7 The grammar of Thun is very simple. A Thun expression is zero or more
8 Thun terms separated by blanks. Terms can be integers in decimal
9 notation, Booleans `true` and `false`, lists enclosed by square brackets
10 `[` and `]`, or symbols (names of functions.)
14 term ::= integer | bool | '[' joy ']' | symbol
16 integer ::= 0 | [ '-' ] ('1'...'9') ('0'...'9')*
18 bool ::= 'true' | 'false'
22 char ::= <Any non-space other than '[' and ']'.>
24 Symbols can be composed of any characters except blanks and square
25 brackets. Integers can be prefixed with a minus sign to denote negative
26 numbers. The symbols `true` and `false` are reserved to denote their
27 respective Boolean values.
29 That's it. That's the whole of the grammar.
32 ![Thun Grammar Railroad Diagram](https://git.sr.ht/~sforman/Thun/blob/trunk/docs/html/images/grammar.png)
37 The original Joy has several datatypes (such as strings and sets)
38 but the Thun dialect currently only uses four:
40 * Integers, signed and unbounded by machine word length (they are
41 [bignums](https://en.wikipedia.org/wiki/Arbitrary-precision_arithmetic).)
42 * Boolean values ``true`` and ``false``.
43 * Lists quoted in `[` and `]` brackets.
47 ## Stack, Expression, Dictionary
49 Thun is built around three things: a __stack__ of data items, an
50 __expression__ representing a program to evaluate, and a __dictionary__
56 [stack-based](https://en.wikipedia.org/wiki/Stack-oriented_programming_language).
57 There is a single main __stack__ that holds data items, which can be
58 integers, bools, symbols (names), or sequences of data items enclosed in
59 square brackets (`[` or `]`).
61 We use the terms "stack", "quote", "sequence", "list", and others to mean
62 the same thing: a simple linear datatype that permits certain operations
63 such as iterating and pushing and popping values from (at least) one end.
65 > In describing Joy I have used the term quotation to describe all of the
66 > above, because I needed a word to describe the arguments to combinators
67 > which fulfill the same role in Joy as lambda abstractions (with
68 > variables) fulfill in the more familiar functional languages. I use the
69 > term list for those quotations whose members are what I call literals:
70 > numbers, characters, truth values, sets, strings and other quotations.
71 > All these I call literals because their occurrence in code results in
72 > them being pushed onto the stack. But I also call [London Paris] a
73 > list. So, [dup \*] is a quotation but not a list.
75 From ["A Conversation with Manfred von Thun" w/ Stevan Apter](http://archive.vector.org.uk/art10000350)
79 A Thun __expression__ is just a sequence or list of items. Sequences
80 intended as programs are called "quoted programs". Evaluation proceeds
81 by iterating through the terms in an expression putting all literals
82 (integers, bools, or lists) onto the main stack and executing functions
83 named by symbols as they are encountered. Functions receive the current
84 stack, expression, and dictionary and return the next stack, expression,
89 The __dictionary__ associates symbols (names) with Thun expressions that
90 define the available functions of the Thun system. Together the stack,
91 expression, and dictionary are the entire state of the Thun interpreter.
96 The Thun interpreter is extremely simple. It accepts a stack, an
97 expression, and a dictionary, and it iterates through the expression
98 putting values onto the stack and delegating execution to functions which
99 it looks up in the dictionary.
101 ![Joy Interpreter Flowchart](https://git.sr.ht/~sforman/Thun/blob/trunk/joy_interpreter_flowchart.svg)
103 All control flow works by
104 [Continuation Passing Style](https://en.wikipedia.org/wiki/Continuation-passing_style).
105 __Combinators__ (see below) alter control flow by prepending quoted programs to the pending
106 expression (aka "continuation".)
109 ## Literals, Functions, Combinators
111 Terms in Thun can be categorized into **literals**, simple **functions**
112 that operate on the stack only, and **combinators** that can prepend
113 quoted programs onto the pending expression ("continuation").
117 Literal values (integers, Booleans, lists) are put onto the stack.
118 Literals can be thought of as functions that put accept a stack and
119 return it with the value they denote on top, if you like.
123 Functions take values from the stack and push results onto it. There are
124 a few kinds of functions: math, comparison, list and stack manipulation.
128 __Combinators__ are functions which accept quoted programs on the stack
129 and run them in various ways by prepending them (or not) to the pending
130 expression. These combinators reify specific control-flow patterns (such
131 as `ifte` which is like `if.. then.. else..` in other languages.)
132 Combinators receive the current expession in addition to the stack and
133 return the next expression. They work by changing the pending expression
134 the interpreter is about to execute.
138 Thun has a set of *basis* functions which are implemented in the host
139 language. The rest of functions in the Thun dialect are defined in terms
142 - Combinators: `branch` `dip` `i` `loop`
143 - Stack Chatter: `clear` `dup` `pop` `stack` `swaack` `swap`
144 - List Manipulation: `concat` `cons` `first` `rest`
145 - Math: `+` `-` `*` `/` `%`
146 - Comparison: `<` `>` `>=` `<=` `!=` `<>` `=`
147 - Logic: `truthy` `not`
148 - Programming: `inscribe`
152 Thun can be extended by adding new definitions to the
153 [defs.txt](https://git.sr.ht/~sforman/Thun/tree/trunk/item/implementations/defs.txt)
154 file and rebuilding the binaries. Each line in the file is a definition
155 consisting of the new symbol name followed by an expression for the body
158 The `defs.txt` file is just joy expressions, one per line, that have a
159 symbol followed by the definition for that symbol, e.g.:
163 The definitions form a DAG (Directed Acyclic Graph) (there is actually a
164 cycle in the definition of `genrec` but that's the point, it is a cycle
165 to itself that captures the cyclical nature of recursive definitions.)
167 I don't imagine that people will read `defs.txt` to understand Thun code.
168 Instead people should read the notebooks that derive the functions to
169 understand them. The reference docs should help, and to that end I'd
170 like to cross-link them with the notebooks. The idea is that the docs
171 are the code and the code is just a way to make precise the ideas in the
174 ### Adding Functions to the Dictionary with `inscribe`
176 You can use the `inscribe` command to put new definitions into the
177 dictionary at runtime, but they will not persist after the program ends.
178 The `inscribe` function is the only function that changes the dictionary.
179 It's meant for prototyping. (You could abuse it to make variables by
180 storing "functions" in the dictionary that just contain literal values as
183 [foo bar baz] inscribe
185 This will put a definition for `foo` into the dictionary as `bar baz`.
192 Nothing prevents you from using symbols as data:
197 But there's a potential pitfall: you might accidentally get a "bare"
198 unquoted symbol on the stack:
205 That by itself won't break anything (the stack is just a list.)
206 But if you were to use, say, `dip`, in such a way as to put the symbol
207 back onto the expression, then when the interpreter encounters it, it
208 will attempt to evaluate it, which is almost certainly not what you want.
215 At the very least you get an "Unknown" error, but if the symbol names a
216 function then the interpreter will attempt to evaluate it, probably
219 I don't see an easy way around this. Be careful? It's kind of against
220 the spirit of the thing to just leave a footgun like that laying around,
221 but perhaps in practice it won't come up. (Because writing Thun code by
222 derivation seems to lead to bug-free code, which is the kinda the point.)
225 ### Variations between Interpreters
227 There are several small choices to be made when implementing a Thun
228 interpreter (TODO: make a comprehensive list), for example, the Python
229 interpreter keeps all of its functions in one dictionary but most of the
230 other interpreters have a `case` or `switch` statement for the built-in
231 functions and a separate hash table for definitions. Additionally, of
232 the interpreters that have hash tables most of them check the hash table
233 after the `case` statement. This means that one cannot "shadow" built-in
234 functions is some interpreters. You can `inscribe` them, but the
235 interpreter will not look for them.
237 I haven't yet formally made a decision for how Thun *shall* work.
238 Letting built-ins be shadowed is fun and useful for exploration, and
239 letting them be inviolate is useful for unsurprising behaviour.
241 Another choice is how to handle duplicate definitions in general. Should
242 you be able to reuse a name? Or should `inscribe` throw some sort of
247 --------------------------------------------------
249 Copyright © 2014 - 2023 Simon Forman
251 This file is part of Thun