https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83419
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The difference is in shortcut_cond_expr, which does:
3640 bool then_se = then_ && TREE_SIDE_EFFECTS (then_);
3641 bool else_se = else_ && TREE_SIDE_EFFECTS (else_);
Now, else_ in one case is:
0;
which is NOP_EXPR around INTEGER_CST, with !TREE_SIDE_EFFECTS.
With -gstatement-frontiers, else_ is:
# DEBUG BEGIN STMT;
0;
which is a STATEMENT_LIST, containing
# DEBUG BEGIN STMT;
(which doesn't have TREE_SIDE_EFFECTS) and
0; (neither), but the STATEMENT_LIST holding both does have TREE_SIDE_EFFECTS:
as make_node has:
switch (type)
{
case tcc_statement:
if (code != DEBUG_BEGIN_STMT)
TREE_SIDE_EFFECTS (t) = 1;
break;
So, perhaps we need those:
bool then_se = then_ && TREE_SIDE_EFFECTS (then_);
bool else_se = else_ && TREE_SIDE_EFFECTS (else_);
then_se = then_ && TREE_SIDE_EFFECTS (then_);
else_se = else_ && TREE_SIDE_EFFECTS (else_);
&& !TREE_SIDE_EFFECTS (then_)
&& !generic_expr_could_trap_p (then_)
&& !TREE_SIDE_EFFECTS (else_)
&& !generic_expr_could_trap_p (else_))
replace with a function that will ignore TREE_SIDE_EFFECTS on STATEMENT_LIST
and instead check the statements in it?
