From: Eric Mahurin <eric.mahurin@...>
Date: 2008-09-24T22:46:44+09:00
Subject: Re: ANTLR Target for Ruby

------=_Part_59476_24056085.1222264011128
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Content-Disposition: inline

On Wed, Sep 24, 2008 at 8:32 AM, James Gray <james@grayproductions.net>wrote:

> On Sep 24, 2008, at 7:23 AM, Eric Mahurin wrote:
>
>  I'm sounds like I
>> won't be able to convince you why a DSL is better than a specific format,
>> and you vice-versa.  A quick summary is: although you lose the exact
>> syntax
>> that you may want, you gain in power and flexibility.  Using Grammar
>> (directly), you could template Grammars (method/lambda could return a
>> Grammar based on Grammar args and other params), use plain ruby for extra
>> control when specifying a Grammar tree, or simply embed Grammar objects in
>> a
>> ruby program (like Regexp objects can be - although they parse the grammar
>> from a string).
>>
>
> When I wrote GhostWheel, I was convinced the DSL was better too.  Then I
> tried using it and changed my mind.  :)
>
> The DSL adds a lot of Ruby noise that makes it harder to think about and
> understand the language being parsed.  It keeps you from focusing on the
> right things.
>
> For example, compare this JSON parser using GhostWheel's DSL:
>
> http://ghostwheel.rubyforge.org/svn/trunk/test/example/tc_json_ruby.rb
>
> and this using a grammar:
>
>
> http://ghostwheel.rubyforge.org/svn/trunk/test/example/tc_json_ghost_wheel.rb
>
> I'm not arguing my grammar is perfect either.  I think it could get even
> more readable.  But have a look at how clear rules like value_space,
> value_content, value, and value_with_array_sep play out.
>
> I also think most of the advantages you list can be gained with non-DSL
> grammars.  I mean, we can always dynamically build up a String in Ruby too
> and that's the dumbest thing I could possibly think of.
>
> Anyway, the two don't have to be exclusive.  As you see, GhostWheel
> supports both.


I think my DSL is still fairly clean because you can simply use variable
assignments for "rules" and use the "+" and "|" operators for sequences and
alternations.  Those help a lot.  Ideally I would use a space instead of "+"
to match BNF, but it doesn't seem that bad.

But, I think I'll also have converters (PEG, BNF, regex) to give the non-DSL
approach.

------=_Part_59476_24056085.1222264011128--