Hillside Logo

On the Design of the UNIX operating System


Home

Articles

Canterbury Tour

Contact form


Why is UNIX successful?
The computer world seems to have gone `UNIX mad', and it is hard to understand why. One good reason is the portability of the system but there must be more to it than that. Most people who use the UNIX system seem to like it even though it is full of idiosyncrasies, is terse to the point of unhelpfulness and consists of a very large number of totally forgettable commands. I think that the success of the system is summed up by the following paragraph.

The UNIX system is successful because the minimum number of keystrokes achieve the maximum effort. In addition, the system says very little to explain errors and relies on the intelligence of the user to deduce reasons for failure.

The statement describes UNIX V6, which we all know is the parent of the UNIX systems running today. History tells us that the guys who designed it did their own typing into the machine. It seems to me that because of this, the main reason that UNIX enjoys/suffers from terse input and output is not through any intellectual design decisions made at some early stage but because the UNIX designers were just bad typists working on slow peripherals.

Let us examine the evidence.

Ancient history
First, there are all those incomprehensible and forgettable two letter commands:
ls, cp, mv, pr, etc, etc. What delight it was to type three letter commands: cat, cdb, and dsw. It is interesting to note that if you were to use the text formatters, it was assumed that you could type. So, to use nroff you had to make five painful keystrokes, admittedly only four different letters were used. Of course, all the commands were in lower case because the designers couldn't find the shift key on the keyboard, never mind the shift-lock.

Secondly, look at the main means of the input of text, another two letter command, ed. Every command in the editor is a single letter, which are just about mnemonic. Every error message is a single letter, `?', which is just about meaningless. (OK folks, I didn't forget about the immortal TMP error message and the fact that `??' means something special.) However, only a very few commands are needed to do basic editing operations. If the user wants to do something complicated, then something complicated must be typed in. This complicated thing is often very short and incomprehensible to anyone other than the author. Is there anything more minimal than regular expressions in ed? Learning to use ed is a continuous voyage of discovery, users find that less and less typing is required as time goes on. What joy it is to finally find out how \( is used. Ed is written in C and is available for hacking, and the alterations made to it by many hackers in many places were thought to be required to make the editor more powerful. That is, you could do more in less keystrokes. So, ed is aimed at bad typists, and it developed to support bad typists.

C is another example of design for bad typists; integer? - no, int; structure? - no, struct; begin/end? - no, `{ }'; `:='? - no `=', after all it's easy to type `=='; long variable names? - yes, but only 8 characters were significant, so it wasn't worth typing any more; and the list could go on and on. C is so terse that it is enigmatic, and was obviously designed for coding by bad typists who are actually typing the code themselves. Most other languages seem to be designed for people who are paid by the number of inches of code they write on paper for other people to input into the machine.

Apart from the famous patented setuid bit, the notion of the use and exploitation of software tools is the main feature of the UNIX system which has become important and perhaps even academically respectable. But, did those guys really sit down over a cup of coffee one day and say ``hey, I've had this really neat idea, let's code a bunch of simple programs onto this experimental file system, provide some way of joining the output of one program to the input of another - and yes, a good name would be UNIX''. Heck no, one of them said to the other ``I'm really up to here with this typing, can't we use the output of that program into the input of this, it'll save me lots of time, which I can spend in playing some really neat games''. So, software tools were born.

So, the evidence is very strong that bad typing played a large role in the design of the UNIX system. With its terse input and terse output, UNIX V6 was a joy to use - I am a bad typist too.

More recent history
A number of improvements were seen in UNIX V7. Notably, better typists were employed to design and implement things. The Bourne shell,
sed and awk are examples of this. Those original guys had got a bit better typists too, there are a few more comments in the kernel and some longer names. C had altered and generally requires more typing. All the new things have long names: typedef, unsigned variables, and the type checking of function returns meant the input of function specifications at the start of the program.

UNIX was ported to a number of machines using the very wordy portable C compiler. However, ports of the system to machines with long names were not successful and the enduring port was to the VAX architecture, because VAX is short enough for everyone to type. And then came the Computer Systems Research Group of the University of California, Berkeley.

It is very fashionable to be rude about UCB systems these days. This rudeness all stems from typing envy, because those people at UCB can really type. It seems that applicants for posts at CSRG, UCB have to pass a typing test to get in; the test consists of typing the full title of the institution without using the delete key.

These really great typists contributed mightily to the development of UNIX. For example, they virtually invented the useful comment in the kernel code. It is suspected that the people who understood the comment You are not expected to understand this don't understand why comments are useful. However, UCB hackers tend to show off their skill somewhat and embarrass others by producing really big programs. Most of the rudeness about UCB system stems from the scorn poured onto these big programs, people seem to forget that other UNIX influences have generated really big programs. For instance, you can't get the original f77 V7 compiler into a PDP-11 unless it has separate I/D space.

AT&T headed towards re-organisation and the oddly named UNIX System III made its appearance. It was odd, too, because it is based on the PWB UNIX system which not many people had used. It didn't feel like UNIX because lots of things had moved and had their names changed. With System V, AT&T tell us that this is THE UNIX operating system and hopes it will rule the world. On initial inspection, System V is well typed and has long manuals. Error messages are now wordy and it is almost possible to say that you don't need the source to find out why something happens the way it does. In truth, the current situation is that the UNIX systems which the outside world can obtain are in the hands of people who employ others to type - or so it seems. Typists have moved into the UNIX world, and those of us who are still bad typists complain that UNIX V6 was the last real UNIX system.

Meanwhile, back in Bell Labs, those guys have conquered the typing problem by designing high quality terminals where all you need to do is point. But that's another story.....
On to The UNIX Cult


Peter Collinson Last change: 13th November 2008