"Joel E. Denny" <[EMAIL PROTECTED]> writes: > Thus, after an error during the split parse, it should be sufficient > to traverse only one of these trees in order to call user > destructors on all existing semantic values. Does this makes sense?
It does to me, though I'd also like Paul Hilfinger's opinion. In the meantime I installed the patch you sent in, and added your test case. Thanks again.
