Yet another PEG parser combinator library in Python. Selling points:
The optional concrete syntax for grammars incorporates semantic actions in a concise host-language-independent way. A Parson grammar won't tie you to Python.
Whole grammars can be analyzed and compiled, even if built at runtime using combinators. (Contrast with a monadic library, where this is uncomputable.)
Semantic actions take and return values in a kind of point-free style.
You can use the concrete syntax with about as little ceremony as re.match.
You can parse non-string sequences.
Anti-selling points:
This library's in fluid design still, undocumented, utterly untuned, etc. I'd like you to use it if you think you might give feedback on the design; otherwise, no promises.
Semantic actions work in a nontraditional way that may remind you of Forth and which I haven't yet tried to make play well in typed languages like Haskell. It's concise and just right for parsing, but maybe in the end it'll turn out too cute and make me rip it out if I want this to be used.
I don't intend to make grammars work in other host languages before the design settles. (I have done this a bit for the Peglet library, a more basic and settled expression of the same approach to actions: it has Python and JavaScript ports.)
I guess the most similar library out there is LPEG, and that's way way more polished.
For now, see all the eg_whatever.py files here. eg_calc.py, eg_misc.py, eg_wc.py, and eg_regex.py have the smallest ones. eg_trees.py shows parsing of tree structures, OMeta-style. Other examples include programming languages and other somewhat-bigger stuff.
Basic things still to explain:
Examples of where I've used it for more than examples:
There's a way to make a grammar automatically skip whitespace and comments and such ('FNORD' rules), which probably should be done differently.
It should be made easy to use with a separate lexer, and I haven't tried this enough to say it's ready (it's probably not).
It should also be easy to write a 'real' compiler, where source-location info gets added to all the AST nodes or whatever representation you're building. This is doable but should be more automated.
After these design issues, this ought to be ported to a different-enough language to bring out issues of working nicely with multiple languages.
After that, I think it'd be time to tackle quality of implementation.
RetroSearch is an open source project built by @garambo | Open a GitHub Issue
Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo
HTML:
3.2
| Encoding:
UTF-8
| Version:
0.7.4