Let the Foreigners In!

In Learning Ruby by Extension I claimed that learning a programming by writing extensions was a worthwhile exercise to really get to grips with a language. This is because to write a natural-feel extension you are forced to learn a fair bit about the internals of the language. I was wrong. Well sort of, let me explain.

Last week I bemoaned the poor network programming support in my Common Lisp implementation and so this week I resolved to get my beloved multicast sockets working by breaking the abstraction a little and just implementing the missing socket option call as a foreign function call. I was looking forward to it because I had anticipated that it would teach me something about Lisp. It didn’t. Well at least not in the way I was expecting. Absolutely nothing, and there’s a reason for this.

The foreign function interface for Lisp is at a higher level of abstraction than what I was doing with Ruby or Python. With Python, especially, I was actually creating Python objects and classes within C++ modules that Python would allow me to load as if they were native modules. This meant I had to delve into topics about Python’s garbage collection, threading and more. With Lisp’s CFFI (and all of the other FFI’s I’ve seen so far) things don’t work quite like that because they don’t need to. The language allows for much of the mechanics to be completely abstracted away and become almost transparent to the user. Let’s take a look at this by looking at what I did:

(require 'cffi)
(use-package :cffi)
(defcstruct ip-mreq
  (multiaddr :uint32)
  (interface :uint32))

(defcfun "setsockopt" :int
  (socket :int)
  (level :int)
  (option :int)
  (option-value :pointer)
  (option-length :int))

(defcfun ("inet_addr" inet-addr) :int
  (cp :string))

(defun socket-add-mcast-membership (socket-fd mcast-addr iface-addr)
  (with-foreign-object (mreq 'ip-mreq)
               (setf (foreign-slot-value mreq 'ip-mreq 'multiaddr)
                 (inet-addr mcast-addr)
                     (foreign-slot-value mreq 'ip-mreq 'interface)
                 (inet-addr iface-addr))
               ;; level = 0 = IPPROTO_IP,  option = 12 = IP_ADD_MEMBERSHIP
               (setsockopt socket-fd 0 12 mreq
                   (foreign-type-size 'ip-mreq))))

That's it, no external C modules or makefiles it's all done inside Lisp. I'd make this a bit cleaner before I incorporated it into my application because I've intentionally cut some corners to show the relative simplicity. However, it goes to show how easy it is to do. For example if you look at the inet_addr C function definition it only requires a two-line symbolic-expression to have a callable version inside my program . Indeed this solution is on about the same level of complexity as defining an entry point in .NET which couldn't really be any simpler. And like .NET there's no C 'glue' libraries to write because CFFI is doing all the heavy lifting. All arguments and return values are being marshalled into and out of dynamically allocated intermediate values. Which is pretty neat because it means all I have to do is concentrate on the abstraction. Even better is that CFFI allows the partial definition of structures which is great when you have a large and complicated struct which you only care about some of the values for. But that's another story.

Now if we're going to be fair to Ruby, Ruby actually has two ways to write non-local extensions. The first is to use the supplied C programming interface, similar (but simpler) to how Python does it. The second goes a step further and has a gem that allows you to embed inline C directly into the Ruby code. Although I don't think I could have used it for the tibcorv project (because I needed to support callbacks) that's also pretty impressive. Having said that Ruby can only achieve that by having access to a C compiler.

At the risk of stating the obvious a simple method for the definition and invocation of foreign functions is a requirement, I think, for any language to become and remain successful. The easier it is to get the foreigners in the more possibilities you can exploit because there's a C library somewhere for just about any task you can imagine. However, I qualify my original statement that writing extensions can teach us a lot about a language. More specifically I now think that writing extensions will teach you a lot about a language only if the level of abstraction that you have to work at to get the job done forces you to. And as far as I'm concerned, the more abstract the better.


Talking to Donkeys

Too much choice is challenging, to say the least. Well, it is to me anyway. I get hung up on trying to pick the best choice from a huge list of items and end up doing nothing at all. All too often just making a choice, any choice, would be the best idea. However when you’re writing software and trying to pick from an array of available libraries the penalties of a wrong choice can be amplified later. This penalty comes from having to bet early on your chosen horse, but here’s the kicker, you could well have a lot of capital in the race before you realise you backed a donkey.


For instance, if you choose the wrong core library for the “Next Big Thing” now, you run the risk of tying yourself to an implementation that might have a flaw in it that you can’t fix because it’s not open source or because you don’t have the resources to undo it. So you have to adopt a work-around. Thankfully, most of the time things don’t turn out so bad. But if I stop talking to the donkeys for a minute and listen very carefully I can hear millions of developers happily using Microsoft’s tools and libraries without a care in the world. They don’t have to think about this sort of problem very often because Microsoft thought about it for them and gave them the library as part of the SDK. Those developers are real happy with what they’ve got and who can blame them, it’s like having an older brother who looks out for you. So what if he steals your lunch money, it’s part of his charm.

I don’t want to bash Microsoft here, by the way, I think they have a massively important role to play and I have to admit that some of the more recent tools they have made have been pretty awesome achievements for a gigadollar-company. It seems obvious, to me at least, that the higher Microsoft raises the bar in terms of tools the better it is for pretty much everyone. But the (real?) world outside of Redmond is a little more confusing, I know from (old) experience that when I want to use a SMTP client in Java I’ve got to go through a process of evaluation. In .NET the client that comes free in System.Net.Mail will do just fine, nothing to add. Sure sometimes .NET will sometimes be found lacking, but I’ve been pretty happy with the standard libraries in .NET so far and it’s only when I want something a little unusual that I need to go elsewhere (except for the annoying lack of a built-in Set class but that’s a different story).

Well all I wanted to do this week on my continuing Lisp learning exercise was send a multicast packet from my Common Lisp project. Guess what, network programming in Common Lisp is simply not possible. Gasp. It seems that the Common Lisp standard was set in 1994 and so predated the massive expansion of the internet. So CL projects like trivial-sockets and usocket have been created as CL vendor neutral APIs. Unfortunately, it seems trivial sockets is abandonware so I guess usockets is the way to go. But then, it seems broadcast networking is still work in progress for usocket so it seems fair to say that Common Lisp network programming for multicast will be some time off. In a platform independent way at least.

Even that’s ok. I don’t mind. I’m not producing any world-changing libraries here. Hell I might even pitch in on the usocket effort to help out. But right now I’m only messing about trying out some ideas to see what might fly, it’s what Lisp is good at after all. So I’m happy to use whatever multicast support SBCL has. Crap. They don’t seem to have implemented all the socket options needed . Then I read this very illuminating 2005 paper that lays the CL network programming problem out beautifully but when you scroll down to the “Proposed Solution” it says:

… To be written

If you stop talking to your own donkeys for a second and listen you might be able to hear me laughing. Very faintly.

Ok, I know what I’ll do. I’ll take a look at what I need to do make a foreign-function interface work. Yeah, that’s good, I’ll do that. Crap. I’ve got to chose between UFFI & CFFI. So I already know that I’m going to have to take a look at both to see what’s going on. A short while later and I’m starting to believe that CFFI is the cleaner of the two but then I stumble upon Verrazano and SWIG and my head starts to spin.

At this point it’s hard to be cross or frustrated about anything. This isn’t really anything that I can really blame CL for, it’s just how it is, if time-travel had of been available when the 1994 ANSI Common Lisp standard had been accepted then I’m sure they would have added a sockets layer. I can’t blame OpenSource because I can’t stop applauding it. I bow down to those people that have sacrificed their own personal time to make my life easier / better. I know that the fact that I have to do a little bit of work in the decision making process should make me happy because I have a choice to begin with. And anyway it’s better than having to do the whole thing from scratch myself.

But here’s the thing. When I stop shouting and beating my donkeys for being a bit lame and look up there’s millions of Microsoft developers cheerfully waving at the crazy guy who talks (and occasionally beats) his donkeys. I know if I was one of them, I’d think I was crazy too.

I think we donkey beaters can turn this around. I still believe that in general what I can hold in my hand for free is far more powerful than anything I can buy. The problem is essentially one of continuing improvement and better communication. All I have to do is stop cursing the donkeys and do something about it.