Symbolics / LISP machines

"Fred N. van Kempen" wrote: oh, don't get me be *started*! :-) there are/where many different lisp machines. On the surface, they were basically all specialized microcoded cpu's which were built to run lisp. Almost all had high resolution b&w displays, mice and large keyboards. Disks and ethernet also. In some sense they where the first "single user workstation". The first machines came from MIT - the CONS and CADR. Later LMI and Symbolics made commericial machines (TI later took over some of LMI's machines). Symbolics went the furthest, reducing the machine to a VLSI cpu which was reasonably quick for it's day (the 'ivory'). The machine in question, a 3650 is a 36 bit microcoded cpu. It has a "front end processor" (a 68000) which is smart enough to initialize the console and boot the microcode from the disk (mostly likely a giant 14" SMD disk, approx 200-400mbytes in size). These machines revolve around a lisp compiler, which reduces the lisp code to macroinstructions. The microcode runs the macroinstructions. They are (on the 36xx) 18 bits wide, so two macroinstructions per 36 bit word. The macroinstruction are close to 1-to-1 for many lisp instructions. There are a *large* number of instructions. Risc this is not. Once the microcode is loaded the FEP will load a "world", which is a essentially off the compiled lisp code need to get going. It then fires up the microcode and off you go. (theses are also called "bands" on other machines). There is a hierarchical file system, and processes, but the way it all works is not like unix. You compile lisp code and it gets loaded into your world. Your world just grows and grows, making heavy use of virtual memory. Everything is there when you need it (and even if you don't need it :-) You can look at everything and check out the source for everything. Don't like it? change it. I'm only scratching the surface. The hardware has many many features which make lisp work better. Each data item is 'tagged' with a 4-8 bit tag which says what type of data it is. So, a function can say "hey, I wanted an integer and you passed me a list". Also, the memory has magic "invisible pointers" which allow for some amount of transparent indirection/virtualization. And lots of support for garbage collection which I won't go into. It's a truely wonderful development machine. You can poke an object and ask what it is and where is came from and pop up the source code. The editor is (essentially) emacs. The network code is very, very smart. An easy to use window system. Very little modality. Crisp b&w display. Extremly active mouse. Command line completion. you name it. And it's all lisp. Even the microcode and PAL's where done in lisp. The PAL code is pretty cool - a nice use of symbolic computing. At the time I think I was using PALASM. Wish I'd had a lisp machine :-) I'll make some quick snaps of my 3640 so you can get an idea. It's basically the size of a small refrigerator. There is a backplane with very large (like 2' x 2') cards packed with TTL logic. I think my machine has about 5 Mwords of memory (drams where small then :-). There are some very cool movies on the web showing the UI in action. I think this works http://lispm.dyndns.org:8000/ Some folks (including myself) are working on various projects to run microcode and macrocode of various lisp machines. The goal is to run the environment on a pc. Nothing yet, but someday... Also, you can buy a commericial emulator for the Symbolics machine which runs on an Alpha. I believe it cost $18k (don't ask me why it's so expensive). There are rumors of one coming out for the G4 (another 64 bit machine). -brad