gen: Generate getters for optional primitives

[abhinav]

The majority use case for accessing values of optional primitives is to
dereference it as-is or use the default/zero value if not.

This modifies code generation to include accessor functions for optional
primitive fields which return non-pointer versions of those fields.

So instead of,

var foo string
if s.Foo != nil {
    foo = *s.Foo
}
doSomething(foo)

Users can do,

doSomething(s.GetFoo())

This should also ease migration for users coming from Apache Thrift.

Resolves #295

Depends on #324

[akshayjshah]

😻 I'm very, very happy that we can make it easier to migrate from Apache Thrift.

[akshayjshah]

nice - named return makes the generated code simpler.

[akshayjshah]

Does generating the same type of accessors for non-primitive types introduce much extra complexity?

[abhinav]

No, but you don't gain anything by doing it. Non-primitive types are always passed around as pointers so the accessor will simply return the value of the field. x.Foo will be no different from x.GetFoo(), so users have no reason to use GetFoo.

[akshayjshah]

Ah, makes sense.

[kriskowal]

I think this is overall a win. Some services may fail to compile after an idl fetch in the unlikely event of an existing conflict. I think it’s a worthwhile risk.

[kriskowal]

This is potentially a breaking change, right?

[abhinav]

Yeah, this is potentially breaking but for unlikely conflicts, we've taken the policy of suggesting "use go.name in case of conflict" rather than cutting a 2.0.

From a quick look at all IDLs in our system, I don't see anything this would actually cause conflicts for. When people use "getFoo" in the request struct, "foo" is usually an optional field in the response struct.