Compiling Expression Modules?

[rocky]

In Combinatorica 2.0 there are a few compiled modules, and none of them can get compiled.

Here is an example:

RP = Compile[{{n, _Integer}},
             Module[{p = Range[n],i,x,t},
	            Do [x = Random[Integer,{1,i}];
		        t = p[[i]]; p[[i]] = p[[x]]; p[[x]] = t,
		        {i,n,2,-1}
		    ];
	            p
	     ]
        ]

The error message we get back is:

Compile::comperr: Expression Module[{p = Range[n], i, x, t}, Do[x = Random[Integer, {1, i}] ; t = p[[i]] ; p[[i]] = p[[x]] ; p[[x]] = t, {i, n, 2, -1}] ; p] could not be compiled.

A "could not be compiled" message occurs for all of them, so I suspect this is more about Expression Modules than it is about the construct inside that needs compiling.

By the way, what would be a workaround for this?

[mmatera]

Maybe when the expression can not be compiled, Compile should return a pure Function

[rocky]

Maybe when the expression can not be compiled, Compile should return a pure Function

Ok - thanks for doing.

[GarkGarcia]

I'm not sure how compiling works in Mathics, but why can't a module be compiled?

[rocky]

There is definitely something we can do better here. I just don't think anyone has, yet, investigated doing this.

Using Function is just workaround for compatibility now so we can pull in more external packages.

What "Compile" means is a bit weird. Whatever is done in Mathematica probably doesn't work here since this is not Objective C based. We could simply bytecompile the function using Python's compile() function. Or some do involved thing with LLVM, or somehow hook into Cython. And there are probably other possibilities.

[mmatera]

Indeed, the current implementation is based on llvm. We could change it, or define a pymathics that overwrite the default behavior. The problem is by now that there are certain structures that are not implemented. This was just the simpler workaround I could think when the issue comes up

[rocky]

Compiled functions have been improved. The example still fails, but for another reason. Will open an issue for that specifically when it has been narrowed.