> I would be grateful if Haskell implementors could share their opinion
> (and/or experience) of the idea of using one of the following
> linear-time unification algorithms in the type inferencer for any of
> the current Haskell implementations:
>
> - Martelli-Montanari (the linear-time algorithm, which
> I believe was never published)
ACM TOPLAS 4(2):258-282, April 1982.
> - Huet
> - Patterson and Wegman
There's a treatment of Martelli-Montanari/Patterson-Wegmen in
"Efficient Type Inference Using Monads",
Hammond, K.,
Proc. 1991 Glasgow Workshop on Functional Programming,
Portree, August 1991.
Springer-Verlag Workshops in Computer Science, pp. 146--157.
Available from ftp.dcs.glasgow.ac.uk in
pub/glasgow-fp/authors/Kevin_Hammond/Efficient_Type_Inference.ps.Z
or /Type_Inference_Full.ps.Z
Basically, graph unification works well, but the characteristics of
type inference are not the same as those for general unification (e.g.
linear-time unification is not necessarily an issue, because types are
small and shallow).
Regards,
Kevin
