DynParser

• repository
• doc
• rust-crate

A small and simple Dynamic Parser. It's not a compile time parser.

You can create and modify the grammar on runtime.

A parser is something that takes an input, process it with some rules and generate an AST

There are also some tools to manage the AST (pruning, compacting, flattening...)

In order to create the grammar, you can build a set of rules, or you can use macros to use a better syntax. But, the easier way, is to use a peg grammar.

Then we need an additional step.

More info about the peg syntax bellow.

You can also generate rust code from rules generated from peg.

This allow you to avoid the peg step and more...

In fact, in order to use a peg grammar, you have to parse it. How to parse a peg grammar? Well, this is a parser, therefore...

More details about it bellow on section (parsing the parser)

Usage

Add to cargo.toml

[dependencies]
dynparser = "0.4.0"

Watch examples below

Modifications

0.1.0   First version

0.2.0   Fixed some errors
        Rules code for peg parsing generated automatically from peg

0.3.0   Passthrow method on AST

0.4.0   Literals with escape (optional)
        Error constructor on peg grammar
        Flattening the AST

0.4.2   Fixed error managing error("xxx")
        Working on modules

TODO

• move to macros by example 2.0 and improve some
• apply tail recursion parsing rule
• macro for eof

Basic example

Lets create the next grammar:

    main            =   letter letter_or_num+

    letter          =   [a-zA-Z]

    letter_or_num   =   letter
                    /   number

    number          =   [0-9]

This grammar will accept a letter, followed from one or more letters or numbers

Just from peg

Straightforward...

    extern crate dynparser;
    use dynparser::{parse, rules_from_peg};

    fn main() {
        let rules = rules_from_peg(
            r#"

    main            =   letter letter_or_num+

    letter          =   [a-zA-Z]

    letter_or_num   =   letter
                    /   number

    number          =   [0-9]

            "#,
        ).unwrap();

        assert!(parse("a2AA456bzJ88", &rules).is_ok());
    }

If you want to print more information...

    extern crate dynparser;
    use dynparser::{parse, rules_from_peg};

    fn main() {
        let rules = rules_from_peg(
            r#"

    main            =   letter letter_or_num+

    letter          =   [a-zA-Z]

    letter_or_num   =   letter
                    /   number

    number          =   [0-9]

            "#,
        ).map_err(|e| {
            println!("{}", e);
            panic!("FAIL");
        })
            .unwrap();

        println!("{:#?}", rules);

        let result = parse("a2Z", &rules);
        match result {
            Ok(ast) => println!("{:#?}", ast),
            Err(e) => println!("Error: {:?}", e),
        };
    }

The AST produced will be:

Rule(
    (
        "main",
        [
            Rule(
                (
                    "letter",
                    [
                        Val(
                            "a"
                        )
                    ]
                )
            ),
            Rule(
                (
                    "letter_or_num",
                    [
                        Rule(
                            (
                                "number",
                                [
                                    Val(
                                        "2"
                                    )
                                ]
                            )
                        )
                    ]
                )
            ),
            Rule(
                (
                    "letter_or_num",
                    [
                        Rule(
                            (
                                "letter",
                                [
                                    Val(
                                        "Z"
                                    )
                                ]
                            )
                        )
                    ]
                )
            )
        ]
    )
)

The AST type is:

pub enum Node {
    Val(String),
    Rule((String, Vec<Node>)),
    EOF,
}

You can also work with flattened AST.In several cases will be easier to visit a flattened AST.

The Flattened AST is:

pub enum Node {
    Val(String),
    BeginRule(String),
    EndRule(String),
    EOF,
}
}

Just it (remember, more information about the peg grammar bellow)

Example 2

You will configure a set of rules to parse.

The rule is composed of a name followed by an arrow and an expression to be parsed.

A basic example

Lets create the next grammar:

   main    =   'a' ( 'bc' 'c'
                   / 'bcdd'
                   / b_and_c  d_or_z
                   )

   b_and_c =   'b' 'c'
   d_or_z  =   'd' / 'z'

Just from peg 2

extern crate dynparser;
use dynparser::{parse, rules_from_peg};

fn main() {
    let rules = rules_from_peg(
        r#"

    main    =   'a' ( 'bc'  'c'
                    /  'bcdd'
                    / b_and_c  d_or_z
                    )

    b_and_c =   'b'   'c'
    d_or_z  =   'd' / 'z'

        "#,
    ).unwrap();

    assert!(parse("abcz", &rules).is_ok());
    assert!(parse("abcdd", &rules).is_ok());
    assert!(parse("abcc", &rules).is_ok());
    assert!(parse("bczd", &rules).is_err());
}

The exit will be the next AST

   Rule(
       (
           "main",
           [
               Val(
                   "a"
               ),
               Rule(
                   (
                       "b_and_c",
                       [
                           Val(
                               "b"
                           ),
                           Val(
                               "c"
                           )
                       ]
                   )
               ),
               Rule(
                   (
                       "d_or_z",
                       [
                           Val(
                               "d"
                           )
                       ]
                   )
               )
           ]
       )
   )

This is a dynamic parser, you can add rules at execution time.

pending: example

Generating the rules by hand with macros

You can create this grammar and parse the string "abcd" with macros like:

#[macro_use]
extern crate dynparser;
use dynparser::parse;

fn main() {
    let rules = rules!{
       "main"   =>  and!{
                        lit!("a"),
                        or!(
                            and!(lit!("bc"), lit!("c")),
                            lit!("bcdd"),
                            and!(
                                ref_rule!("b_and_c"),
                                ref_rule!("d_or_z")
                            )
                        )
                    },
        "b_and_c"  => and!(lit!("b"), lit!("c")),
        "d_or_z"  => or!(lit!("d"), lit!("z"))
    };

    let result = parse("abcd", &rules);
    match result {
        Ok(ast) => println!("{:#?}", ast),
        Err(e) => println!("Error: {:?}", e),
    };
}

Adding a rule on execution time:

#[macro_use]  extern crate dynparser;
use dynparser::parse;
fn main() {
    let rules = rules!{
       "main"   =>  and!{
                        rep!(lit!("a"), 1, 5),
                        ref_rule!("rule2")
                    }
    };

    let rules = rules.add("rule2", lit!("bcd"));

    assert!(parse("aabcd", &rules).is_ok())
}

Of course, you could need to add (or merge) several rules at once

And of course, you can add several rules at once

#[macro_use]  extern crate dynparser;
use dynparser::parse;
fn main() {
    let r = rules!{
       "main"   =>  and!{
                        rep!(lit!("a"), 1, 5),
                        ref_rule!("rule2")
                    }
    };
    let r = r.merge(rules!{"rule2" => lit!("bcd")});
    assert!(parse("aabcd", &r).is_ok())
}

merge takes the ownership of both set of rules and returns a "new" (in fact modified) set of rules. This helps to reduce mutability

main rule is the entry point.

More information in doc

Calculator example

A parser is not a parser without basic math expression parser example.

Here it is...

extern crate dynparser;
use dynparser::{parse, rules_from_peg};

fn main() {
    let rules = rules_from_peg(
        r#"

    main            =   _  expr  _

    expr            =   add_t       (_  add_op  _   add_t)*
                    /   portion_expr

    add_t           =   fact_t      (_  fact_op _   fact_t)*

    fact_t          =   portion_expr

    portion_expr    =   '('  expr ')'
                    /   item

    item            =   num

    num             =   [0-9]+ ('.' [0-9]+)?
    add_op          =   '+'  /  '-'
    fact_op         =   '*'  /  '/'

    _               =   ' '*

        "#,
    ).map_err(|e| {
        println!("{}", e);
        panic!("FAIL");
    })
        .unwrap();

    let result = parse(" 1 +  2*  3 +(5/5 - (8-7))", &rules);
    match result {
        Ok(ast) => println!(
            "{:#?}",
            ast.compact()
                .prune(&vec!["_"])
                .passthrow_except(&vec!["main", "add_t", "fact_t"])
        ),
        Err(e) => println!("Error: {:?}", e),
    };
}

PEG

Rule elements enumeration

Examples below

token	Description
=	On left, symbol, on right expresion defining symbol
symbol	It's an string without quotes
.	Any char
'...'	Literal delimited by single quotes
"..."	Literal delimited by quotes. It accepts escape chars
space	Separate tokens and Rule concatenation (and operation)
/	Or operation
(...)	A expression composed of sub expressions
?	One optional
*	Repeat 0 or more
+	Repeat 1 or more
!	negate expression
[...]	Match chars. It's a list or ranges (or both)
error(...)	Let us to define specific errors
->	pending...
:	pending...

Let's see by example

Rules by example

The best way to know the peg syntax, is to look the peg grammar. And yes it is on peg syntax :-)

A simple literal string.

main = 'Hello world'

There are two literal types.

No escaped literals, are delimited by '

And escaped literals, delimited by ". "\n" will be transformed in new-line char i.e.

It's possible to represent a char by an hex number. i.e. "0x13"

main   = "Hello\nworld"

main   = "Hello\0x13world"

main   = 'Hello' "\n"    'world'

main   = 'Hello' "\0x13" 'world'

With this two types of literals, it's easy to have " and '

main   = "'"

main   = '"'

It's recomended to use non escaped literals as much as possible and use the escaped literals when necessary.

Concatenation (and)

main = 'Hello '  'world'

Referencing symbols (rule)

Symbol

main = hi
hi   = 'Hello world'

Or /

main = 'hello' / 'hi'

Or multiline

main
    = 'hello'
    / 'hi'
    / 'hola'

Or multiline 2

main = 'hello'
     / 'hi'
     / 'hola'

Or disorganized

main = 'hello'
     / 'hi' / 'hola'

An important note about the or

main    =   'hello'
        /   'hello world'
        /   'hola'

Given the text hello world, the first option will match processing the first word of the input, and the second one will never be executed. It could be fixed, but... doesn't look a great idea.

Fixing the grammar to avoid this problems, it's very easy. Trying to fix the parser to let this kind of grammars, is expensive.

Parenthesis

main = ('hello' / 'hi')  ' world'

Just multiline

Multiline1

main
    = ('hello' / 'hi')  ' world'

Multiline2

main
    = ('hello' / 'hi')
    ' world'

Multiline3

main = ('hello' / 'hi')
     ' world'

It is recommended to use or operator / on each new line and = on first line, like

Multiline organized

main = ('hello' / 'hi')  ' world'
     / 'bye'

One optional

main = ('hello' / 'hi')  ' world'?

Repetitions

main         = one_or_more_a / zero_or_many_b
one_or_more  = 'a'+
zero_or_many = 'b'*

Negation will not move current position

Next example will consume all chars till get an 'a'

Negation

main = (!'a' .)* 'a'

Consume till

//  This is a line comment
/*  This is a
    multiline comment  */
comment = '//' (!'\n' .)*       //  line comment can be at the end of line
        / '/*' (!'*/' .)* '*/'  /*  a multiline comment can start
                                    at any place
                                */

Match a set of chars. Chars can be defined by range.

number  = digit+ ('.' digit+)?
digit   = [0-9]
a_or_b  = [ab]
id      = [_a-zA-Z][_a-zA-Z0-9]*

a_or_b_or_digit  = [ab0-9]

Simple recursion

one or more 'a' recursive

as  = 'a' as
    / 'a'

//  simplified with `+`
ak = 'a'+

Recursion to match parenthesis

Recursion match par

match_par = '(' match_par ')'
          / '(' ')'

That's ok and works fine, but we can inprove error messages...

In order to improve error messages, would be interesting to modify the grammar.

See next section.

In some cases, we can have an error for no termination consuming full input.

The reason is on

...
and_expr        =   compl_expr  (  ' '  _  and_expr)*
...

Showing an error informing that we didn't consume full input, is not the best.

Here, we said, "hey, try to look for a sequence, or not *"

And is not, then the parser say, I matched the rule, I have to continue verifying other previous branches. But there are no previous partial applied brunch. Then the parser ends not consuming all the input.

To improve error messages, would be interesting to have something like:

Errors included on peg grammar also will help in this case (see next section)

Full grammar in peg format bellow (a grammar for the grammar)...

Errors

Errors are very important.

Take a look to this grammar

    main    =   '('  main  ')'
            /   'hello'

It will force to match parenthesis around the word 'hello'

That's great, but what if we write ((hello)

The system will point the error place, but... witch is going to be the message?

We would like to have a message like unbalanced parenthesis

We can...

    main    =   '('  main  ( ')'  /  error("unbalanced parenthesis") )
            /   'hello'

With this constructor, we can improve our error messages :-)

And we also can remove errors kind of not consumed full input

Remember.The best way to know the peg syntax, is to look the peg grammar. And yes it is on peg syntax :-)

Full exmample...

extern crate dynparser;
use dynparser::{parse, rules_from_peg};
fn main() {
    let rules = rules_from_peg(
        r#"

    main    =   '('  main   ( ')'  /  error("unbalanced parenthesis") )
            /   'hello'

        "#,
    ).unwrap();

    match parse("((hello)", &rules) {
        Ok(_) => panic!("It should fail"),
        Err(e) => assert!(e.descr == "unbalanced parenthesis"),
    }
}

Text

Hey, I'm a text parser, I need a text to parse ;-P

If you want to parse text indentation sensitive, I recommend you the lib indentation_flattener

A grammar for the grammar

A grammar to define the grammar to be parsed by de parser. ;-P

I will define the grammar using the this parser grammar definition rules.

A grammar is a set of rules.

A rule, is a symbol followed by = and an expression

grammar = rule+
rule    = symbol '='  expr

Here we relax the verification to keep the grammar as simple as possible. It's missing also the non significant spaces.

About the expression.

As you know, it's important to accept valid inputs, but also it's important to build an AST with proper priority.

Next grammar:

main    =  'A' 'B'  /  'B' 'C'

It's equivalent to:

main    =  ('A' 'B')  /  ('B' 'C')

But not to:

main    =  (('A' 'B')  /  'B') 'C'

To represent this priority, the expression rule has to be defined in a descendant priority way:

expr            =   or_expr

or_expr         =   and_expr     ('/'  or_expr)*

and_expr        =   simpl_expr   (' '  and_expr)*

simpl_expr      =   '!' atom_or_par
                /   simpl_par ('*' / '+')

atom_or_par     =   (atom / parenth_expr)


parenth_expr    =   '('  expr ')'

Descendant definition

expr	Description
atom_or_par	It's an atom or a parenthesis expression
rep_or_neg	It's not a composition of and or or expressions. It can have negation or repetitions
parenth	It's an expressions with parenthesis
and	Sequence of expressions separated by space
or	Sequence of expression separated by '/'

Now, it's the atom turn:

atom    =   literal
        /   match
        /   dot
        /   symbol

literal =   "\""  (!"\"" .)*  "\""
match   =   '['  ((.  '-'  .)  /  (.))+   ']'
dot     =   '.'
symbol  =   [a-zA-Z0-9_]+

Hey, what about comments?

What about non significate spaces and carry return?

It will be defined on '_' symbol

This is the general idea. The peg used by the parser will evolve to add error control, vars, scape on strings, and other ideas.

As the parser will generate the code from peg to parse itself... It's easy to keep updated the peg grammar used to parse from peg.

    main            =   grammar

    grammar         =   rule+

    rule            =   _  symbol  _  '='  _  expr  _eol _

    expr            =   or

    or              =   and         ( _  '/'  _  or  )*

    and             =   rep_or_neg  ( _1 _ !(symbol _ '=') and )*

    rep_or_neg      =   atom_or_par ('*' / '+' / '?')?
                    /   '!' atom_or_par

    atom_or_par     =   (atom / parenth)

    parenth         =   '('  _  expr  _  ')'

    atom            =   literal
                    /   match
                    /   dot
                    /   symbol

    literal         =  lit_noesc  /  lit_esc

    lit_noesc       =   _'   (  !_' .  )*   _'
    _'              =   "'"

    lit_esc         =   _"
                            (   esc_char
                            /   hex_char
                            /   !_" .
                            )*
                        _"
    _"              =   '"'

    esc_char        =   '\r'
                    /   '\n'
                    /   '\\'
                    /   '\"'

    hex_char        =   '\0x' [0-9A-F] [0-9A-F]

    symbol          =   [_a-zA-Z0-9] [_'"a-zA-Z0-9]*

    eol             =   ("\r\n"  /  "\n"  /  "\r")
    _eol            =   ' '*  eol

    match           =   '['
                            (
                                (mchars  mbetween*)
                                / mbetween+
                            )
                        ']'

    mchars          =   (!']' !(. '-') .)+
    mbetween        =   (.  '-'  .)

    dot             =   '.'

    _               =   (  ' '
                            /   eol
                        )*

    _1              =   (' ' / eol)

Parsing the parser

Or... how to parse yourself

Remember, we started with the concept of a simple parser...

Starting with a set of rules and the input to process, we will generate the AST

On this, we added an additional step to generate the rules from a peg grammar avoiding written by hand on code.

Then, we have a parser that accepts peg grammars.

Now instead of giving the set of rules, we can provide a peg definition and the input to generate the AST

But the input peg grammar has to be processed (parsed). We have to writte rules_from_peg code to parse the input peg

Who's gonna parse the grammar peg? A parser?

Let me think... Ummmm!!!

I'm a parser!!!!!

We have a feature that allows us to generate the Rust code for an AST tree generated from a peg grammar. Oh?!

So the code to parse the peg grammar will be generated automatically with this parser

Then we will generate automatically rules_from_peg recursively.

Once this is done, we can now use the parser in a classic way

Remember, a normal parsing, we have two inputs.

1. The peg grammar
2. The input text

Now, for start with, both inputs will be a peg grammar defining it self (a peg grammar defining a peg grammar)

1. input: peg grammar defining itself
2. running rules_from_peg to generate a set of rules for this peg grammar
3. With the two previous points, we will parse creating the AST for the peg grammar
4. Now we will call ast::generate_rust to generate the code for rules_from_peg
5. We will insert this code on the parser
6. And we are ready to parse an input with a peg grammar to generate the AST

The point rules_from_peg it's special.

pub fn rules_from_peg(peg: &str) -> Result {
    let ast = parse(peg, &rules::parse_peg())?;
    let nodes = ast.compact().prune(&["_", "_1", "_eol"]).flatten();

    rules_from_flat_ast(&nodes)
}

As you can see, we parse the peg grammar (in this case a peg defining the peg grammar).

After it, we transform the AST compacting, removing nodes, and flattening.

An AST flattened, is just something to be parsed, but instead chars, we work with tokens, and it's a LL(1) parser.

Errors will be found and registered in the previous parsing.

Then, we have to write by hand the LL(1) parser, but it's easy (not necessary to control errors, not working with chars, just LL(1))

Why to do that?

First, it's possible and a great test.

Second. If we want to modify our peg grammar, it's boring and error prone to write the code manually.

Using a peg file to generate automatically rules_from_peg, keeps document and code as one (always synchronized)

diagrams generation

echo "[ input peg ] -- rules_from_peg --> [ rules ][ input text ], [ rules ] --> { end: back,0; } [ AST ]" | graph-easy --dot | dot -Tpng -o doc_images/basic.png

echo "[ rules ][ input text ], [ rules ] --> { end: back,0; } [ AST ]" | graph-easy --dot | dot -Tpng -o doc_images/simple_parser.png

echo "
[input peg \\n
  for peg grammar ] -- [rules_from_peg] { shape: none; } -->
                [rules_peg_gramm] { label: rules\\n
                                        for peg grammar }
[input peg \\n
  for peg grammar ], [ rules_peg_gramm ] -- parse --> { end: back,0; } [ AST ]

[AST] ~~ generate rust ~~> [rules_from_peg] { shape: none; }

" | graph-easy --dot | dot -Tpng -o doc_images/automatic_diagram.png