RFC: Generalised dot syntax for postfix application

[ziman]

Here's a more radical idea: .(expr) and .name are postfix applications of expr/name that bind tighter than application.

Some examples:

• printLn user.name.length
• "hello".length.(+3).printLn
• map (.length.show) ["foo", "booo"]
• mySet.(insert "hello").(insert "world") <+> otherSet
• 5 .div 3
• 5 .(div {ty=Int}) 3
• foo .(NS.(>>=) {lin = True}) bar

This patch did not require any changes to the existing code in the stdlib and tests (other than the recent record test).

One issue I see with this is that idiomatic functions put the main arguments last so you can't write myList.map (+1) and you have to write myList.(map (+1)) instead. There is myList.for printLn but that works only in Applicative contexts.

[gallais]

I have a love/hate relationship with this proposal. But mostly love.

[ziman]

If we do this, we should tell the elaborator not to generalise names after dots to avoid foo.n -> T being translated as {0 n : _} -> n foo -> T.

[gallais]

I expect foo is the one that would be implicitly bound as it is the one
in argument position.

[edwinb]

I'm a bit scared of this one, although I do kind of like it... I think... maybe if we are going to have it, it should be hidden behind a %language pragma?

I think we've overdone syntactic trickery and experimentation in the past, but somehow I still want to play with this...

[andylokandy]

I'll feel a bit awkward if they're identical:

• "hello".(printLn . (+3) . length)
• "hello".length.(printLn . (+3))
• "hello".length.(+3).printLn