.gnu.org/g:78d6e3832905ce5e028d8acef21eff08e8b50819
commit r17-5064-g78d6e3832905ce5e028d8acef21eff08e8b50819
Author: Jerry DeLisle <[email protected]>
Date: Wed Jun 10 08:36:59 2026 -0700
fortran: [PR125720] ICE assigning an allocatable character to an
allocatable component
On 10/11/26 2:28 AM, Mikael Morin wrote:
Le 10/10/2026 à 21:57, Jerry DeLisle a écrit :
See attached patch. A structure constructor that gives an explicit-length
allocatable character component an allocatable or deferred-length value
ICEs in fold_convert_loc. gfc_trans_subcomponent_assign falls through to
its final branch, which fold_converts the value, a pointer to the
string, to the component's character type. The patch copies the string
with gfc_trans_string_copy instead, which also truncates or blank-pads
it to the declared length.
The fix is the new BT_CHARACTER branch in gfc_trans_subcomponent_assign
in trans-expr.cc; the rest is the new test. The bug was found with the
Fujitsu compiler test suite; the test is new.
Regression tested on x86_64.
OK for mainline?
Regards,
Jerry
---
fortran: [PR125720] ICE assigning an allocatable character to an allocatable
component
In a structure constructor, copy a value into an explicit-length
allocatable character component with gfc_trans_string_copy instead of
a scalar fold_convert.
Assisted-by: Claude Opus 5.5
PR fortran/125720
gcc/fortran/ChangeLog:
* trans-expr.cc (gfc_trans_subcomponent_assign): Use
gfc_trans_string_copy for an explicit-length allocatable
character component.
gcc/testsuite/ChangeLog:
* gfortran.dg/structure_constructor_19.f90: New test.
OK.
