Sub-block terminator can cause parent block termination

[ejones]

This occurs, apparently, when the line that terminates the sub-block expression is indented by exactly one space EDIT: an odd number of spaces...? from the initial indentation of the parent block, and the contents of the sub-block are indented past that, and the line following the terminator is at the original indentation:

# These don't work
->
  (
    a
   )
  b
->
  (
      a
     )
  b
###
(function() {
  return a;
});
b;
###

# These work correctly:
->
  (
    a
  )
  b
->
  (
     a
    )
  b
->
  (
    a
   )
   b
###
(function() {
  a;
  return b;
});
###

[satyr]

exactly one space

Any amount of under-dedentation works actually:

$ coffee -bsp
->
  ->
      a
    b
  c

(function() {
  (function() {
    return a;
  });
  return b;
});
c;

Unlike Python, Coffee doesn't force {in,de}dent counts to match.

[michaelficarra]

Unlike Python, Coffee doesn't force {in,de}dent counts to match.

I don't really see how this behaviour is helpful. Its purpose has always confused me and I have never (intentionally) used it. Is it time for a change?

[jashkenas]

Right -- if the dedent doesn't quite make it out to the level of the previous indent, it counts as if it had. This is just to make it a little more forgiving -- but if the less strict syntax causes more problems than it solves, perhaps we should remove it.

I can guarantee we'll have a bunch of folks come out of the woodwork with freshly broken programs if we make that change, however.

[michaelficarra]

It would be something for which I'd be okay holding out until 2.0 so some backwards-incompatibilities are expected. It's not immediately problematic.

[ejones]

@satyr - as of the version on coffeescript.org and npm's coffee-script, your example actually gets processed correctly:

(function() {
  (function() {
    return a;
  });
  b;
  return c;
});

I did discover, though, that in these cases, its not specifically one but rather any odd number of spaces.

Although I acknowledge the non-matcing indent-dedent rule (and disagree with it), the parser's behaviour is nevertheless in these cases inconsistent, or seemingly so.

[autotelicum]

A warning of odd/insufficient spacing would be backwards compatible (cf. examples in closed issue #1512).