Lexer: Indenting within object literal produces unexpected behavior, unhelpful error message

[FiloSottile]

This code compiles

{ a: 1
b: 2 }

while this other Syntax Errors

{ a: 1
  b: 2 }

(And generally every time you indent inside a {} literal.)

Not sure it is a bug, but I can't find why in the docs.
Test it

[vendethiel]

You have no "base" indentation so that's expected.

[FiloSottile]

I don't think I understand.

Shouldn't this work?

o = { a: 1,
      b: 2 }

[vendethiel]

The first member's indentation is taken as base indent. You can use

a =
 b: c
 D: e

[epidemian]

Shouldn't this work?

o = { a: 1,
      b: 2 }

I personally would expect that to work, especially given that:

a = [ foo
      bar
      baz ]

And, more importantly

o = { foo
      bar
      baz }

Both work as expected (with or without separator commas).

[FiloSottile]

Indeed, I understand what works and what not, but I'm questioning this behavior.

[satyr]

Shouldn't this work?

Would work if there weren't implicit objects. The code in question is rewritten as:

{ a: 1
 {b: 2} }

and becomes invalid.

[vendethiel]

To be clearer this is lexed as

a = {b: c
  {d: e}
}

Whereas {a b} isn't triggering implicit objects rules. (That may be counted as a bug actually, but that's how INDENT obj works)

[FiloSottile]

Ok, makes sense, sorry for the trouble.
(But it's not clear to the newcomer...)

[epidemian]

Why closing this? @satyr and @Nami-Doc gave an explanation of why this happens, but i don't think they suggested that it's not a bug, or that it's a nice behavior of the compiler. I think it's an inconsistent behavior and should be considered a bug. At least the confusing error message should be considered a bug IMO:

o = { a: 1,
      b: 2, 
      c: 3 }

Error: Parse error on line 2: Unexpected '{'

WAT? There's no '{' at line 2! That error message is as useful as "Parse error: please move your code around a bit until it parses correctly", which is basically what i'd do in that situation =P

[FiloSottile]

Fair enough.

I fear the parser issue in unfixable, as it would require ignoring indentation generated objects inside others objects. And there's no reason for such a behavior.

However, a better error message is needed.

[vendethiel]

Something more future-proof would be to display rewritten coffee, maybe

[satyr]

→ #2160

[mhart]

Just chiming in - I've been bitten by this a number of times and there's nothing that I can see that's ambiguous about the syntax.

It also makes wrapping objects quite messy unless you're going to list each property on a separate line - which in a number of cases is overkill.

[xixixao]

Is it unreasonable to require

{ a: b
    c: d
} # for ({a: b({c: d})});

? See from @epidemian's examples that the whitespace is already special cased inside the brackets, because

[ a
  b ] # works
a()
  b() # doesn't

I couldn't find an open issue for:

{ a: b
  b: c } # ({a: b({c: d})});

but I argue it should be fixed alongside this. (following #3269; this issue was closed as a duplicate of the error message issue, not the original syntax issue)

[xixixao]

So I meant to write this:

I'd like to open this and fix the original issue. Even with better error message (#3319) this still seems like a surprising behavior; and to have same rules to #3305.

here, and I see I have already asked to get this reopened.

Also mentioned here: #1769 (comment)

[xixixao]

Also

{ a: b,
  b: c }

should work as well.

[GeoffreyBooth]

Currently

{ a: 1
  b: 2 }

throws

[stdin]:2:6: error: unexpected indentation
  b: 2 }
     ^

which I believe is correct. The b line is indented when it shouldn’t be. It would be better if the caret pointed at the b, but that’s a quibble. The error message is specific enough.

As a general policy I don’t think we should be creating more exceptions where CoffeeScript ignores indentation. We’re a whitespace-significant language, and people just need to get used to that.

If you want the keys to start on the same column, just write it like:

{
  a: 1
  b: 2
}