Re: Front-end

David Abrahams wrote:
> on Sat Jun 30 2007, John Moeller <fishcorn-AT-gmail.com> wrote:
> 
>> Daniel Wallin wrote:
>>> John Moeller wrote:
>>>> I was wondering about the front-end a little bit, and since it would be 
>>>> essentially a DSEL, has any thought been given toward using Proto for 
>>>> the Langbinding front-end?
>>> I don't know which DSEL you're thinking of. 
>> The class_<>, def(), etc.  It's a DSEL that describes a class.
>>
>>> IIRC we use type erasure
>>> everywhere, in which case proto won't help. 
>> Type erasure.  That's the term I was looking for, thank you.  :)
>>
>> There's only type erasure (AFAICT) at the point where xxx binds.  Tuple 
>> transformations and other compile-time operations are used internally in 
>> langbinding.  All I'm saying is that it may be useful.  If you don't 
>> think so, so be it.
> 
> It would be useful if we needed to statically generate
> language-specific bindings from the user's binding code.  However,
> it's our hope to avoid that.

Ok.  I'll try to invent another use for proto outside langbinding.  It's 
just so neat!

>> Additionally, I've been wondering if type erasure is strictly necessary 
>> anyway.  A while back, you and Dave had discussed a system where static 
>> type information was utilized.  
> 
> Static type information will always be "utilized."

Hm.  That was vague, wasn't it?  Let me modify that to say, "utilized to 
determine xxx argument types."  I see now that that goes against the 
"compile-once-bind-anywhere" idea.

>> It seems that the idea of static converter generators was abandoned
>> in favor of type erasure, but I wasn't able to discern why.  
> 
> It enables us to have a single compiled front-end language binding
> that can be used with any number of backend languages.

That makes sense.

Ugh.  The "frontend"/"backend" thing is making me dizzy.  I tried to 
come up with alternatives ("native" to replace "frontend", for example), 
but I couldn't think of any role-independent terms for backend languages.

>> That's the point that I was trying to get at in my other post.  I
>> wanted to revisit the issue, if possible.
> 
> I don't think it's possible to get any advantage out of keeping more
> static information around... but I still have to get to your other
> posts ;-)

Ok.  Sorry if I sounded urgent.  :-)  I know that it takes time.

>>> We might have some parts
>>> that doesn't use type erasure, like the policy specification syntax, but
>>>  I don't think we'd want to do any tree transformations on that anyway..
>> Policy specification syntax?  Are you referring to something in 
>> Boost.Python or in Luabind?
> 
> Something proposed for Langbinding.

Cool!  Let's get it codified.  :-)  Was that one of the discussions I 
found in the archives, and I just didn't absorb it?  And by "policy" do 
you mean Alexandrescu-like policies, or something else?  I know there 
was a debate over the term "policy."

-- 

John Moeller
[email protected]


-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/