foo (x: (T)) => void {
print "in inference";
}
foo (x: int) => void {
print "in int.";
}
bar (x: (T)) => void {
foo x;
}
entrypoint => void {
bar "bar";
bar 42;
}
This though had a bit of a sticking point. Visual Studio compiles the basic type check as something like this:
IL_0001: ldtoken [mscorlib]System.Int32
IL_0006: call class [mscorlib]System.Type [mscorlib]System.Type::GetTypeFromHandle(valuetype [mscorlib]System.RuntimeTypeHandle)
IL_000b: ldarga.s 'value'
IL_000d: constrained. !!fooT
IL_0013: callvirt instance class [mscorlib]System.Type [mscorlib]System.Object::GetType()
IL_0018: call bool [mscorlib]System.Type::op_Equality(class [mscorlib]System.Type, class [mscorlib]System.Type)
Note this weird constrained. prefix. It is some magic that lets CIL call generic parameters without boxing them. The problem comes with the ldarga.s instruction above it. Instead of pulling out the parameter value like everything else under the sun, this is getting the address of that parameter. If you don't get the address and make that call, you get some nondescript ExecutionEngineException during runtime. Awesome.
The problem for me was that at that point of the compilation process, I don't know the parameter's index since it was abstracted a few layers above. So for now, Tangent just boxes the thing and goes about its business.
I'm not sure what is next on the docket. Perhaps improving this to allow partial specialization, perhaps local variables, perhaps .NET imports. We'll see what strikes my fancy.
No comments:
Post a Comment