Le 27/08/2026 à 18:50, Jerry D a écrit :
See attached patch. I reworked this one from that posted in the PR some
time ago.
This is a wrong code of the worst kind because gfortran was silently
giving incorrect results.
Regression tested on x86_64.
OK for mainline, we should consider backporting this because of the
nature of the error.
Best regards,
Jerry
OK for mainline and backport.
One micro-optimization nit:
+ tmp = TREE_TYPE (gfc_conv_descriptor_data_get (dest));
You may consider using GFC_TYPE_ARRAY_DATAPTR_TYPE (TREE_TYPE (dest))
instead, which avoids creating new tree nodes.
Thanks for the patch.
