Hi,

I try to learn D and as a testproject I wanted to implement combinator parsing for D.

That's what I currently have: https://gist.github.com/736456#file_parser.d

It is a simple combinator library including simple matchers, alternatives, sequences, optional, transformations of the results and so on.

included is a unit-test that realizes a very simple expression evaluator (lexing, parsing and evaluation are all put together into one).

My questions are now:
* Starting from this, how should such a module be named?
* In which package structure should it be put?
* How could the language be made more convinient (e.g. I overloaded ~ and | operations to be able to write something like match("super") | match("not super"))?


Next steps for me would be to add support for different input that char[] aka strings.

happy to hear from you

christian

Nice.

Maybe add support for predicated alternatives?

If you *really* want to make it more convenient, you could make a string mixin front for it, so that one could do something like

mixin(maek_parser("
  start -> s1 s2;

  s1 -> (`ab`|`cd`) `ef`;

  s2 -> (`1`|`2`|`3`);
"));

I don't recommend you do that, though, given D's current state of affairs regarding ctfe.

Though a runtime version would be doable.

Instead of adding support for char[], wchar[], string, etc, I would recommend you just make it support a dchar range interface.  This hits all of the above, and then some.

I looked at your code a bit, but I'm lazy. Why is ParseSuccess.fResults a Variant[] ?

Lastly, I can't help you with naming :)

On 12/10/2010 10:58 AM, Christian Köstlin wrote:
> Hi,
>
> I try to learn D and as a testproject I wanted to implement combinator
> parsing for D.
>
> That's what I currently have: https://gist.github.com/736456#file_parser.d
>
> It is a simple combinator library including simple matchers,
> alternatives, sequences, optional, transformations of the results and so
> on.
>
> included is a unit-test that realizes a very simple expression evaluator
> (lexing, parsing and evaluation are all put together into one).
>
> My questions are now:
> * Starting from this, how should such a module be named?
> * In which package structure should it be put?
> * How could the language be made more convinient (e.g. I overloaded ~
> and | operations to be able to write something like match("super") |
> match("not super"))?
>
>
> Next steps for me would be to add support for different input that
> char[] aka strings.
>
> happy to hear from you
>
> christian

thanks for your feedback!

> Maybe add support for predicated alternatives?
what is this :) ?

> If you *really* want to make it more convenient, you could make a string
> mixin front for it, so that one could do something like
>
> mixin(maek_parser("
> start -> s1 s2;
>
> s1 -> (`ab`|`cd`) `ef`;
>
> s2 -> (`1`|`2`|`3`);
> "));
>
> I don't recommend you do that, though, given D's current state of
> affairs regarding ctfe.
>
> Though a runtime version would be doable.
>
> Instead of adding support for char[], wchar[], string, etc, I would
> recommend you just make it support a dchar range interface. This hits
> all of the above, and then some.
yes ... i also think so ... i will try this today (but i want to go first with arrays of objects (templated) because then you could put a lexer in front of the parsers).

> I looked at your code a bit, but I'm lazy. Why is ParseSuccess.fResults
> a Variant[] ?
mhh .. thats one of the things i was not sure of. i wanted to be able to return different types from the parsers. e.g. an integer parser should return an array of ints (or in my case a variantarray containing ints). the use of the variants was the second thing which came to my mind and allowed for mixed returns from the parsers ... e.g. the result of parsing "private int i = 5" could be (by classes) [Modifier, Type, string, int].

> Lastly, I can't help you with naming :)
sure :) i just dont get it right now how people have their sources and libraries organized. because it seems that dsss is not any more for d2?

regards

christian