[antlr-interest] Attribute Access in Tree Grammar Actions

[Benjamin Niemann]

Randall R Schulz wrote:

> On Friday 29 June 2007 15:28, Gavin Lambert wrote:
>> At 10:16 30/06/2007, Randall R Schulz wrote:
>>  >atomicSentence
>>  >returns [ Formula f ]
>>  >
>>  >  : equation
>>  >
>>  >    {
>>  >      f = $equation;
>>  >    }
>>  >
>>  >  | atom
>>  >
>>  >    {
>>  >      f = $atom;
>>  >    }
>>  >  ;
>>  >
>>  >
>>  >I get these diagnostics from ANTLR:
>>  >
>>  >error(117): CLIFBuilder.g:1157:3: missing attribute access on
>>  >rule scope: equation
>>  >error(117): CLIFBuilder.g:1162:3: missing attribute access on
>>  >rule scope: atom
>>
>> I think you need to use $equation.value and $atom.value.  Or
>> something like that :)
> 
> Evidently it's "or something like that."
> 
> When I look at the generated source code, equation(), atom() and
> atomicSentence() all declare plain, unadorned Formulas, as their return
> types.
> 
> How to I get at the return values of rules when those rules appear on
> the right-hand-side of another rule?

If equation looks like

equation returns [ Formula f ]
...

you'll have to use "$equation.f" to get at the return value.

>> And I think you're also supposed to say $f rather than just
>> f.  Gives the code generator the opportunity to move things around.
> 
> Hmm... OK. I see that's what I did in an earlier (non-tree) grammar I
> wrote. But "move things around?" Why would that be a good thing? Order
> of execution is kind of important in imperative languages...

Stuff is not moved around, but ANTLR may translate '$f = ...' to 
'f = ...' or 'retval.f = ...', depending on the use of other attributes.
So 'f = ...' may work, but break later on a seemingly unrelated change to
the rule (e.g. if some other rule far away in the grammar used
$atomicSentence.start).

-- 
Benjamin Niemann
Email: pink at odahoda dot de
WWW: http://pink.odahoda.de/