Data Driven / Configurable Translator Architecture

[jjtolton]

Let's imagine you have a translation task to translate string S into data structure D. For the sake of imagination, let's imagine that S is a binary encoded string similar to (for example) bencode, except that you can configure quasi-types with some configuration C. So for instance, you could specify perhaps a functor with a name, arrity 2, and arg 0 would be a string and arg 1 would be yet another nested functor. How this would be serialized linearly, something like:

4=foo %% 4 is code for functor/group
7=2  %% 7 is code for arrity, which = 2
12=X %% 12 is code for string
17=96 %% maybe upper bits are indicating functor and lower bits indicate offset where functor info is located? not sure

and maybe that would all be packed together with a header, length, data for binary serialization.

This is just an example of a possible S. Let's say the desired D was a valid Prolog term. And let's say that there was a configuration C that communicated the valid quasi-types to be parsed from S. In fact, C is really a quasi-grammar.

So, normally, if I were writing a parser, the DCGs themselves would be the grammar. But in this case, I'm wondering how you would write a generic parser with a configurable grammar. Does anyone have any ideas or resources on this?

[hurufu]

Compile your quasi grammar to DCG rules and then use those, or do you need something even more dynamic?

[jjtolton]

Ah, no that's a great idea! Could either compile it or interpret it. That seems like a really clean approach!

[hurufu]

An interesting problem would be to infer matching grammar (C) from a small sample of different S and D. It would require inductive logic programming.

[jjtolton]

Ah that is very interesting and would be really helpful to a project I am currently working on. Might need to look into that!