On 02.11.2010, at 10:20, Simon Marlow wrote:
It's not really a question of priority, rather that we don't know of a good way to fix it!
I would not have guessed that there exists a Haskell related problem that cannot immediately be fixed by the ghc headquarters ; ) If I understand correctly, the problem is to keep the selectors recognizable while still performing optimizations that might "destroy" the selector structure. In Bertram's example the resulting expression looked as follows.
l : case q of (_, ys) -> case ys of [] -> [] _:zs -> splitBy' p zs
Is it correct that the selector is not recognizable in this case because the right hand side fo the outermost case expression is not a simple variable but a case expression? It is probably quite naive to assume that the problem can be solved by looking for a structure like case q of (_,ys) -> e where ys is a free variable in e. There are probably cases where further optimizations prevent this. Or do I miss the real problem in identifying selectors? Cheers, Jan