On 11/04/2013 05:03 PM, David Malcolm wrote:
I believe it would be easiest to do them one subclass at-a-time e.g.
convert all gimple_call_* functions to accepting a gimple_call, then
another subclass etc - assuming that this approach is blessed by the
core reviewers.
I think the strategy for reworking the gimple stmts needs quite a bit of
experimenting to determine what would have to be done and what it would
look like, and to get buy in from more people that the end result is
indeed acceptable. There are complex situation, and I dont see how we
can avoid virtual functions in some cases (at least not easily without
resorting to ugly as_a/is_a/dyn_cast functions). So I thinkit needs
research.
Andrew
