The problem here is that the objc front-end produces trees like:
COND_EXPR<a,BIND_EXPR,d> but since BIND_EXPR is not of the exceptions
in make_node_stat to mark its TREE_SIDE_EFFECTS, we make an empty
statement
for this expression which causes us to just to fail.
This patch marks BIND_EXPR as always having SIDE_EFFECTs as it does of
declaring a variable and containing other code.
OK? Bootstrapped and tested on powerpc-darwin and x86_64-linux-gnu
with no regressions. The testcases are already in the testsuite.
Thanks,
Andrew Pinski
ChangeLog:
* tree.c (make_node_stat): BIND_EXPR always have side-effects.