On Wed, Dec 14, 2016 at 08:10:10PM -0700, Martin Sebor wrote:
> Do you approve backporting this fix to the 6.x branch?
Yes, but please wait a little bit with that (i.e. only commit it after
6.3 is released next week).
I'm also surprised this code does not care about what TREE_CODE inner is,
handles all kinds of trees that have at least one operand. Generally,
operands of different trees may mean different things.
You could have also fixed the formatting bugs right next to it:
&& 0 < TREE_OPERAND_LENGTH (inner)
This should be && TREE_OPERAND_LENGTH (inner) > 0
&& reject_gcc_builtin (TREE_OPERAND (inner, 0)))
return error_mark_node;
and this line should be indented 2 positions after the above if,
i.e. 2 positions before &&. Plus the whole block uses spaces instead of tabs.
Jakub
