Soul of a tiny new machine. More thorough tests → More comprehensible and rewrite-friendly software → More resilient society.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
Kartik K. Agaram 38a9370673 typo in shell Readme 5 months ago
.. typo in shell Readme 5 months ago shell: reliable image rendering 11 months ago
data.limg . 9 months ago rename grapheme to code-point-utf8 8 months ago . 11 months ago rename grapheme to code-point-utf8 8 months ago rename grapheme to code-point-utf8 8 months ago snapshot 1 year ago better error handling on stray commas 11 months ago rename grapheme to code-point-utf8 8 months ago start guessing parentheses based on indentation 1 year ago shell: support loading 128x128px images 11 months ago rename grapheme to code-point-utf8 8 months ago shell: literal images 11 months ago shell: support loading 128x128px images 11 months ago rename grapheme to code-point-utf8 8 months ago rename grapheme to code-point-utf8 8 months ago rename grapheme to code-point-utf8 8 months ago
vimrc.vim 7842 - new directory organization 1 year ago

A prototype shell for the Mu computer

Currently runs a tiny dialect of Lisp. Steps to run it from the top-level:

  1. Build it:
./translate shell/*.mu      # generates code.img

You can now already run it (under emulation):

qemu-system-i386 code.img

But let's add some more 'meat' to play with.

  1. Create a data disk with a library of functions.
dd if=/dev/zero of=data.img count=20160         # approximately 10MB
dd if=shell/data.limg of=data.img conv=notrunc

Equivalently, use tools/image-data.

tools/image-data 10 < shell/data.limg
  1. Run with data disk (and 2GB of RAM):
qemu-system-i386 -m 2G -hda code.img -hdb data.img
screenshot of the Mu shell

The Mu computer has a fixed-size screen, which the shell environment partitions into two major regions, with a context-sensitive menu of keyboard shortcuts along the bottom. (No mouse support at the moment.) On the left, two-thirds of the screen is for editing functions and viewing documentation on available primitives. On the right is a REPL where you can try out expressions and see their output. The REPL also has a little toy screen and keyboard for interactively playing with side effects of expressions.

Try typing in some expressions at the REPL and hitting ctrl-s to see their results. Hit ctrl-m to focus on the ... after a run, and browse how the trace of how the results were computed. Here's a 2-minute demo.

To navigate to a function to edit, hit ctrl-g. The list of functions doubles as a bare-bones “file system,” and functions are rendered in order of most recently touched, like in Tiddlywiki. There is no scrolling anywhere yet.

  1. If your Qemu installation supports them, an -accel argument will speed up emulation. Try -accel help to list your options.

Once you select an accelerator, I recommend also adjusting the responsiveness mask in shell/, which controls how frequently the fake screen updates within the REPL. Smaller values will seem more responsive, larger values will run your programs faster. I like to see the fake screen update about once a second. Some suggested values depending on how fast your Qemu is running:

  • -accel kvm on a T420s running Linux: 0xffff/responsiveness=64k
  • -accel tcg on a 2019 Mac: 0xfff/responsiveness=4k

Putting it all together, here's the command I typically use on Linux:

qemu-system-i386 -m 2G -accel kvm -hda code.img -hdb data.img


The Mu shell is a Lisp under the hood. However, you'll see a lot fewer parentheses than most Lisps because it can often automatically insert them based on indentation.

If you're already used to Lisp and always type in all parens, everything will continue to work. In particular, paren-insertion is disabled inside explicitly added parens. Once Mu sees a (, it stops trying to be smart until it sees a ).

I recommend tastefully only removing parens from top-level (def, mac, define) and control-flow words (if, while, for, etc.) Continue using parens for most real function calls. When in doubt, insert parens.

The rule for when parens are inserted is:

Multi-word lines without leading parens are implicitly grouped with later indented lines

For example:

if (> n 0)      =>      (if (> n 0)
  34                      34)

No indented lines after? Parens go around a single line:

f a             =>      (f a)
f b                     (f b)

Lines with a single word are never wrapped in parens:

def (foo)       =>      (def (foo)
  42                      42)

Lines with a leading paren never get more parens:

def (foo x)     =>      (def (foo x)
  (print x) x             (print x) x)


The Mu shell supports infix operators:

(3 + 1)
=> 4

You don't need spaces around infix operators:

=> 4

Operator precedence is not hardcoded. Instead, there is just one rule: operators surrounded by whitespace have lower precedence than operators that are not.

To see how an expression is parsed, quote it:

=> (+ 3 1)

You can create your own infix operators:

def (a <> b)
  (not (a = b))

To permit arbitrary infix operators, the Mu shell partitions the space of characters (technically Unicode code-points) between operators and regular symbols. As a result, you can't define symbols mixing the two.

=> ((* global) . *)                   # probably not what you want

=> (- (- uppercase char) p)           # probably not what you want

'(char> a p)
=> ((char . >) a p)                   # probably not what you want

Infix operators also work in prefix position:

(+ 3 1)
=> 4

As a special case, operators can be unary. A silly example:

def (+++ x)

=> 5

To pass operators to higher-order functions, wrap them in parens

(map1 (+++) '(1 2 3))
=> (2 3 4)

Known issues

  • No mouse support.

  • Mu currently assumes access to 2GB of RAM. To increase that, modify the definition of Heap in 120allocate.subx, and then modify the -m 2G argument in the Qemu commands above. Mu currently has no virtual memory. If your Heap is too large for RAM, allocating past the end of RAM will succeed. However, accessing addresses not backed by RAM will fail with this error:

    lookup: failed