Felipe Magno de Almeida wrote:
On Sat, Jul 4, 2009 at 2:29 PM, Eric Niebler<eric@boostpro.com> wrote:
There is really only one good reason to specify a grammar when defining a domain, and that is to disable troublesome operator overloads. My suggestion would be to not specify your dsel's grammar up front, but rather only validate full expressions at the point of evaluation. You lose no safety that way.
Defining a grammar loosely is harder than specifying it fully. Doing it fully, all I need to do is: shift(create, table, table_name, columns) and I'm done.
Agreed.
But if I do it for all partial expressions, I might forget something. I'm trying to define a SQL grammar. Having it checked earlier, helps me traversing it later, since I know it is right.
I don't see it. You also know it's right if you check the full expression against your full grammar before you traverse the expression.
Also, it prohibits most expressions right away, giving better error messages.
This is not true. "Checking early" means dropping operator overloads via SFINAE, leading to error messages like "no operator << matches with arguments [complicated type 1] and [complicated type 2]" followed by a so-called candidate list, the list of every operator overload that didn't match. This is usually a huge and impenetrable error. But if you do a late check against the full expression with BOOST_MPL_ASSERT and proto::matches, the error is much shorter and takes users directly to a line of code that looks like this: BOOST_MPL_ASSERT((proto::matches< SqlExpression, SqlGrammar >)); That makes the error much more understandable.
I wouldn't have to check later on if there's a non-sensical terminal anywhere.
create << table << create would have no overloads.
All I need is to traverse a grammar, and define another grammar with all sub-expressions. Since the grammar is not very big, it shouldn't be too hard on the compiler (I hope). Though I'm not very experienced with proto yet.
Is transforming a grammar a really difficult task?
You're setting yourself up to climb a mountain, and I don't think you'll like the view from the top. My $0.02, -- Eric Niebler BoostPro Computing http://www.boostpro.com