Re: C++ PATCH: perform_implicit_conversion

[Mark Mitchell <mark at codesourcery dot com>]

>>>>> "Nathan" == Nathan Sidwell <nathan@acm.org> writes:

    Nathan> That would be optimizing the compiler on the (hopefully)
    Nathan> rare failure path to the detriment of the frequent success
    Nathan> path. Ok, not by much.  implicit_conversion copes with an
    Nathan> error_mark_node and returns NULL. (I should have said
    Nathan> that)

Good point.  But, we're so far from having any idea where the
performance hot spots are in the C++ front-end (and I bet you they're
not in implicit_conversion), that I don't think we need to worry about
this.  I think simple code is the way to the go, until the compiler is
considerably more correct than it is now.  Most routines should just
return error_mark_node when confronted with it; that will cause to
quickly unwind to the next statement and keep going sensibly.

    Nathan> Would you still prefer early checking for error_mark_node?

Yes.

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com