Sounds great! Here are a few feature requests: The I/O pair ([-1,4,-6],[1,4,6]) could yield "map abs" as a suggestion :) The pairs ('a',False) and ('*',True) could yield "not . isAlpha" etc. I realize that this (at least the composition part) will probably increase complexity beyond what's acceptable... /Jonas Duregård On 18 March 2010 09:13, Paul Brauner wrote:

Hi,

I was looking at hoogle documentation when I remembered that there is some nice, but quite unusable, feature of squeak (smalltalk) which allows you to search function in the library by giving a list of pairs of inputs/ouputs.

When I'm saying that it is quite unusable, I mean that squeak has to try _every_ function, some of which may be very slow to deliver a result, or require some side effects.

But, piggibacking such a feature on top of hoogle would surely be more efficient:

 1. infer types for arguments and outout  2. look for matching functions using google  3. test them

Has anyone tried that before? If not I would be glad to.

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

On Mar 18, 2010, at 9:13 PM, Paul Brauner wrote:

Hi,

I was looking at hoogle documentation when I remembered that there is some nice, but quite unusable, feature of squeak (smalltalk) which allows you to search function in the library by giving a list of pairs of inputs/ouputs.

The MethodFinder lets you _seek_ or search _for_ methods. It can even cope with permuted arguments. The query 2. 8. 4. produces answers 8 / 2 --> 4 8 // 2 --> 4 8 quo: 2 --> 4

When I'm saying that it is quite unusable, I mean that squeak has to try _every_ function, some of which may be very slow to deliver a result, or require some side effects.

Actually, no it _doesn't_ try every function. It has a long set of methods *not* try, whether because they are dangerous or because they are slow. And it is by design quite easy to extend this set. The query MethodFinder methodFor: #( ('time' 'timeOfDay') true) (intended answer #beginsWith: with the arguments swapped) runs in 24 milliseconds. Smalltalk has exception handling and timeouts, so methods that crash on the given arguments aren't a problem, and methods that take a long time are not a problem either. And they would not be a problem for Haskell. Methods that make unpleasant changes to the system state *could* be a problem but so far I haven't found it so; here of course Haskell does very well. I don't know whether the Haskell version of this considers permuting the arguments; perhaps not.