In the months since I wrote this essay, it has become largely moot. yEnc has taken over as the main encoding scheme used in binary Usenet groups. A lot of people said, “Come up with something better, then,” and I did just that, even figuring out how to do it within the existing MIME spec. And, as I predicted, despite hundreds of people telling me I was wrong, the predominant response was, “Why should we care? yEnc is already here.”
I was surprised by the rather extraordinary popularity of this essay. I wrote it thinking my audience was a few news admins and some regulars in alt.binaries.news-server-comparison. It ended up on Slashdot, the URL was posted to Usenet probably thousands of times (and still is being posted), and I received more email than I could hope to reply to. If I didn't reply to yours, don't be offended, you're in good company of several thousand others. I did, however, read them all, and will continue to do so.
Anyway, I still think yEnc was the wrong way to do it, but at this point, this essay serves mostly historical purposes. There's nothing to do about it now. My thanks to all the people who supported me. Very surprisingly, the overwhelming majority of end-users, the non-power-users, the weekend downloaders, the modem users, were totally against yEnc. Almost every programmer of news software I heard from agreed with me strongly. But, as always, the loudest voices won, and Usenet goes on.
News administrator Curt Welch posted an excellent message about what's wrong with yEnc, which I have posted here with permission. He says some things better than I have.
The introduction of the yEnc encoding scheme has led to some “interesting dialogue” on Usenet. Unfortunately, with the discussions being fragmented in many newsgroups, the people who will ultimately determine the acceptance (or non-acceptance) of yEnc -- the end-users -- are largely unaware of the issues and problems raised by this situation. yEnc has been passed off as a solution, but the problem was never really defined. I have created this page in the hope that the other side of this issue might be understood.
I have actually seen comments on Usenet, about yEnc, to the effect that “if they're writing the code, they must know what they're doing.” As someone who writes code, I appreciate the vote of confidence, but knowing how to program isn't the same thing as knowing what you're doing. You shouldn't take my word for it either. Read what I have to say and decide for yourself. I do this for a living; I am a news administrator for one of the largest sites on Usenet. That doesn't mean I'm always right, but I think it does at least mean I know what I'm talking about.
The main premise behind yEnc is lower overhead for encoded binaries. The usual methods of encoding a binary for Usenet transmission, uuencode and base64, both result in more than a 33% increase in file size after encoding. yEnc, on the other hand, results in only a tiny overhead, meaning the posts are smaller.
I'm not here trying to tell you that was a bad idea. I think it's a good idea. It doesn't make sense to use inefficient encoding methods on Usenet, because Usenet is very nearly 8-bit clean.
In fact, it was largely my idea. Some time ago, a discussion in the newsgroup alt.binaries.news-server-comparison led me to the idea of a low-overhead binary encoding system. I wrote some code and made some test posts using an encoding system very much like the one used in yEnc. It worked. I didn't go out and get people to start using it right away, though, for reasons which will become clear shortly. And I now regret having done it in public, because the yEnc implementor, Jürgen Helbing, took what I did and turned it into yEnc.
Please note, I am not accusing him of stealing anything. I had no intention of using my idea for any proprietary purpose; in fact, I was and continue to be happy for it to be used freely, and I am probably not even the first one to have thought of it (there were mentions of something similar on the MIME mailing list a number of years ago, though it was never pursued). My objection to yEnc is because it was done poorly, not because it was done by Jürgen, and I certainly have nothing against him. Had he done it right, I would be thanking him right now.
yEnc creates significantly smaller encoded binaries than either uuencode, base64, or binhex. That means faster downloads and faster uploads. It means people using metered Usenet service can download more for the same amount of money. These are good things, of course. So what's the problem?
The biggest technical problems with yEnc can be boiled down to this: it preserves everything that is wrong with uuencoding, by re-inventing things which have already been invented better. It is the result of the logical fallacy, “We must do something; this is something; therefore, we must do this.”
Uuencoding relies on searching for “magic strings” in the message body of a Usenet post. This is unreliable, error-prone, and has already led to problems with certain client software. It is absolutely the wrong way to go about tagging message content, because what you really want is something reliably machine-readable and precisely specified. However, yEnc also relies upon magic strings in the body. There was an excuse for doing it this way when uuencode was invented, but there is none now, because reliable, machine-readable content tagging has already been invented. It's called MIME, and it works.
yEnc also offers a way to detect corrupted files or corrupted parts of a multi-part post. That's fine, but that, too, had already been invented and specified. It's called Content-MD5.
With a uuencoded multi-part post, client software typically uses the Subject line of the post to attempt to determine the filename, and to tell where the segment falls in the sequence. This is obviously a terrible way to do it. Clients must parse the Subject line for commonly-used conventions, and hope it works. Sure, it works out most of the time, but it is imprecise and error prone (especially when spaces are used in filenames), and offers no actual reliable means of reassembling a multipart, in particular if parts need to be reposted. How is the software to know that this message, posted days later by a different person, actually contains part 15 of that 30-part post from two days ago? It can't. It's trial and error.
yEnc continues to use the Subject line for this. It relies upon “yEnc” being in the Subject to find messages which contain a yEnc binaries, and the filename and part indication must be present. The structure is a bit more specified, but it is still a kludgy hack.
And this is also an already-solved problem. A way to identify what a message contains, and to specify the filename and other attributes such as sequence of a multi-part, has already been developed. But yEnc ignores all of that, and instead uses the Subject line method.
The filename specification is horribly imprecise. In its current form, it essentially is restricted to us-ascii characters in order to remain at all reliable. It claims that non-ascii characters may be used, but it recommends that the filename be placed in the Subject line of the message. When non-ascii characters are used in message headers, software currently just has to guess what they mean. Jürgen's filename specification cannot even be used to reliably reproduce his own name. This issue may not be immediately obvious, because people use high-bit characters in message headers frequently and it seems to work, but the fact is that, when it works, it's mere coincidence.
Beyond that, though, there are other issues with filenames. The syntax specified for the Subject line of multipart binaries puts the filename into quotes, but gives no method to specify a filename which happens to contain quotes, which is not uncommon. The specification seems to rely on the principle of “it'll probably just work most of the time.” I don't think that's a very good premise upon which to base a standard.
And the bandwidth savings? That's an illusion. A smaller encoding scheme gives us exactly one benefit: faster downloads and uploads for the users. It is not going to make Usenet smaller. It is not going to allow servers to increase retention. Do you really think people aren't going to post more, if they can do it faster? Of course they are. They're always going to post more, with or without yEnc. And, with yEnc, they are even more likely to post more, because posting the same amount of material will take a shorter time, and because people who can't use yEnc will ask for reposts in uuencode.
The growth of Usenet volume is more or less exponential, and has been for quite some time. So let's just say I'm wrong about people, and they really will post less. Let's say that, overnight, all of the binaries on Usenet start getting posted in yEnc, and people post exactly the same amount they would have posted with uuencode, resulting in less total volume. All you have done, in that far-fetched scenario, is create a one-time volume savings. Usenet will continue to grow at the same rate it has been growing, and after a few months, it will be just as large as it was before. And it will get bigger from there. So all you have done is moved the graph back by a few months. Big deal.
So what's the problem we're trying to solve, again?
At this point, you may be thinking, so what? So what if it's not perfect? We can use it for now, and when something better comes along, we can switch to that. But therein lies the problem.
Programmers of Usenet software now have to implement yEnc in their code. Not just once, either. The specification is, as I write this, up to version 1.3, and there will be future revisions. So everyone has to go back and update their code every time the spec is updated. And they don't just have to change it, they have to continue to support the older spec as well, because updates to a new version won't happen overnight. And because the spec is imprecise, programmers are forced to create and maintain even more ugly code in their software, when they could be spending time making more worthwhile improvements. There is a good reason for new standards to be discussed at length and incorporate feedback from experts -- so that you don't have to keep going back and fixing it. And, even when something better comes along, all that yEnc code can't just go away; it will still be there, and still have to be maintained. People won't stop using yEnc overnight. It would take years to become uncommon.
Meanwhile, the transition creates confusion for the users. People don't know what yEnc is, and they have to re-learn how to download binaries. Users of many newsreaders are forced (for a time) to manually decode posts using external software. But, surprising as it may sound, this is not actually one of my arguments against yEnc. Well, it is, in a way, but not in the way you may be thinking.
A transition period to a new method of binary posting is going to cause confusion and some amount of difficulty. There's no way around it. It's just something you have to live with. No big deal.
But, the problem here is that we are undergoing too much transition period for too little benefit. How many times are the programmers going to go back in and update their yEnc code before they get tired of it? How many times are the users going to deal with the confusion before they get tired of it? What is going to happen if and when someone comes up with a real standard to improve the posting of binaries on Usenet (one which will almost certainly be more difficult to implement)?
Is everyone just going to happily switch again? Or are they going to say, hey, I just figured out this yEnc thing a little while ago, and now you're making me go through this crap again, and for what? Why can't you just leave it alone?
Encoding isn't going to get a whole lot smaller than yEnc. Smaller encoding is the easy part. The problem here isn't that we need a smaller encoding scheme, the problem is that we need a better way to post binaries on Usenet.
But the “sexy” feature for the users is the faster downloads, and they will already have that. A lot of the technical stuff in a “real” standard won't mean a whole lot to the users. They will ask, hey, the last time we went through this crap, we got 30% faster downloads. Are you doing that for us again? No? Get lost, then, we don't care. The programmers will say, hey, you want me to add something else to my code? Wasn't it enough the last time?
Is everybody going to happily go through another, equally difficult period of transition?
So, basically, we are going through all of this for a “standard” which probably has a useful life of only a few years, but will require support for far longer than that, and make the next change more difficult than it needed to be. This is not the way to go about introducing a standard.
yEnc was developed and implemented very quickly. Sure, there was some discussion in news.software.nntp and elsewhere about it, but when people who know what they're talking about pointed out what was wrong with the plan, they were essentially ignored. When Jürgen found that going through an actual standardization process within MIME would take time, he chose to ignore MIME in favor of getting something out there right away. He has bragged in Usenet posts about how quickly yEnc has spread. What was the problem that was so bad that it needed to be solved right now? What was so broken? Nothing. Usenet was working just fine, and people were posting and downloading binaries just fine. Was he more concerned with improving Usenet, or with getting his name on something?
Now, he seems to be planning to update the spec to include a means of using yEnc with MIME, which is the way everyone has told him it should be done. But he says he's going to do it within a few weeks! You can't add something to MIME in a few weeks, and there are good reasons for that. So, in reality, what he may be planning to do is bypass the standards process and simply publish a specification. This is very bad. The problem is that doing yEnc within MIME the way he suggested it will break the current MIME specification, and very likely cause problems with existing software. You cannot add yEnc to MIME without first having two small changes made in the MIME specs. Bypassing the standards process would basically sabotage MIME by making it so that coding to the spec will produce software that doesn't work in the real world.
Why not go through the process of updating MIME? Because, to him, it will take too long. I say, what's the rush? If he has the time to work on this, and the desire to get it done, I think that's wonderful, but doing something which will cause long-term harm is worse than not doing anything at all.
I, and others, have repeatedly pointed out what is wrong with yEnc, and we have been ignored. Unfortunately, it has the big selling point of smaller encoding, so getting users to accept it and to demand support for it from their newsreaders has been remarkably easy, and yEnc has spread across Usenet like a plague.
If you agree with me, what can you do to help? If you are the author of a Usenet newsreader, you probably have to implement yEnc at least for decoding, but you can leave out posting support to try to prevent this from spreading any more. At the very least, if a MIME yEnc specification arrives but bypasses the standards process, please, don't implement it. If you are a user, you can refuse to post in yEnc.
If you still don't agree with me, well, thanks for reading this far. I tried.