Re: type inference for python
skaller <skaller <at> users.sourceforge.net>
2005-02-01 06:58:08 GMT
On Tue, 2005-02-01 at 16:26, Philippe Fremy wrote:
> Hi,
>
> I would like to implement something similar to the type inference of
> ocaml for the python language. I have always found it very impressive
> (although I have only used caml light).
>
> I have no experience with the topic,
I do. I did considerable work on this. The project
was called Vyper. The key idea was to build an
interpreter which was used to initialise all the
dictionaries, then analyse them for type information,
and generate code that used that type information.
I got as far as getting the interpreter to work very well,
running most standard pure python plus some emulations of
popular C extensions, and with reasonable performance.
In addition, I made some extensions, many of which
got into later versions of Python .. including proper
lexical scoping, garbage collection (well, my version
used Ocaml for that .. ) and a few other things.
However, the type analysis was never done, and I lost
interest in the project, primarily because it would
never support continuation passing in the style of
Stackless Python. (And the source is lost now).
A few technical problems also got in the way:
(Continue reading)
Do you have any hints or where I could build up my knowledge (code, books, article, ...) to do that kind of thing. I imagine that it works in a kind of three way pass: 1. analyse all the constraints of the code Ex: def f(a): a.append(1) def g(a): a=a+1; f(a) g('coucou') => a must support append => a must also be an int 2. cross-validate the constraints consistency => inconsistent assertions 3. validate the constraints against reality => g('coucou') will not work
RSS Feed