Hi Harald, The Linaro people caught that as well. Thanks.
Interestingly, I was about to re-submit the patch for PR113363, in which all the invalid accesses and memory leaks are fixed but requires this patch to do so. The final transfer was thrown in because it seemed to be working out of the box but should be checked anyway. Inserting your print statements, my test shows the difference in size(chr_a) but prints chr_a as "abcdefgjj" no matter what I do. Needless to say, the latter was the only check that I did. The problem, I suspect, lies somewhere in the murky depths of trans-array.cc(gfc_alloc_allocatable_for_assignment) or in the array part of intrinsic_transfer, untouched by either patch, and is present in 13- and 14-branches. I am onto it. Cheers Paul On Wed, 8 May 2024 at 22:06, Harald Anlauf <anl...@gmx.de> wrote: > Hi Paul, > > this looks mostly good, but the new testcase transfer_class_4.f90 > does exhibit a problem with your patch. Run it with valgrind, > or with -fcheck=bounds, or with -fsanitize=address, or add the > following around the final transfer: > > print *, storage_size (star_a), storage_size (chr_a), size (chr_a), len > (chr_a) > chr_a = transfer (star_a, chr_a) > print *, storage_size (star_a), storage_size (chr_a), size (chr_a), len > (chr_a) > print *, ">", chr_a, "<" > > This prints for me: > > 40 40 2 5$ > 40 40 4 5$ > >abcdefghij^@^@^@^@^@^@^@^@^@^@<$ > > So since the physical representation of chr_a is sufficient > to hold star_a (F2023:16.9.212), no reallocation with a wrong > calculated size should happen. (Intel and NAG get this right.) > > Can you check again? > > Thanks, > Harald > > > Am 08.05.24 um 17:01 schrieb Paul Richard Thomas: > > This fix is straightforward and described by the ChangeLog. Jose Rui > > Faustino de Sousa posted the same fix for the ICE on the fortran list > > slightly more than three years ago. Thinking that he had commit rights, I > > deferred but, regrettably, the patch was never applied. The attached > patch > > also fixes storage_size and transfer for unlimited polymorphic arguments > > with character payloads. > > > > OK for mainline and backporting after a reasonable interval? > > > > Paul > > > > Fortran: Unlimited polymorphic intrinsic function arguments [PR84006] > > > > 2024-05-08 Paul Thomas <pa...@gcc.gnu.org> > > > > gcc/fortran > > PR fortran/84006 > > PR fortran/100027 > > PR fortran/98534 > > * trans-expr.cc (gfc_resize_class_size_with_len): Use the fold > > even if a block is not available in which to fix the result. > > (trans_class_assignment): Enable correct assignment of > > character expressions to unlimited polymorphic variables using > > lhs _len field and rse string_length. > > * trans-intrinsic.cc (gfc_conv_intrinsic_storage_size): Extract > > the class expression so that the unlimited polymorphic class > > expression can be used in gfc_resize_class_size_with_len to > > obtain the storage size for character payloads. Guard the use > > of GFC_DECL_SAVED_DESCRIPTOR by testing for DECL_LANG_SPECIFIC > > to prevent the ICE. Also, invert the order to use the class > > expression extracted from the argument. > > (gfc_conv_intrinsic_transfer): In same way as 'storage_size', > > use the _len field to obtaining the correct length for arg 1. > > > > gcc/testsuite/ > > PR fortran/84006 > > PR fortran/100027 > > * gfortran.dg/storage_size_7.f90: New test. > > > > PR fortran/98534 > > * gfortran.dg/transfer_class_4.f90: New test. > > > >
