Fix error when parsing let x=-.1; and co.

[anmonteiro]

This is a follow-up to #1642 employing the strategy highlighted in the following
comment: #1642 (comment)

It removes the special case around LABEL_WITH_EQUAL, making other valid inputs
parseable too.

cc @IwanKaramazow

[IwanKaramazow]

Now that I think of it, can you add a test to verify that =+ etc. as infix operators are still usable?

[anmonteiro]

Yeah I think I might have broken those.

I'll add tests and make sure they're working. Thanks for the feedback

[anmonteiro]

This is actually trickier than I envisioned, and possibly even unsolvable.

Imagine the following:

let (=+) = (a, b) => a + b;
let (=+-) = (a, b) => b;

let foo = a =+-b;

now what is really foo? It could either be parsed as a =+- b or a =+ (-b)

[anmonteiro]

I think the answer to my question is infix precedence.

I managed to fix this issue in the end and added some regression tests around it.

This was definitely a tricky one, and I think we can make it work!

[anmonteiro]

there's another one around line 190

[chenglou]

Man I wish we can disable userland infix operator one day, and then provide a set of e.g. 5-10 good ones, whose precedence we can hard-code, and whose symbol won't be constrained by the first char.

Anyway, is this diff ready?

[anmonteiro]

I really agree with that "first character" statement.

Yeah this is ready to go. It's unfortunate that it complicates the grammar a little bit but both Iwan and I can't see another immediate way to support this case.

[chenglou]

Alright then, merging this one. Thanks for the hard work! Doesn't seem trivial