leo/grammar/abnf-grammar.md

Leo Library



Copyright (C) 2021 Aleo Systems Inc.



Author: Alessandro Coglio (acoglio on GitHub)


--------


Format Note
-----------

The ABNF standard requires grammars to consist of lines terminated by CR LF
(i.e. carriage return followed by line feed, DOS/Windows-style),
as explained in the background on ABNF later in this file.
This file's lines are therefore terminated by CR LF.
To avoid losing this requirement across systems,
this file is marked as 'text eol=crlf' in .gitattributes:
this means that the file is textual, enabling visual diffs,
but its lines will always be terminated by CR LF on any system.

Note that this CR LF requirement only applies to the grammar files themselves.
It does not apply to the lines of the languages described by the grammar.
ABNF grammar may describe any kinds of languages,
with any kind of line terminators,
or even without line terminators at all (e.g. for "binary" languages).


--------


Introduction
------------

This file contains an initial draft of
a complete ABNF (Augmented Backus-Naur Form) grammar of Leo.
Background on ABNF is provided later in this file.

The initial motivation for creating an ABNF grammar of Leo was that
we have a formally verified parser of ABNF grammars
(at https://github.com/acl2/acl2/tree/master/books/kestrel/abnf;
also see the paper at https://www.kestrel.edu/people/coglio/vstte18.pdf)
which we have used to parse this ABNF grammar of Leo
into a formal representation, in the ACL2 theorem prover,
of the Leo concrete syntax.
The parsing of this grammar file into an ACL2 representation
happens every time the ACL2 formalization of Leo is built.

In addition to that initial motivation,
this ABNF grammar has now the additional and primary purpose of
providing an official definition of the syntax of Leo
that is both human-readable and machine-readable.
This grammar will be part of the (upcoming) Leo language reference,
of the Leo Language Formal Specification
(i.e. the LaTeX document in the leo-semantics repo),
and of the ACL2 formalization of Leo (which was the initial motivation).
It has also been suggested that it may be used to generate tests.


--------


Background on ABNF
------------------

ABNF is an Internet standard:
see https://www.rfc-editor.org/info/rfc5234
and https://www.rfc-editor.org/info/rfc7405.
It is used to specify the syntax of JSON, HTTP, and other standards.

The following is copied (and "un-LaTeX'd") from the aforementioned paper
(at https://www.kestrel.edu/people/coglio/vstte18.pdf).

ABNF adds conveniences and makes slight modifications
to Backus-Naur Form (BNF),
without going beyond context-free grammars.

Instead of BNF's angle-bracket notation for nonterminals,
ABNF uses case-insensitive names consisting of letters, digits, and dashes,
e.g. HTTP-message and IPv6address.
ABNF includes an angle-bracket notation for prose descriptions,
e.g. <host, see [RFC3986], Section 3.2.2>,
usable as last resort in the definiens of a nonterminal.

While BNF allows arbitrary terminals,
ABNF uses only natural numbers as terminals,
and denotes them via:
(i) binary, decimal, or hexadecimal sequences,
e.g. %b1.11.1010, %d1.3.10, and %x.1.3.A
all denote the string '1 3 10';
(ii) binary, decimal, or hexadecimal ranges,
e.g. %x30-39 denotes any string 'n' with 48 <= n <= 57
(an ASCII digit);
(iii) case-sensitive ASCII strings,
e.g. "Ab" denotes the string '65 98';
and (iv) case-insensitive ASCII strings,
e.g. %i"ab", or just "ab", denotes
any string among
'65 66',
'65 98',
'97 66', and
'97 98'.
ABNF terminals in suitable sets represent ASCII or Unicode characters.

ABNF allows repetition prefixes n*m,
where n and m are natural numbers in decimal notation;
if absent,
n defaults to 0, and
m defaults to infinity.
For example,
1*4HEXDIG denotes one to four HEXDIGs,
*3DIGIT denotes up to three DIGITs, and
1*OCTET denotes one or more OCTETs.
A single n prefix
abbreviates n*n,
e.g. 3DIGIT denotes three DIGITs.

Instead of BNF's |, ABNF uses / to separate alternatives.
Repetition prefixes have precedence over juxtapositions,
which have precedence over /.
Round brackets group things and override the aforementioned precedence rules,
e.g. *(WSP / CRLF WSP) denotes strings
obtained by repeating, zero or more times,
either (i) a WSP or (ii) a CRLF followed by a WSP.
Square brackets also group things but make them optional,
e.g. [":" port] is equivalent to 0*1(":" port).

Instead of BNF's ::=, ABNF uses = to define nonterminals,
and =/ to incrementally add alternatives
to previously defined nonterminals.
For example, the rule BIT = "0" / "1"
is equivalent to BIT = "0" followed by BIT =/ "1".

The syntax of ABNF itself is formally specified in ABNF
(in Section 4 of the aforementioned RFC 5234,
after the syntax and semantics of ABNF
are informally specified in natural language
(in Sections 1, 2, and 3 of the aforementioned RFC 5234).
The syntax rules of ABNF prescribe the ASCII codes allowed for
white space (spaces and horizontal tabs),
line endings (carriage returns followed by line feeds),
and comments (semicolons to line endings).


--------


Structure
---------

Language specifications often define the syntax of their languages via
(i) a lexical grammar that describes how
sequence of characters are parsed into tokens, and
(ii) a syntactic grammar that described how
tokens are parsed into expressions, statements, etc.
(The adjectives 'lexical' and 'syntactic' are
the ones used in the Java language specification;
other terms may be used by other languages,
but the essence is similar.)
The situation is sometimes more complex,
with multiple passes (e.g. Unicode escape processing in Java),
but the division between lexical and syntactic (in the sense above) stands.

In the aforementioned language specifications,
both the lexical and syntactic grammars
are normally written in a context-free grammar notation,
augmented with natural language that may assert, for instance,
that tokenization always takes the longest sequence that constitutes a token.

This dual structure appears to be motivated by the fact that
concerns of white space, line endings, etc.
can be handled by the lexical grammar,
allowing the syntactic grammar to focus on the more important structure.
Handling both aspects in a single context-free grammar may be unwieldy,
so having two grammars provides more clarity and readability.

In contrast, PEG (Parsing Expression Grammar) formalisms like Pest
naturally embody a procedural interpretation
that can handle white space and tokenization in just one manageable grammar.
However, this procedural interpretaion may be sometimes
less clear and readable to humans than context-free rules.
Indeed, context-free grammar are commonly used to documentat languages.

ABNF is a context-free grammar notation,
with no procedural interpretation,
and therefore it makes sense to define
separate lexical and syntactic ABNF grammars for Leo.
Conceptually, the two grammars define two subsequent processing phases,
as detailed below.
However, a parser implementation does not need to perform
two strictly separate phases (in fact, it typically does not),
so long as it produces the same final result.


--------


Operator Precedence
-------------------

We formulate the grammar rules for expressions
in a way that describes the relative precedence of operators,
as often done in language syntax specifications.

For instance, consider the rules



```
multiplicative-expression =
        exponential-expression
      / multiplicative-expression "*" exponential-expression
      / multiplicative-expression "/" exponential-expression
```



```
additive-expression =
        multiplicative-expression
      / additive-expression "+" multiplicative-expression
      / additive-expression "-" multiplicative-expression
```



this rule tells us that the additive operators '+' and '-' have
lower precedence than the multiplicative operators '*' and '/',
and that both the additive and multiplicative operators associate to the left.
This may be best understood via the examples given below.

According to the rules, the expression



```
x + y * z
```



can only be parsed as



```
  +
 / \
x   *
   / \
  y   z
```



and not as



```
    *
   / \
  +   z
 / \
x   y
```



because a multiplicative expression cannot have an additive expression
as first sub-expression, as it would in the second tree above.

Also according to the rules, the expression



```
x + y + z
```



can only be parsed as



```
    +
   / \
  +   z
 / \
x   y
```



and not as



```
  +
 / \
x   +
   / \
  y   z
```



because an additive expression cannot have an additive expression
as second sub-expression, as it would in the second tree above.


--------


Naming Convention
-----------------

For this ABNF grammar, we choose nonterminal names
that consist of complete English words, separated by dashes,
and that describe the construct the way it is in English.
For instance, we use the name 'conditional-statement'
to describe conditional statements.

At the same time, we attempt to establish
a precise and "official" nomenclature for the Leo constructs,
by way of the nonterminal names that define their syntax.
For instance, the rule



```
group-literal = product-group-literal
              / affine-group-literal
```



tells us that there are two kinds of group literals,
namely product group literals and affine group literals.
This is more precise than describing them as
integers (which are not really group elements per se),
or points (they are all points, just differently specified),
or being singletons vs. pairs (which is a bit generic).

The only exception to the nomenclature-establishing role of the grammar
is the fact that, as discussed above,
we write the grammar rules in a way that determines
the relative precedence and the associativity of expression operators,
and that therefore we have rules like



```
unary-expression = primary-expression
                 / "!" unary-expression
                 / "-" unary-expression
```



In order to allow the recursion of the rule to stop,
we need to regard, in the grammar, a primary expression as a unary expression.
However, this is just a grammatical artifact:
ontologically, a primary expression is not really a unary expression,
because a unary expression is one that consists of
a unary operator and an operand sub-expression.
These terminological "exceptions" should be easy to identify in the rules.


--------


Lexical Grammar
---------------

A Leo file is a finite sequence of Unicode characters,
represented as Unicode code points,
which are numbers in the range form 0 to 10FFFFh.
These are captured by the ABNF rule 'character' below.

The lexical grammar defines how, conceptually,
the sequence of characters is turned into
a sequence of tokens, comments, and whitespaces.

As stated, the lexical grammar is ambiguous.
For example, the sequence of characters '**' (i.e. two stars)
could be equally parsed as two '*' symbol tokens or one '**' symbol token
(see rule for 'symbol' below).
As another example, the sequence or characters '<CR><LF>'
(i.e. carriage return followed by line feed)
could be equally parsed as two line terminators or one
(see rule for 'newline').

Thus, as often done in language syntax definitions,
the lexical grammar is disambiguated by
the extra-grammatical requirement that
the longest possible sequence of characters is always parsed.
This way, '**' must be parsed as one '**' symbol token,
and '<CR><LF>' must be parsed as one line terminator.
(We should formalize this requirement,
along with the other extra-grammatical requirements given below,
and formally prove that they indeed make
the lexical grammar of Leo unambiguous.)

As mentioned above, a character is any Unicode code point.
This grammar does not say how those are encoded in files (e.g. UTF-8):
it starts with a decoded sequence of Unicode code points.
Note that we allow any value,
even though some values may not be used according to the Unicode standard.

<a name="character"></a>
```abnf
character = %x0-10FFFF   ; any Unicode code point
```

We give names to certain ASCII characters.

<a name="horizontal-tab"></a>
```abnf
horizontal-tab = %x9
```

<a name="line-feed"></a>
```abnf
line-feed = %xA
```

<a name="carriage-return"></a>
```abnf
carriage-return = %xD
```

<a name="space"></a>
```abnf
space = %x20
```

<a name="double-quote"></a>
```abnf
double-quote = %x22
```

We give names to complements of certain ASCII characters.
These consist of all the Unicode characters except for one or two.

<a name="not-double-quote"></a>
```abnf
not-double-quote = %x0-22 / %x24-10FFFF   ; anything but "
```

<a name="not-star"></a>
```abnf
not-star = %x0-29 / %x2B-10FFFF   ; anything but *
```

<a name="not-line-feed-or-carriage-return"></a>
```abnf
not-line-feed-or-carriage-return = %x0-9 / %xB-C / %xE-10FFFF
                                   ; anything but LF or CR
```

<a name="not-star-or-slash"></a>
```abnf
not-star-or-slash = %x0-29 / %x2B-2E / %x30-10FFFF   ; anything but * or /
```

Lines in Leo may be terminated via
a single carriage return,
a line feed,
or a carriage return immediately followed by a line feed.
Note that the latter combination constitutes a single line terminator,
according to the extra-grammatical rule of the longest sequence.

<a name="newline"></a>
```abnf
newline = line-feed / carriage-return / carriage-return line-feed
```

Go to: _[line-feed](#user-content-line-feed), [carriage-return](#user-content-carriage-return)_;


Line terminators form whitespace, along with spaces and horizontal tabs.

<a name="whitespace"></a>
```abnf
whitespace = space / horizontal-tab / newline
```

Go to: _[horizontal-tab](#user-content-horizontal-tab), [newline](#user-content-newline), [space](#user-content-space)_;


There are two kinds of comments in Leo, as in other languages.
One is block comments of the form '/* ... */',
and the other is end-of-line comments '// ...'.
The first kind start at '/*' and end at the first '*/',
possibly spanning multiple (partial) lines;
they do no nest.
The second kind start at '//' and extend till the end of the line.
The rules about comments given below are similar to
the ones used in the Java language specification.

<a name="comment"></a>
```abnf
comment = block-comment / end-of-line-comment
```

Go to: _[block-comment](#user-content-block-comment), [end-of-line-comment](#user-content-end-of-line-comment)_;


<a name="block-comment"></a>
```abnf
block-comment = "/*" rest-of-block-comment
```

Go to: _[rest-of-block-comment](#user-content-rest-of-block-comment)_;


<a name="rest-of-block-comment"></a>
```abnf
rest-of-block-comment = "*" rest-of-block-comment-after-star
                      / not-star rest-of-block-comment
```

Go to: _[not-star](#user-content-not-star), [rest-of-block-comment-after-star](#user-content-rest-of-block-comment-after-star), [rest-of-block-comment](#user-content-rest-of-block-comment)_;


<a name="rest-of-block-comment-after-star"></a>
```abnf
rest-of-block-comment-after-star = "/"
                                 / "*" rest-of-block-comment-after-star
                                 / not-star-or-slash rest-of-block-comment
```

Go to: _[rest-of-block-comment-after-star](#user-content-rest-of-block-comment-after-star), [rest-of-block-comment](#user-content-rest-of-block-comment), [not-star-or-slash](#user-content-not-star-or-slash)_;


<a name="end-of-line-comment"></a>
```abnf
end-of-line-comment = "//" *not-line-feed-or-carriage-return newline
```

Go to: _[newline](#user-content-newline)_;


Below are the keywords in the Leo language.
They cannot be used as identifiers.

<a name="keyword"></a>
```abnf
keyword = "address"
        / "as"
        / "bool"
        / "circuit"
        / "console"
        / "const"
        / "else"
        / "false"
        / "field"
        / "for"
        / "function"
        / "group"
        / "i8"
        / "i16"
        / "i32"
        / "i64"
        / "i128"
        / "if"
        / "import"
        / "in"
        / "input"
        / "let"
        / "mut"
        / "return"
        / "Self"
        / "self"
        / "static"
        / "string"
        / "true"
        / "u8"
        / "u16"
        / "u32"
        / "u64"
        / "u128"
```

The following rules define (ASCII) digits
and (uppercase and lowercase) letters.

<a name="digit"></a>
```abnf
digit = %x30-39   ; 0-9
```

<a name="uppercase-letter"></a>
```abnf
uppercase-letter = %x41-5A   ; A-Z
```

<a name="lowercase-letter"></a>
```abnf
lowercase-letter = %x61-7A   ; a-z
```

<a name="letter"></a>
```abnf
letter = uppercase-letter / lowercase-letter
```

Go to: _[lowercase-letter](#user-content-lowercase-letter), [uppercase-letter](#user-content-uppercase-letter)_;


An identifier is a non-empty sequence of letters, digits, and underscores,
starting with a letter.
It must not be a keyword: this is an extra-grammatical constraint.

<a name="identifier"></a>
```abnf
identifier = letter *( letter / digit / "_" )   ; but not a keyword
```

Go to: _[letter](#user-content-letter)_;


A package name consists of one or more segments separated by single dashes,
where each segment is a non-empty sequence of lowercase letters and digits.

<a name="package-name"></a>
```abnf
package-name = 1*( lowercase-letter / digit )
               *( "-" 1*( lowercase-letter / digit ) )
```

An address starts with 'aleo1'
and continues with exactly 58 lowercase letters and digits.
Thus an address always consists of 63 characters.

<a name="address"></a>
```abnf
address = "aleo1" 58( lowercase-letter / digit )
```

A format string is a sequence of characters, other than double quote,
surrounded by double quotes.
Within a format string, substrings '{}' are distinguished as containers
(these are the ones that may be matched with values
whose textual representation replaces the containers
in the printed string).
There is an implicit extra-grammatical requirements that
the explicit 'formatted-string-container' instances include
all the occurrences of '{}' in the parsed character sequence:
that is, there may not be two contiguous 'not-double-quote' instances
that are '{' and '}'.

<a name="formatted-string-container"></a>
```abnf
formatted-string-container = "{}"
```

<a name="formatted-string"></a>
```abnf
formatted-string = double-quote
                *( not-double-quote / formatted-string-container )
                double-quote
```

Go to: _[double-quote](#user-content-double-quote)_;


Here is (part of this ABNF comment),
an alternative way to specify format strings,
which captures the extra-grammatical requirement above in the grammar
but is more complicated:



```
not-double-quote-or-open-brace = %x0-22 / %x24-7A / %x7C-10FFFF
```



```
not-double-quote-or-close-brace = %x0-22 / %x24-7C / %x7E-10FFFF
```



```
formatted-string-element = not-double-quote-or-open-brace
                      / "{" not-double-quote-or-close-brace
                      / formatted-string-container
```



```
formatted-string = double-quote *formatted-string-element double-quote
```



It is not immediately clear which approach is better; there are tradeoffs.
Regardless, we should choose one eventually.

Annotations are built out of names and arguments, which are tokens.
Two names are currently supported.
An argument is a sequence of one or more letters, digits, and underscores.

<a name="annotation-name"></a>
```abnf
annotation-name = "@" identifier
```

Go to: _[identifier](#user-content-identifier)_;


A natural (number) is a sequence of one or more digits.
Note that we allow leading zeros, e.g. '007'.

<a name="natural"></a>
```abnf
natural = 1*digit
```

An integer (number) is either a natural or its negation.
Note that we also allow leading zeros in negative numbers, e.g. '-007'.

<a name="integer"></a>
```abnf
integer = [ "-" ] natural
```

Go to: _[natural](#user-content-natural)_;


An untyped literal is just an integer.

<a name="untyped-literal"></a>
```abnf
untyped-literal = integer
```

Go to: _[integer](#user-content-integer)_;


Unsigned literals are naturals followed by unsigned types.

<a name="unsigned-literal"></a>
```abnf
unsigned-literal = natural ( "u8" / "u16" / "u32" / "u64" / "u128" )
```

Go to: _[natural](#user-content-natural)_;


Signed literals are integers followed by signed types.

<a name="signed-literal"></a>
```abnf
signed-literal = integer ( "i8" / "i16" / "i32" / "i64" / "i128" )
```

Go to: _[integer](#user-content-integer)_;


Field literals are integers followed by the type of field elements.

<a name="field-literal"></a>
```abnf
field-literal = integer "field"
```

Go to: _[integer](#user-content-integer)_;


There are two kinds of group literals.
One is a single integer followed by the type of group elements,
which denotes the scalar product of the generator point by the integer.
The other is a pair of integer coordinates,
which are reduced modulo the prime to identify a point,
which must be on the elliptic curve.
It is also allowed to omit one (not both) coordinates,
with an indication of how to infer the missing coordinate
(i.e. sign high, sign low, or inferred).

<a name="product-group-literal"></a>
```abnf
product-group-literal = integer "group"
```

Go to: _[integer](#user-content-integer)_;


<a name="group-coordinate"></a>
```abnf
group-coordinate = integer / "+" / "-" / "_"
```

Go to: _[integer](#user-content-integer)_;


<a name="affine-group-literal"></a>
```abnf
affine-group-literal = "(" group-coordinate "," group-coordinate ")" "group"
```

Go to: _[group-coordinate](#user-content-group-coordinate)_;


<a name="group-literal"></a>
```abnf
group-literal = product-group-literal / affine-group-literal
```

Go to: _[product-group-literal](#user-content-product-group-literal), [affine-group-literal](#user-content-affine-group-literal)_;


Note that the rule for group literals above
allows no whitespace between coordinates.
If we want to allow whitespace,
e.g. '(3, 4)group' as opposed to requiring '(3,4)group',
then we should define affine group literals
in the syntactic grammar instead of in the lexical grammar.
We should have a notion of atomic literal in the lexical grammar,
and (non-atomic) literal in the syntactic grammar.
The lexical grammar should define a token for ')group'
if we want no whitespace between the closing parenthesis and 'group'.
More precisely, the rule for 'literal' below in the lexical grammar
would be replaced with



```
atomic-literal = ... / product-group-literal
```



where the '...' stands for all the '...-literal' alternatives
in the current rule for 'literal' below, except 'group-literal'.
Furthermore, the rule for 'symbol' below in the lexical grammar
would be extended to



```
symbol = ... / ")group"
```



where '...' stands for the current definiens of the rule.
We would also have to adjust the rule for 'token' below in the lexical grammar
to reference 'atomic-literal' instead of 'literal' in the definiens.
We would then add to the syntactic grammar the following rules



```
affine-group-literal = "(" group-coordinate "," group-coordinate ")group"
```



```
literal = atomic-literal / affine-group-literal
```



which would now define literals in the syntactic grammar.
Note that now an affine group literal would have the form



```
( <ow> <coordinate> <ow> , <ow> <coordinate> <ow> )group
```



where <ow> is optional whitespace;
however, no whitespace is allowed between the closing ')' and 'group'.

Boolean literals are the usual two.

<a name="boolean-literal"></a>
```abnf
boolean-literal = "true" / "false"
```

An address literal is an address wrapped into an indication of address,
to differentiate it from an identifier.

<a name="address-literal"></a>
```abnf
address-literal = "address" "(" address ")"
```

Go to: _[address](#user-content-address)_;


The ones above are all the literals, as defined by the following rule.

<a name="literal"></a>
```abnf
literal = untyped-literal
        / unsigned-literal
        / signed-literal
        / field-literal
        / group-literal
        / boolean-literal
        / address-literal
```

Go to: _[unsigned-literal](#user-content-unsigned-literal), [untyped-literal](#user-content-untyped-literal), [field-literal](#user-content-field-literal), [group-literal](#user-content-group-literal), [address-literal](#user-content-address-literal), [boolean-literal](#user-content-boolean-literal), [signed-literal](#user-content-signed-literal)_;


After defining the (mostly) alphanumeric tokens above,
it remains to define tokens for non-alphanumeric symbols such as "+" and "(".
Different programming languages used different terminologies for these,
e.g. operators, separators, punctuators, etc.
Here we use 'symbol', for all of them, but we can do something different.
We could give names to all of these symbols,
via rules such as



```
equality-operator = "=="
```



and defining 'symbol' in terms of those



```
symbol = ... / equality-operator / ...
```



This may or may not make the grammar more readable,
but it would help establish a terminology in the grammar,
namely the exact names of some of these token.
On the other hand, at least some of them are perhaps simple enough
that they could be just described in terms of their symbols,
e.g. 'double dot', 'question mark', etc.

<a name="symbol"></a>
```abnf
symbol = "!" / "&&" / "||"
       / "==" / "!="
       / "<" / "<=" / ">" / ">="
       / "+" / "-" / "*" / "/" / "**"
       / "=" / "+=" / "-=" / "*=" / "/=" / "**="
       / "(" / ")"
       / "[" / "]"
       / "{" / "}"
       / "," / "." / ".." / "..." / ";" / ":" / "::" / "?"
       / "->" / "_"
```

Everything defined above, other than comments and whitespace,
is a token, as defined by the following rule.

<a name="token"></a>
```abnf
token = keyword
      / identifier
      / literal
      / package-name
      / formatted-string
      / annotation-name
      / symbol
```

Go to: _[literal](#user-content-literal), [annotation-name](#user-content-annotation-name), [formatted-string](#user-content-formatted-string), [symbol](#user-content-symbol), [identifier](#user-content-identifier), [keyword](#user-content-keyword), [package-name](#user-content-package-name)_;



--------


Syntactic Grammar
-----------------

The processing defined by the lexical grammar above
turns the initial sequence of characters
into a sequence of tokens, comments, and whitespaces.
The purpose of comments and whitespaces, from a syntactic point of view,
is just to separate tokens:
they are discarded, leaving a sequence of tokens.
The syntactic grammar describes how to turn
a sequence of tokens into concrete syntax trees.

There are unsigned and signed integer types, for five sizes.

<a name="unsigned-type"></a>
```abnf
unsigned-type = "u8" / "u16" / "u32" / "u64" / "u128"
```

<a name="signed-type"></a>
```abnf
signed-type = "i8" / "i16" / "i32" / "i64" / "i128"
```

<a name="integer-type"></a>
```abnf
integer-type = unsigned-type / signed-type
```

Go to: _[unsigned-type](#user-content-unsigned-type), [signed-type](#user-content-signed-type)_;


The integer types, along with the field and group types,
for the arithmetic types, i.e. the ones that support arithmetic operations.

<a name="field-type"></a>
```abnf
field-type = "field"
```

<a name="group-type"></a>
```abnf
group-type = "group"
```

<a name="arithmetic-type"></a>
```abnf
arithmetic-type = integer-type / field-type / group-type
```

Go to: _[field-type](#user-content-field-type), [group-type](#user-content-group-type), [integer-type](#user-content-integer-type)_;


The arithmetic types, along with the boolean and address types,
form the scalar types, i.e. the ones whose values do not contain (sub)values.

<a name="boolean-type"></a>
```abnf
boolean-type = "bool"
```

<a name="address-type"></a>
```abnf
address-type = "address"
```

<a name="scalar-type"></a>
```abnf
scalar-type =  boolean-type / arithmetic-type / address-type
```

Go to: _[arithmetic-type](#user-content-arithmetic-type), [address-type](#user-content-address-type), [boolean-type](#user-content-boolean-type)_;


Circuit types are denoted by identifiers and the keyword 'Self'.
The latter is only allowed inside a circuit definition,
to denote the defined circuit.

<a name="self-type"></a>
```abnf
self-type = "Self"
```

<a name="circuit-type"></a>
```abnf
circuit-type = identifier / self-type
```

Go to: _[self-type](#user-content-self-type), [identifier](#user-content-identifier)_;


A tuple type consists of zero, two, or more component types.

<a name="tuple-type"></a>
```abnf
tuple-type = "(" [ type 1*( "," type ) ] ")"
```

Go to: _[type](#user-content-type)_;


An array type consists of an element type
and an indication of dimensions.
There is either a single dimension (a number),
or a tuple of one or more dimensions
(we could restrict  the latter to two or more dimensions).

<a name="array-type"></a>
```abnf
array-type = "[" type ";" array-dimensions "]"
```

Go to: _[type](#user-content-type), [array-dimensions](#user-content-array-dimensions)_;


<a name="array-dimensions"></a>
```abnf
array-dimensions = natural
                 / "(" natural *( "," natural ) ")"
```

Go to: _[natural](#user-content-natural)_;


Circuit, tuple, and array types form the aggregate types,
i.e. types whose values contain (sub)values
(with the corner-case exception of the empty tuple value).

<a name="aggregate-type"></a>
```abnf
aggregate-type = tuple-type / array-type / circuit-type
```

Go to: _[array-type](#user-content-array-type), [tuple-type](#user-content-tuple-type), [circuit-type](#user-content-circuit-type)_;


Scalar and aggregate types form all the types.

<a name="type"></a>
```abnf
type = scalar-type / aggregate-type
```

Go to: _[scalar-type](#user-content-scalar-type), [aggregate-type](#user-content-aggregate-type)_;


As often done in grammatical language syntax specifications,
we define rules for different kinds of expressions,
which also defines the relative precedence
of operators and other expression constructs,
and the (left or right) associativity of binary operators.

The primary expressions are self-contained in a way,
i.e. they have clear deliminations.
Some consist of single tokens:
identifiers, the keywords 'self' and 'input', and literals.
Primary expressions also include parenthesized expressions,
i.e. any expression may be turned into a primary one
by putting parentheses around it.
The other kinds are defined and explained below.

<a name="primary-expression"></a>
```abnf
primary-expression = identifier
                   / "self"
                   / "input"
                   / literal
                   / "(" expression ")"
                   / tuple-expression
                   / array-expression
                   / circuit-expression
                   / function-call
```

Go to: _[function-call](#user-content-function-call), [expression](#user-content-expression), [circuit-expression](#user-content-circuit-expression), [identifier](#user-content-identifier), [literal](#user-content-literal), [array-expression](#user-content-array-expression), [tuple-expression](#user-content-tuple-expression)_;


There are tuple expressions to construct and deconstruct tuples.
A construction consists of zero, two, or more component expressions.
A deconstruction uses a component index (zero-indexed).
Note that constructions are delimited by closing parentheses
and deconstructions are delimited by natural tokens.
The rule below, and similar rules for other aggregate types,
use the perhaps more familiar 'access',
but note that 'deconstruction' has a nice symmetry to 'construction';
the term 'destructor' has a different meaning in other languages,
so we may want to avoid it, but not necessarily.

<a name="tuple-construction"></a>
```abnf
tuple-construction = "(" [ expression 1*( "," expression ) ] ")"
```

Go to: _[expression](#user-content-expression)_;


<a name="tuple-expression"></a>
```abnf
tuple-expression = tuple-construction
```

Go to: _[tuple-construction](#user-content-tuple-construction)_;


The are array expressions to construct and deconstruct arrays.
There are two kinds of constructions:
one lists the element expressions (at least one),
including spreads (via '...') which are arrays being spliced in;
the other repeats (the value of) a single expression
across one or more dimensions.
There are two kinds of deconstructions:
one selects a single element by index (zero-indexed);
the other selects a range via two indices,
the first inclusive and the second exclusive --
both are optional,
the first defaulting to 0 and the second to the array length.
Note that these expressions are all delimited
by closing square brackets.

<a name="array-inline-construction"></a>
```abnf
array-inline-construction = "["
                            array-inline-element
                            *( "," array-inline-element )
                            "]"
```

Go to: _[array-inline-element](#user-content-array-inline-element)_;


<a name="array-inline-element"></a>
```abnf
array-inline-element = expression / "..." expression
```

Go to: _[expression](#user-content-expression)_;


<a name="array-repeat-construction"></a>
```abnf
array-repeat-construction = "[" expression ";" array-dimensions "]"
```

Go to: _[expression](#user-content-expression), [array-dimensions](#user-content-array-dimensions)_;


<a name="array-construction"></a>
```abnf
array-construction = array-inline-construction / array-repeat-construction
```

Go to: _[array-inline-construction](#user-content-array-inline-construction), [array-repeat-construction](#user-content-array-repeat-construction)_;


<a name="array-expression"></a>
```abnf
array-expression = array-construction
```

Go to: _[array-construction](#user-content-array-construction)_;


There are circuit expressions to construct and deconstruct circuit values.
A construction lists values for all the member variables (in any order);
there must be at least one member variable currently.
A single identifier abbreviates
a pair consisting of the same identifier separated by dot;
note that, in the expansion, the left one denotes a member name,
while the right one denotes an expression (a variable),
so they are syntactically identical but semantically different.
A deconstruction selects a member variable by name.
Note that these expressions are delimited,
by closing curly braces or identifiers.

<a name="circuit-construction"></a>
```abnf
circuit-construction = circuit-type "{"
                       circuit-inline-element *( "," circuit-inline-element )
                       "}"
```

Go to: _[circuit-inline-element](#user-content-circuit-inline-element), [circuit-type](#user-content-circuit-type)_;


<a name="circuit-inline-element"></a>
```abnf
circuit-inline-element = identifier ":" expression / identifier
```

Go to: _[identifier](#user-content-identifier), [expression](#user-content-expression)_;


<a name="circuit-expression"></a>
```abnf
circuit-expression = circuit-construction
```

Go to: _[circuit-construction](#user-content-circuit-construction)_;


There are three kinds of function calls:
top-level function calls,
instance (i.e. non-static) member function calls, and
static member function calls.
What changes is the start, but they all end in an argument list,
delimited by a closing parenthesis.

<a name="function-arguments"></a>
```abnf
function-arguments = "(" [ expression *( "," expression ) ] ")"
```

Go to: _[expression](#user-content-expression)_;


Postfix expressions have highest precedence.
They apply to primary expressions.
Contains access expressions for arrays, tuples, and circuits.
Contains function call types.
postfix-expression = primary-expression
                 / postfix-expression "." natural
                 / postfix-expression "." identifier
                 / identifier function-arguments
                 / postfix-expression "." identifier function-arguments
                 / circuit-type "::" identifier function-arguments
                 / postfix-expression "[" expression "]"
                 / postfix-expression "[" [expression] ".." [expression] "]"

Unary operators have the highest operator precedence.
They apply to postfix expressions
and recursively to unary expressions.

<a name="unary-expression"></a>
```abnf
unary-expression = postfix-expression
                 / "!" unary-expression
                 / "-" unary-expression
```

Go to: _[postfix-expression](#user-content-postfix-expression), [unary-expression](#user-content-unary-expression)_;


Next in the operator precedence is casting.
The current rule below makes exponentiation left-associative,
cast-expression = unary-expression
                / cast-expression "as" type

Next in the operator precedence is exponentiation,
following mathematical practice.
The current rule below makes exponentiation left-associative,
i.e. 'a ** b ** c' must be parsed as '(a ** b) ** c'.
This is easy to change if we want it to be right-associative instead.

<a name="exponential-expression"></a>
```abnf
exponential-expression = cast-expression
                       / exponential-expression "**" cast-expression
```

Go to: _[cast-expression](#user-content-cast-expression), [exponential-expression](#user-content-exponential-expression)_;


Next in precedence come multiplication and division, both left-associative.

<a name="multiplicative-expression"></a>
```abnf
multiplicative-expression = exponential-expression
                          / multiplicative-expression "*" exponential-expression
                          / multiplicative-expression "/" exponential-expression
```

Go to: _[exponential-expression](#user-content-exponential-expression), [multiplicative-expression](#user-content-multiplicative-expression)_;


Then there are addition and subtraction, both left-assocative.

<a name="additive-expression"></a>
```abnf
additive-expression = multiplicative-expression
                    / additive-expression "+" multiplicative-expression
                    / additive-expression "-" multiplicative-expression
```

Go to: _[additive-expression](#user-content-additive-expression), [multiplicative-expression](#user-content-multiplicative-expression)_;


Next in the precedence order are ordering relations.
These are not associative, because they return boolean values.

<a name="ordering-expression"></a>
```abnf
ordering-expression = additive-expression
                    / additive-expression "<" additive-expression
                    / additive-expression ">" additive-expression
                    / additive-expression "<=" additive-expression
                    / additive-expression ">=" additive-expression
```

Go to: _[additive-expression](#user-content-additive-expression)_;


Equalities return booleans but may also operate on boolean,
so we make them left-associative.

<a name="equality-expression"></a>
```abnf
equality-expression = ordering-expression
                    / equality-expression "==" ordering-expression
                    / equality-expression "!=" ordering-expression
```

Go to: _[equality-expression](#user-content-equality-expression), [ordering-expression](#user-content-ordering-expression)_;


Next come conjunctive expressions, left-associative.

<a name="conjunctive-expression"></a>
```abnf
conjunctive-expression = equality-expression
                       / conjunctive-expression "&&" equality-expression
```

Go to: _[equality-expression](#user-content-equality-expression), [conjunctive-expression](#user-content-conjunctive-expression)_;


Next come disjunctive expressions, left-associative.

<a name="disjunctive-expression"></a>
```abnf
disjunctive-expression = conjunctive-expression
                       / disjunctive-expression "||" conjunctive-expression
```

Go to: _[disjunctive-expression](#user-content-disjunctive-expression), [conjunctive-expression](#user-content-conjunctive-expression)_;


Finally we have conditional expressions.

<a name="conditional-expression"></a>
```abnf
conditional-expression = disjunctive-expression
                        / conditional-expression
                         "?" expression
                         ":" conditional-expression
```

Go to: _[conditional-expression](#user-content-conditional-expression), [expression](#user-content-expression), [disjunctive-expression](#user-content-disjunctive-expression)_;


These are all the expressions.
Recall that conditional expressions
may be disjunctive expressions,
which may be conjunctive expressions,
and so on all the way to primary expressions.

<a name="expression"></a>
```abnf
expression = conditional-expression
```

Go to: _[conditional-expression](#user-content-conditional-expression)_;


There are various kinds of statements,
including blocks, which are
possibly empty sequences of statements surounded by curly braces.

<a name="statement"></a>
```abnf
statement = expression-statement
          / return-statement
          / variable-definition-statement
          / conditional-statement
          / loop-statement
          / assignment-statement
          / console-statement
          / block
```

Go to: _[variable-definition-statement](#user-content-variable-definition-statement), [loop-statement](#user-content-loop-statement), [block](#user-content-block), [assignment-statement](#user-content-assignment-statement), [console-statement](#user-content-console-statement), [conditional-statement](#user-content-conditional-statement), [return-statement](#user-content-return-statement), [expression-statement](#user-content-expression-statement)_;


<a name="block"></a>
```abnf
block = "{" *statement "}"
```

An expression (that returns the empty tuple)
can be turned into a statement by appending a semicolon.

<a name="expression-statement"></a>
```abnf
expression-statement = expression ";"
```

Go to: _[expression](#user-content-expression)_;


A return statement always takes an expression,
and does not end with a semicolon (but we may want to change that).

<a name="return-statement"></a>
```abnf
return-statement = "return" expression
```

Go to: _[expression](#user-content-expression)_;


There are two kinds of variable definition statements,
which only differ in the starting keyword.
The variables are either a single one or a tuple of two or more;
in all cases, there is just one optional type
and just one initializing expression.

<a name="variable-definition-statement"></a>
```abnf
variable-definition-statement = ( "let" / "const" )
                                identifier-or-identifiers
                                [ ":" type ] "=" expression ";"
```

Go to: _[identifier-or-identifiers](#user-content-identifier-or-identifiers), [type](#user-content-type), [expression](#user-content-expression)_;


<a name="identifier-or-identifiers"></a>
```abnf
identifier-or-identifiers = identifier
                          / "(" identifier 1*( "," identifier ) ")"
```

Go to: _[identifier](#user-content-identifier)_;


A conditional statement always starts with a condition and a block
(which together form a branch).
It may stop there, or it may continue with an alternative block,
or possibly with another conditional statement, forming a chain.
Note that we require blocks in all branches, not merely statements.

<a name="branch"></a>
```abnf
branch = "if" expression block
```

Go to: _[block](#user-content-block), [expression](#user-content-expression)_;


<a name="conditional-statement"></a>
```abnf
conditional-statement = branch
                      / branch "else" block
                      / branch "else" conditional-statement
```

Go to: _[block](#user-content-block), [branch](#user-content-branch), [conditional-statement](#user-content-conditional-statement)_;


A loop statement implicitly defines a loop variable
that goes from a starting value (inclusive) to an ending value (exclusive).
The body is a block.

<a name="loop-statement"></a>
```abnf
loop-statement = "for" identifier "in" expression ".." expression block
```

Go to: _[expression](#user-content-expression), [block](#user-content-block), [identifier](#user-content-identifier)_;


An assignment statement is straightforward.
Based on the operator, the assignment may be simple (i.e. '=')
or compound (i.e. combining assignment with an arithmetic operation).

<a name="assignment-operator"></a>
```abnf
assignment-operator = "=" / "+=" / "-=" / "*=" / "/=" / "**="
```

<a name="assignment-statement"></a>
```abnf
assignment-statement = expression assignment-operator expression ";"
```

Go to: _[assignment-operator](#user-content-assignment-operator), [expression](#user-content-expression)_;


Console statements start with the 'console' keyword,
followed by a console function call.
The call may be an assertion or a print command.
The former takes an expression (which must be boolean) as argument.
The latter takes either no argument,
or a format string followed by expressions,
whose number must match the number of containers '{}' in the format string.
Note that the console function names are identifiers, not keywords.
There are three kind of printing.

<a name="console-statement"></a>
```abnf
console-statement = "console" "." console-call
```

Go to: _[console-call](#user-content-console-call)_;


<a name="console-call"></a>
```abnf
console-call = assert-call
             / print-call
```

Go to: _[assert-call](#user-content-assert-call), [print-call](#user-content-print-call)_;


<a name="assert-call"></a>
```abnf
assert-call = "assert" "(" expression ")"
```

Go to: _[expression](#user-content-expression)_;


<a name="print-function"></a>
```abnf
print-function = "debug" / "error" / "log"
```

<a name="print-arguments"></a>
```abnf
print-arguments = "(" [ formatted-string *( "," expression ) ] ")"
```

Go to: _[formatted-string](#user-content-formatted-string)_;


<a name="print-call"></a>
```abnf
print-call = print-function print-arguments
```

Go to: _[print-function](#user-content-print-function), [print-arguments](#user-content-print-arguments)_;


An annotation consists of an annotation name (which starts with '@')
with optional annotation arguments.
Note that no parentheses are used if there are no arguments.

<a name="annotation"></a>
```abnf
annotation = annotation-name
             [ "(" identifier *( "," identifier ) ")" ]
```

Go to: _[identifier](#user-content-identifier), [annotation-name](#user-content-annotation-name)_;


A function declaration defines a function.
This could be called 'function-definition' instead,
but see the comments about the 'declaration' rule below.
The output type is optional (it defaults to the empty tuple type).
In general, a function input consists of an identifier and a type,
with an optional 'const' modifier.
However, functions inside circuits
may start with a 'mut self' or 'self' parameter,
which may be the only parameter in fact.
Furthermore, any function may end with an 'input' parameter,
which may be the only parameter in fact.

<a name="function-declaration"></a>
```abnf
function-declaration = *annotation "function" identifier
                       "(" [ function-parameters ] ")" [ "->" type ]
                       block
```

Go to: _[function-parameters](#user-content-function-parameters), [type](#user-content-type), [block](#user-content-block), [identifier](#user-content-identifier)_;


<a name="function-parameters"></a>
```abnf
function-parameters = self-parameter [ "," input-parameter ]
                    / self-parameter "," function-inputs [ "," input-parameter ]
                    / function-inputs [ "," input-parameter ]
                    / input-parameter
```

Go to: _[input-parameter](#user-content-input-parameter), [function-inputs](#user-content-function-inputs), [self-parameter](#user-content-self-parameter)_;


<a name="self-parameter"></a>
```abnf
self-parameter = ["mut"] "self"
```

<a name="function-inputs"></a>
```abnf
function-inputs = function-input *( "," function-input )
```

Go to: _[function-input](#user-content-function-input)_;


<a name="function-input"></a>
```abnf
function-input = [ "const" ] identifier ":" type
```

Go to: _[identifier](#user-content-identifier), [type](#user-content-type)_;


<a name="input-parameter"></a>
```abnf
input-parameter = "input"
```

A circuit member variable declaration consists of an identifier and a type.
A circuit member function declaration consists of a function declaration.
We could call these 'member-definition' etc.,
but see the comments about the 'declaration' rule below.

<a name="member-declaration"></a>
```abnf
member-declaration = member-variable-declaration
                   / member-function-declaration
```

Go to: _[member-variable-declaration](#user-content-member-variable-declaration), [member-function-declaration](#user-content-member-function-declaration)_;


<a name="member-variable-declaration"></a>
```abnf
member-variable-declaration = identifier ":" type
```

Go to: _[identifier](#user-content-identifier), [type](#user-content-type)_;


<a name="member-function-declaration"></a>
```abnf
member-function-declaration = function-declaration
```

Go to: _[function-declaration](#user-content-function-declaration)_;


A circuit declaration defines a circuit type. It is straightforward.
This could be called 'circuit-definition' instead,
but see the comments about the 'declaration' rule below.

<a name="circuit-declaration"></a>
```abnf
circuit-declaration = *annotation "circuit" identifier
                      "{" member-declaration *( "," member-declaration ) "}"
```

Go to: _[member-declaration](#user-content-member-declaration), [identifier](#user-content-identifier)_;


An import declaration consists of the 'import' keyword
followed by a package path, which may be one of the following:
a single wildcard;
an identifier, optionally followed by a local renamer;
a package name followed by a path, recursively;
or a parenthesized list of package paths,
which are "fan out" of the initial path.
Note that we allow the last element of the parenthesized list
to be followed by a comma, presumably for convenience.

<a name="import-declaration"></a>
```abnf
import-declaration = "import" package-path
```

Go to: _[package-path](#user-content-package-path)_;


<a name="package-path"></a>
```abnf
package-path = "*"
             / identifier [ "as" identifier ]
             / package-name "." package-path
             / "(" package-path *( "," package-path ) [","] ")"
```

Go to: _[identifier](#user-content-identifier), [package-name](#user-content-package-name), [package-path](#user-content-package-path)_;


Finally, we define a file as a sequence of zero or more declarations.
This is why we used 'function-declaration' and 'circuit-declaration'
instead of 'function-definition' and 'ciruit-definition':
this way, they are all declarations of some sort.
An import declaration cannot really called an import definition,
because it does not define anything.
But we may revisit this, and use 'definition' instead of 'declaration'.

<a name="declaration"></a>
```abnf
declaration = import-declaration
            / function-declaration
            / circuit-declaration
```

Go to: _[import-declaration](#user-content-import-declaration), [function-declaration](#user-content-function-declaration), [circuit-declaration](#user-content-circuit-declaration)_;


<a name="file"></a>
```abnf
file = *declaration