[Grace-core] Indentation in Grace

[Andrew P. Black]

On 17 Jul 2013, at 16:59 , Andrew P. Black wrote:

I've added in our decisions, as of the telecom on 30 July 2013

> I've just had an enlightening conversation with Michael Adams about indentation-sensitive parsing.  The executive summary is
> 
> 	-- It's much more complicated than I thought
> 	-- depending on th details of the rules that we adopt for Grace, we can probably avoid a lot of the complexity
> 	-- There are a number of decisions that we need to make about what we WANT indentation to do in Grace
> 
> Unfortunately, the details of what sort of indentation we want will affect (to some extent) the combinators that we need to implement indentation.
> 
> The decisions that we need to make include:
> 
> 	- What productions in Grace's grammar start new indentation scopes:
> 		method
> 		object 
> 		class
> 		block
> 		argument-list
> 	  Are there any others?

That's all (for now!)

> 	
> 	- When the indentation increases, does it increase relative to the line on which the indentation scope starts, or relative to the enclosing indentation scope?
> 	  Michael's mail message dated 12 July 2013 08:34 discusses these alternatives.  For example, consider
> 			
> 		object { method doit {
> 		      foobar(x)
> 		    }
> 		}
> 
> 	  Is this legal?   foobar(x) is syntactically inside the method doit, so it should be indented with respect to that method.  It is indented with respect to
> 	  the indentation of the line on which the method starts, but not with respect to the indentation of the enclosing indentation scope, which starts 
> 	  with the keyword method.  
> 
> 	  One problem with line-based indentation is that adding newlines matters, so while the above would be accepted, the following would be rejected:
> 
> 		object { 
> 		             method doit {
> 		      foobar(x)
> 		    }
> 		}

Indentation is with respect to the whole line on which the current indentation scope starts.  

> 
> 	- Another choice is whether to require a newline after the token that starts a new indentation scope, unless the end of the scope is on the same line.
> 	  this would outlaw 
> 
> 		object { method doit {
> 
> 	  altogether, and force us to write
> 		
> 		object { 
> 		    method doit {
> 
> 	  although it would still allow single-line objects and single-line methods if everything fits on one line.   The main objection to this is that it may make it 
> 	  harder to fit code snippets into papers with a hard page limit!   It's a good idea if you like to use proportionally-spaced fonts for your programs 
> 	  (because all indentations are now determined by the number of leading spaces, and all spaces have equal width).  It also means that the 
>  	  distinction between "current indentation scope" and "current line" goes away.

We can't really require a new line for each unit that starts a new indentation scope, because blocks and arg list start a new scope.
> 
> 	- Do we allow semicolons at all, in multi-line layout, or do we require one statement on each line?   So can I write:
> 
> 		method doIt {
> 			stmt1 
> 			stmt2; stmt3
> 			stmt4
> 		}

Yes we allow semicolons.
> 
> 	  If we allow semicolons, is a trailing semicolon allowed?  

Yes, they are allowed.  Horrible, but they should be allowed.

> 	- Do we want to get rid of the braces around class, object and method?  As Michael pointed out, we can't make them optional, because that
>           introduces an ambiguity.

I think that for now, at lead, we keep the braces.


> 	- Are there any productions in Grace's grammar that "forget" about indentation.  An example is Landin's "except possibly for [parenthesized] 
> 	  subsegments" rule, enclosed by Adam's circle-star combinator, which allows
> 
> 	      x + v where
> 	          x = -(
> 	      y + z) + w
> 

There is no such construct in Grace!

> That may be all for now

So all we need to do now is to write the grammar!

	Andrew