Whoops, the definition of chainl1 should of course be chainl1 p op = p >>= rest where rest x = do f <- op y <- p rest (f x y) +++ return x i.e. rest (f x y), rather than return (f x y). Maybe I should write a test suite too, while I'm at it. Is there any specific location in the cvs tree where test suites should be put?

John Meacham wrote:

2. give Fail a String argument and have it preserve the string passed to 'fail'.

I don't think either will adversly affect performance,

I haven't studied the code for readP, but from experience I can tell that one needs to be very careful to not affect performance when adding error messages. For example, if one wants to combine failure messages from different alternatives, you could force too much evaluation. i.e. When adding this, be very careful not to break the laziness of the current implementation. -- Daan.

John Meacham writes:

possible to use ReadP as a simple standard (as in comes with the libraries) very lightweight parsing monad

Errm, "comes with the libraries" does not necessarily make it standard. As the file header says, Text.ParserCombinators.ReadP is non-portable because it uses local universal quantification. Martin Sjögren writes:

While we're discussing changes, is there an overwhelmingly good reason for the use of local quantification beyond making the type signatures simpler? Writing newtype ReadP r a = R ((a -> P r) -> P r) (like in ContT) would make ReadP completely Haskell98 as far as I can tell.

I think you'll find that without the local universal quantifier, you cannot make ReadP a (useful) instance of Monad, because your 'r' parameter would be fixed across >>=, whereas it really needs to be variable. Regards, Malcolm

Martin Sjögren writes:

...

I think you'll find that without the local universal quantifier, you cannot make ReadP a (useful) instance of Monad, because your 'r' parameter would be fixed across >>=, whereas it really needs to be variable.

Uhh. But it works for ContT, so why not for ReadP? The only point time you have to fix the r is when you want to "run" the parser, all other parsers will be polymorphic in r:

OK, cool. If you can successfully convert ReadP to Haskell'98, we'd all be delighted. But Koen and Simon PJ are the people most likely to be able to give relevant comment on whether any subtle properties will be lost. Regards, Malcolm