[Patch, fortran] PR64290 - [F03] No finalization at deallocation of LHS

[Paul Richard Thomas]

Hi All,

This patch was triggered by a thread on clf. Some years ago Tobias and I
discussed the remaining conditions where finalization should be triggered
and is not. Intrinsic assignment was one of the glaring omissions for which
implementation looked like a heavy lift job. As it happens, it wasn't too
bad :-)

Most of the work was suppressing partial finalization, as a prelude to
reallocation on assignment, and ensuring that finalization happened in the
right circumstances. gfc_assignment_finalizer_call does the work for
intrinsic assignment and is straightforward. Care has to be taken to place
the result between evaluation of the rhs and any reallocation of the lhs
that might occur.

I thought it to be a good idea to squeeze this in before Stage 4 and so the
testcase is not yet finished.I will post it separately once complete and
before pushing the patch. The process is a bit tedious since it involves
checking that the finalization is occurring at the correct point in the
assignment, that the results are consistent with my understanding of
7.5.6.3 and that another brand gives the same results.

Regtests on FC33/x86_64 - OK for master? It occurs to me that this should
also be backported to the 10-branch at very least.

Paul

Fortran:Implement finalization on intrinsic assignment [PR64290]

2021-01-14  Paul Thomas  <pault@gcc.gnu.org>

gcc/fortran
PR fortran/64290
* resolve.c (resolve_where, gfc_resolve_where_code_in_forall,
gfc_resolve_forall_body, gfc_resolve_code): Check that the op
code is still EXEC_ASSIGN. If it is set lhs to must finalize.
* trans-array.c (structure_alloc_comps): Add boolean argument
to suppress finalization and use it for calls from
gfc_deallocate_alloc_comp_no_caf. Otherwise it defaults to
false.
(gfc_alloc_allocatable_for_assignment): Suppress finalization
by setting new arg in call to gfc_deallocate_alloc_comp_no_caf.
* trans-array.h : Add the new boolean argument to the prototype
of gfc_deallocate_alloc_comp_no_caf with a default of false.
* trans-expr.c (gfc_trans_scalar_assign): Suppress finalization
by setting new arg in call to gfc_deallocate_alloc_comp_no_caf.
(gfc_assignment_finalizer_call): New function to provide
finalization on intrinsic assignment.
(gfc_trans_assignment_1): Call it and add the block between the
rhs evaluation and any reallocation on assignment that there
might be.

gcc/testsuite/
PR fortran/64290
* gfortran.dg/finalize_38.f90 : New test.
* gfortran.dg/allocate_with_source_16.f90 : The number of final
calls goes down from 6 to 4.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: submit.diff
Type: text/x-patch
Size: 16595 bytes
Desc: not available
URL: <https://gcc.gnu.org/pipermail/gcc-patches/attachments/20210114/016d3666/attachment-0002.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: finalize_38.f90
Type: text/x-fortran
Size: 2299 bytes
Desc: not available
URL: <https://gcc.gnu.org/pipermail/gcc-patches/attachments/20210114/016d3666/attachment-0003.bin>