The following seems to ignore the typespec ("t2::") when calculating how much memory is required to allocate: module m implicit none type t integer :: i end type t type, extends(t) :: t2 integer :: j end type t2 end module m program test use m implicit none integer :: i class(t), save, allocatable :: y(:) allocate (t2 :: y(5)) ! Should malloc 2*4*5 = 40 bytes, mallocs only 20 select type(y) type is (t2) do i = 1, 5 y(i)%i = i ! "Invalid write of size 4" end do end select deallocate(y) ! SIGABRT: Process abort signal. end
Note: It works if one uses a scalar instead of an array.
There is a similar problem for: character(len=:), allocatable :: str(:) allocate (character(len=5) :: str(5)) end * * * To solve this properly, one should put all the logic into gfc_trans_allocate - and avoid duplicating it in gfc_array_init_size (which is called from gfc_array_allocate, which is called from the mentioned gfc_trans_allocate). Note: Also the evaluation of the type-spec length should be done only once for allocate (character(len = f()) :: str(5), str2, str3, str4(10))
Is not this PR supposed to have been fixed (partially) by revision 199528? Revision 199528 Author: burnus Date: Fri May 31 09:41:53 2013 UTC (3 weeks, 6 days ago) Changed paths: 6 Log Message: 2013-05-31 Tobias Burnus <burnus@net-b.de> PR fortran/57456 * trans-array.c (gfc_array_init_size): Use passed type spec, when available. (gfc_array_allocate): Pass typespec on. * trans-array.h (gfc_array_allocate): Update prototype. * trans-stmt.c (gfc_trans_allocate): Pass typespec on. 2013-05-31 Tobias Burnus <burnus@net-b.de> PR fortran/57456 * gfortran.dg/class_array_17.f90: New. Note that the problem in comment #2 is still there at revision 200486.
(In reply to Dominique d'Humieres from comment #3) > Is not this PR supposed to have been fixed (partially) by revision 199528? Yes. The patch fixes the derived-type issue - but it doesn't solve the CHARACTER issue. As written in comment 2, the proper way is use one common code in gfc_trans_allocate, which calculates the byte size, and use it both for scalars (directly in gfc_trans_allocate) and for arrays in gfc_array_allocate -> gfc_array_init_size. (One also should evaluate the LEN= expression only once.)
Remain issue should be fixed with: https://gcc.gnu.org/ml/fortran/2015-03/msg00074.html https://gcc.gnu.org/ml/fortran/2015-03/msg00075.html https://gcc.gnu.org/ml/fortran/2015-03/msg00085.html
Author: vehre Date: Tue Mar 24 10:28:48 2015 New Revision: 221621 URL: https://gcc.gnu.org/viewcvs?rev=221621&root=gcc&view=rev Log: gcc/fortran/ChangeLog 2015-03-24 Andre Vehreschild <vehre@gmx.de> PR fortran/64787 PR fortran/57456 PR fortran/63230 * class.c (gfc_add_component_ref): Free no longer needed ref-chains to prevent memory loss. (find_intrinsic_vtab): For deferred length char arrays or unlimited polymorphic objects, store the size in bytes of one character in the size component of the vtab. * gfortran.h: Added gfc_add_len_component () define. * trans-array.c (gfc_trans_create_temp_array): Switched to new function name for getting a class' vtab's field. (build_class_array_ref): Likewise. (gfc_array_init_size): Using the size information from allocate more consequently now, i.e., the typespec of the entity to allocate is no longer needed. This is to address the last open comment in PR fortran/57456. (gfc_array_allocate): Likewise. (structure_alloc_comps): gfc_copy_class_to_class () needs to know whether the class is unlimited polymorphic. * trans-array.h: Changed interface of gfc_array_allocate () to reflect the no longer needed typespec. * trans-expr.c (gfc_find_and_cut_at_last_class_ref): New. (gfc_reset_len): New. (gfc_get_class_array_ref): Switch to new function name for getting a class' vtab's field. (gfc_copy_class_to_class): Added flag to know whether the class to copy is unlimited polymorphic. Adding _len dependent code then, which calls ->vptr->copy () with four arguments adding the length information ->vptr->copy(from, to, from_len, to_cap). (gfc_conv_procedure_call): Switch to new function name for getting a class' vtab's field. (alloc_scalar_allocatable_for_assignment): Use the string_length as computed by gfc_conv_expr and not the statically backend_decl which may be incorrect when ref-ing. (gfc_trans_assignment_1): Use the string_length variable and not the rse.string_length. The former has been computed more generally. * trans-intrinsic.c (gfc_conv_intrinsic_sizeof): Switch to new function name for getting a class' vtab's field. (gfc_conv_intrinsic_storage_size): Likewise. (gfc_conv_intrinsic_transfer): Likewise. * trans-stmt.c (gfc_trans_allocate): Restructured to evaluate source=expr3 only once before the loop over the objects to allocate, when the objects are not arrays. Doing correct _len initialization and calling of vptr->copy () fixing PR 64787. (gfc_trans_deallocate): Reseting _len to 0, preventing future errors. * trans.c (gfc_build_array_ref): Switch to new function name for getting a class' vtab's field. (gfc_add_comp_finalizer_call): Likewise. * trans.h: Define the prototypes for the gfc_class_vtab_*_get () and gfc_vptr_*_get () functions. Added gfc_find_and_cut_at_last_class_ref () and gfc_reset_len () routine prototype. Added flag to gfc_copy_class_to_class () prototype to signal an unlimited polymorphic entity to copy. gcc/testsuite/ChangeLog 2015-03-24 Andre Vehreschild <vehre@gmx.de> * gfortran.dg/allocate_alloc_opt_13.f90: Added tests for source= and mold= expressions functionality. * gfortran.dg/allocate_class_4.f90: New test. * gfortran.dg/unlimited_polymorphic_20.f90: Added test whether copying an unlimited polymorhpic object containing a char array to another unlimited polymorphic object respects the _len component. * gfortran.dg/unlimited_polymorphic_22.f90: Extended to check whether deferred length char array allocate works, unlimited polymorphic object allocation from a string works and if allocating an array of deferred length strings works. * gfortran.dg/unlimited_polymorphic_24.f03: New test. Added: trunk/gcc/testsuite/gfortran.dg/allocate_class_4.f90 trunk/gcc/testsuite/gfortran.dg/unlimited_polymorphic_24.f03 Modified: trunk/gcc/fortran/ChangeLog trunk/gcc/fortran/class.c trunk/gcc/fortran/gfortran.h trunk/gcc/fortran/trans-array.c trunk/gcc/fortran/trans-array.h trunk/gcc/fortran/trans-expr.c trunk/gcc/fortran/trans-intrinsic.c trunk/gcc/fortran/trans-stmt.c trunk/gcc/fortran/trans.c trunk/gcc/fortran/trans.h trunk/gcc/testsuite/ChangeLog trunk/gcc/testsuite/gfortran.dg/allocate_alloc_opt_13.f90 trunk/gcc/testsuite/gfortran.dg/unlimited_polymorphic_20.f90 trunk/gcc/testsuite/gfortran.dg/unlimited_polymorphic_22.f90
Fixed with commit r221621.