Property initialization doesn't work without space before interpolated string

[Lanayx]

Repro steps

type C() =
    member val Name = "" with get, set

let x = C(Name="123") //works
let y = C(Name=$"123") // doesn't work
let z = C(Name= $"123") // works

Sharplab

Expected behavior

All three cases should work

Actual behavior

Second case doesn't work

Known workarounds

Add space before value

Related information

.NET 8 F# 8

[abonie]

It is not really related to property initialization, simpler repro:

> let x =$"123";;

  let x =$"123";;
  ------^^

stdin(2,7): error FS0035: This construct is deprecated: '$' is not permitted as a character in operator names and is reserved for future use

and not so much to interpolated strings either:

> let x =-1;;

  let x =-1;;
  ------^^

stdin(3,7): error FS0010: Unexpected infix operator in binding. Expected '=' or other token.

[Martin521]

Well, the last one I think is a different thing, and correctly reported as error. Actually, if I think about it, also =$ is probably according to F# syntax recognized as an operator, and correctly rejected. Same thing as let x = 1+-2. A better error message might be in order. (In any case, following the formatting guidelines and using Fantomas is helpful ;-))

[abonie]

@Martin521 it is kind of the same thing, because in both cases lexer thinks it sees INFIX_COMPARE_OP, in one case it throws an error there because $ is not allowed there, in the other case, I am actually not sure when the error is thrown (after lexing I think), but it stems from the way this code got lexed IMO.

And yeah, not quite sure if this is a bug tbh.

[Lanayx]

@abonie Property initialisation is a case where skipping spaces around = makes the most sense. I've met it when writing html dsl, so it looks like <img src="img_girl.jpg"> , note no spaces around equals operator.

[edgarfgp]

I think this can be fixed following the same approach than https://github.com/dotnet/fsharp/pull/15923/files

[edgarfgp]

This is really bad TBH. another glitch that should be addressed

[vzarytovskii]

This is really bad TBH. another glitch that should be addressed

Easier said than done. It is lexed and parsed like an operator, so it needs be "special cased" for binding, fields and properties assignments (also, how should $"foo"=$"bar" be treated in this case?), but I'm pretty sure by the time we parsed it and produced this diagnostics, there's no "way back" to treating it as assignment.

One possible solution is for us to "deprecate" $ for good, removing it from lexing, so it's never treated as part of the operator, but this will need suggestion/discussion.

@auduchinok thoughts?

Also, I think message in Rider is outdated:

[edgarfgp]

Never said it was easy 😅. I played around locally with a similar approach than https://github.com/dotnet/fsharp/pull/15923/files and it "works". But now shows that Name can't be resolved.

[auduchinok]

@auduchinok thoughts?

We could try to special-case this in the parser: do a check after successfully parsing it as an operator and rewrite the resulting tree if it matches this pattern. We do a similar thing in some other cases already.

[vzarytovskii]

@auduchinok thoughts?

We could try to special-case this in the parser: do a check after successfully parsing it as an operator and rewrite the resulting tree if it matches this pattern. We do a similar thing in some other cases already.

It should be relatively safe, since we don't allow defining (=$) operator.