Having written the MonadicBang plugin somewhat recently, where I don't care about the exact-print annotations, I do have some snippets in my code that look like this:

AsPat xa name tok pat -> do
tellName name
AsPat xa name tok <$> traverse (liftMaybeT . evacPats) pat

where the `tok` variable only exists to pass along the exact-print annotations unchanged. So in that context, I would have a slight preference for the exact-print annotations being hidden away in the extension points.

However, I think this also illustrates that the cost to clients is quite manageable. Adding this one variable doesn't make the code unreadable - of course, that's assuming exact-print annotations remain special and not just the first in a long list of properties to eventually be added to each node.

Of course, the cost *is* multiplied by the number of pattern matches on AST nodes you have, which could be a lot given the amount of constructors the types have. In my case, it was very few, because I was able to handle the vast majority of constructors generically via the Data.Data instance.

Jakob

On Thu, Jul 13, 2023 at 1:23 PM Simon Peyton Jones <simon.peytonjones@gmail.com> wrote:

Dear GHC developers

Could you please look at #23447 Where should "tokens" live?

In brief, the question is whether we want to have:

data HsExpr p = ....
   | HsLet (XLet p) (HsLocalBinds p) (LHsExpr p)
or

data HsExpr p = ....
   | HsLet (XLet p) (HsToken "let" p)
           (HsLocalBinds p) (HsToken "in" p) (LHsExpr p)

In the former, if a client wants HsTokes to track the precise source locations of the "let" and "in" keywords, they'd have to put it in the TTG extension field; in the latter, this information is in every syntax tree.

At the moment we have some of each, which is not satisfactory. We need to decide a policy and stick to it. If you use HsSyn, HsExpr, HsPat etc, in any way, you should have an opinion. Please do express it. At the moment we have only a few voices so we risk deciding without enough evidence and use-cases.

Comments with specific use-cases and examples would be particularly helpful.

Thanks!

Simon

_______________________________________________
ghc-devs mailing list
ghc-devs@haskell.org
http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs