I was recently working on an type inference algorithm and to test it I did the following:
  Used the quickcheck Arbitrary typeclass to generate expressions
  Inferred types of the expressions using my algorithm,
  converted the expressions that passed inference to haskell and wrote them to a file (without type signatures)
  converted and wrote the failed expression to a separate file
  compiled the "passed" file with -Wall, and extracted all the type signatures that were spit out as warnings
  parsed the type signatures that -Wall spit out and compare with the signatures generated by my algorithm.
  compiled the "failed" file and make sure I get type errors for all my expressions

My algorithm also infers infinite types (which haskell does not) so I had to test that functionality manually.

Overall it was kinda messy, but it worked ok.

I could possibly send you one of my lists. My test expressions are all very simple, no type classes, only 2 types (function and number), and the only expression components are let, lambda, apply, identifier, and number. If something like that would work, let me know.

Hope that helps,

- Job


On Tue, Oct 20, 2009 at 10:07 AM, Peter Verswyvelen <bugfact@gmail.com> wrote:
For learning, I would like to develop my own implementation of type
inference, based on the paper "Typing Haskell in Haskell".

At first sight, the source code of THIH contains a small number of
tests, but I was wandering if a large test set exist?

Thanks,
Peter
_______________________________________________
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe