https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81111

--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Started with r249158.  Note, I see a bootstrap failure even with this patch on
i686-linux:
0x90ad32a make_decl_rtl(tree_node*)
        ../../gcc/varasm.c:1307
0x88a01be expand_expr_real_1(tree_node*, rtx_def*, machine_mode,
expand_modifier, rtx_def**, bool)
        ../../gcc/expr.c:9816
0x889954a expand_expr_real(tree_node*, rtx_def*, machine_mode, expand_modifier,
rtx_def**, bool)
        ../../gcc/expr.c:8074
0x887aaa3 expand_expr
        ../../gcc/expr.h:276
0x888db6c expand_assignment(tree_node*, tree_node*, bool)
        ../../gcc/expr.c:5259
0x871e1ee expand_gimple_stmt_1
        ../../gcc/cfgexpand.c:3645
0x871e602 expand_gimple_stmt
        ../../gcc/cfgexpand.c:3741
0x8726651 expand_gimple_basic_block
        ../../gcc/cfgexpand.c:5745
0x8727e44 execute
        ../../gcc/cfgexpand.c:6354
Former (flag_sanitize & XXX) && do_ubsan_in_current_function () is certainly
not equivalent to sanitize_flags_p (XXX), but to sanitize_flags_p (XXX) &&
current_function_decl.  The weird thing is that current_function_decl is
non-NULL in the above ICE.

Reply via email to