Re: language neutral parser

[Axel Kittenberger]

> You probably mix up the order by the order that the arrays entries are
> computed in the course of the algorithm:

Maybe I'm too dump to see, but isn't the order bison writes out the tables 
currently for C not the only valid order?

> I just describe a general problem when producing programs that should
> translate binary data into written files. Typically, the data is produced
> in an order different than the order by which it should be written. The
> workaround is to produce various containers that can hold the information
> while it is computed, and write the files afterwards. So rather than then
> writing XML and let an XML tool do that, one produces a language that can
> act on the binary data directly.

I think this evolve slowly to a pure religiouse discussions, wheter it is 
better to act on binary data, beeing faster and more effective, but generally 
beeing more specialised and not so portable, and having to write more code, 
against using strings, in-middle formats, sacrificing effectiveness for 
generality, beeing able to use standard tools, having to maintain less code 
yourself.

In example XSL scripts can walk through the node tree multiple times, this 
costs effectivness yes, but you can access the fields in any order, without 
having to worry inside of bison of the final order. This is what I would tell 
as task seperation between some small applications (""the unix way of 
computing"")

And honestly I don't write the bison goals, but from my view bison 
compiletime speed is one of the most less important features. How often is it 
executed? Everytime you compile a compiler thats not much. How long does it 
take? 10 milliseconds? I would not mind if it takes him 100 ms or even a 
whole second, but then put out better code, or be more portable, or is able 
to do more things, or has a more comfortable input file, or detects more 
failures, etc...

- Axel