LPegRex is a re-implementation of LPeg/ LPegLabel re module with some extensions to make easy to parse language grammars into an AST (abstract syntax tree) while maintaining readability.

LPegRex stands for LPeg Regular Expression eXtended.

The goal of this library is to extend the LPeg re module with some minor additions to make easy parsing a whole programming language grammar to an abstract syntax tree using a single, simple, compact and clear PEG grammar.

For instance is in the goal of the project to parse Lua 5.4 source files with complete syntax into an abstract syntax tree under 100 lines of clear PEG grammar rules while generating an output suitable to be used analyzed by a compiler. This goal was accomplished, see the Lua example section below.

The new extensions should not break any existing re syntax.

This project will be later incorporated in the Nelua programming language compiler. This goal was accomplished, and LPegRex is the new parsing engine for the Nelua compiler.

• New predefined patterns for control characters (%ca %cb %ct %cn %cv %cf %cr).
• New predefined patterns for utf8 (%utf8 %utf8seq %ascii).
• New predefined pattern for spaces independent of locale (%sp).
• New syntax for capturing arbitrary values while matching empty strings (e.g. $true).
• New syntax for optional captures (e.g patt~?).
• New syntax for throwing labels errors on failure of expected matches (e.g. @rule).
• New syntax for rules that capture AST Nodes (e.g. NodeName <== patt).
• New syntax for rules that capture tables (e.g. MyList <-| patt).
• New syntax for matching unique tokens with automatic skipping (e.g. `,`).
• New syntax for matching unique keywords with automatic skipping (e.g. `for`).
• Auto generate KEYWORD rule based on used keywords in the grammar.
• Auto generate TOKEN rule based on used tokens in the grammar.
• Use supplied NAME_SUFFIX rule for generating each keyword rule.
• Use supplied SKIP rule for generating each keyword or token rule.
• Capture nodes with initial and final positions.
• Support using - character in rule names.
• Pre define some useful auxiliary functions:
  • tonil Substitute captures by nil.
  • totrue Substitute captures by true.
  • tofalse Substitute captures by false.
  • toemptytable Substitute captures by {}.
  • tonumber Substitute a string capture by its corresponding number.
  • tochar Substitute a numeric code capture by its corresponding character byte.
  • toutf8char Substitute a numeric code capture by its corresponding UTF-8 byte sequence.
  • foldleft Fold tables to the left (use only with ~>).
  • foldright Fold tables to the right (use only with ->).
  • rfoldleft Fold tables to the left in reverse order (use only with ->).
  • rfoldright Fold tables to the right in reverse order (use only with ~>)

For reference on how to use re and its syntax, please check its manual first.

Here is a quick reference of the new syntax additions:

As you can notice the additional syntax is mostly sugar for common capture patterns that are used when defining programming language grammars.

Often we need to reduce a list of captured AST nodes into a single captured AST node (e.g. when reducing a call chain), here we call this operation folding. The following table demonstrates the four ways to fold a list of nodes:

Where the pattern patt captures a list of tables with a least one capture. Note that depending on the fold operation you must use its correct arrow (-> or ~>).

Sometimes is useful to match empty strings and capture some arbitrary values, the following tables show auxiliary syntax to help on that:

Sometimes is useful to substitute a list of captures by a lua value, the following tables show auxiliary functions to help on that:

By default when capturing a node with <== syntax, LPegRex will set the following 3 fields:

• tag Name of the node (its type)
• pos Initial position of the node match
• endpos Final position of the node match (usually includes following SKIP)

The user can customize and change these field names or disable them by setting it's corresponding name in the defs.__options table when compiling the grammar, for example:

local mypatt = rex.compile(mygrammar, {__options = {
  tag = 'name', -- 'tag' field rename to 'name'
  pos = 'init', -- 'pos' field renamed to 'init'
  endpos = false, -- don't capture node final position
}})

The fields pos and endpos are useful to generate error messages with precise location when analyzing the AST and the tag field is used to distinguish the node type.

In case defs.__options.tag is a function, then it's called and the user will be responsible for setting the tag field and return the node, this flexibility exists in case specific actions are required to be executed on node creation, for example:

local mypatt = rex.compile(mygrammar, {__options = {
  tag = function(tag, node)
    print('new node', tag)
    node.tag = tag
    return node
  end
}})

Note that when this function is called the node children may be incomplete in case the node is being folded.

When using the back tick syntax (e.g. `something`), LPegRex will register its contents as a keyword in case it begins with a letter (or _), or as token in case it contains only punctuation characters (except _).

Both keywords and tokens always match the SKIP rule immediately to skip spaces, thus the rule SKIP must always be defined when using the back tick syntax.

Tokens matches are always unique in case of common characters, that is, in case both . and .. tokens are defined, the rule `.` will match . but not ...

In case a token is found, the rule TOKEN will be automatically generated, this rule will match any token plus SKIP.

In case a keyword is found, the rule NAME_SUFFIX also need to be defined, it's used to differentiate keywords from identifier names.

In most cases the user will need define something like:

NAME_SUFFIX   <- [_%w]+
SKIP          <- %s+

You may want to edit the SKIP rule to consider comments if you grammar supports them. Token and keywords will not capture SKIP rule when using the syntax {`keyword`}.

Often we need to create a rule that capture identifier names while ignoring grammar keywords, let call this rule NAME. To assist doing this the KEYWORD rule is automatically generated based on all defined keywords in the grammar, the user can then use it to define the NAME rule, in most cases something like:

NAME          <-- !KEYWORD {NAME_PREFIX NAME_SUFFIX?} SKIP
NAME_PREFIX   <-- [_%a]
NAME_SUFFIX   <-- [_%w]+
SKIP          <- %s+

Any rule name, keyword, token or string pattern can be preceded by the token @, marking it as an expected match, in case the match is not fulfilled an error label will be thrown using the name Expected_name, where name is the token, keyword or rule name.

Once an error label is found, the user can generate pretty syntax error messages using the function lpegrex.calcline to gather line information, for example:

local patt = lpegrex.compile(PEG)
local ast, errlabel, errpos = patt:match(source)
if not ast then
  local lineno, colno, line = lpegrex.calcline(source, errpos)
  local colhelp = string.rep(' ', colno-1)..'^'
  error('syntax error: '..filename..':'..lineno..':'..colno..': '..errlabel..
        '\n'..line..'\n'..colhelp)
end

Here is a small example parsing JSON into an AST in 12 lines of PEG rules:

local lpegrex = require 'lpegrex'

local patt = lpegrex.compile([[
Json          <-- SKIP (Object / Array) (!.)^UnexpectedSyntax
Object        <== `{` (Member (`,` @Member)*)? @`}`
Array         <== `[` (Value (`,` @Value)*)? @`]`
Member        <== String `:` @Value
Value         <-- String / Number / Object / Array / Boolean / Null
String        <-- '"' {~ ('\' -> '' @ESCAPE / !'"' .)* ~} @'"' SKIP
Number        <-- {[+-]? (%d+ '.'? %d+? / '.' %d+) ([eE] [+-]? %d+)?} -> tonumber SKIP
Boolean       <-- `false` -> tofalse / `true` -> totrue
Null          <-- `null` -> tonil
ESCAPE        <-- [\/"] / ('b' $8 / 't' $9 / 'n' $10 / 'f' $12 / 'r' $13 / 'u' {%x^4} $16) -> tochar
SKIP          <-- %s*
NAME_SUFFIX   <-- [_%w]+
]])

local source = '[{"string":"some\\ntext", "boolean":true, "number":-1.5e+2, "null":null}]'

local ast, errlabel, errpos = patt:match(source)
if not ast then
  local lineno, colno, line = lpegrex.calcline(source, errpos)
  local colhelp = string.rep(' ', colno-1)..'^'
  error('syntax error: '..lineno..':'..colno..': '..errlabel..
        '\n'..line..'\n'..colhelp)
end
-- `ast` should be a table with the JSON
print('JSON parsed with success!')

The above should parse into the following equivalent AST table:

local ast = { tag = "Array", pos = 1, endpos = 73,
  { tag = "Object", pos = 2, endpos = 72,
    { tag = "Member", pos = 3, endpos = 24,
    "string","some\ntext" },
    { tag = "Member", pos = 26, endpos = 40,
    "boolean", true },
    { tag = "Member", pos = 42, endpos = 58,
      "number", -150.0 },
    { tag = "Member", pos = 60, endpos = 71,
      "null", nil }
  }
}

A JSON parser similar to this example can be found in parsers/json.lua.

When prototyping complex grammars you may want to debug the rules that the parser is trying to match and the ones that were successfully matched. You can enable LPegRex debug mode for this by setting lpegrex.debug = true globally. When debug is enabled all compiled grammars will be compiled in debug mode.

When debugging is enabled every attempt to match a rule will print ENTER <rulename> (<lineno>:<colno>) to io.stderr, and every rule successfully matched will print LEAVE <rulename> (<lineno>:<colno>) to io.stderr. Notice that rules failing to match will not print LEAVE.

The following is an example of parsing {"string": JSON chunk using the JSON parser shown above with debugging enabled:

ENTER Json (1:1)
ENTER SKIP (1:1)
LEAVE SKIP (1:1)
ENTER Object (1:1)
ENTER { (1:1)
ENTER Array (1:1)
ENTER [ (1:1)
ENTER SKIP (1:2)
LEAVE SKIP (1:2)
LEAVE [ (1:2)
ENTER Value (1:2)
ENTER String (1:2)
ENTER Number (1:2)
ENTER Object (1:2)
ENTER { (1:2)
ENTER SKIP (1:3)
LEAVE SKIP (1:3)
LEAVE { (1:3)
ENTER Member (1:3)
ENTER String (1:3)
ENTER SKIP (1:11)
LEAVE SKIP (1:11)
LEAVE String (1:11)
ENTER : (1:11)
ENTER SKIP (1:12)
LEAVE SKIP (1:12)
LEAVE : (1:12)

Notice String ENTER at 1:3 and LEAVE at 1:11, this means that we have matched the rule String in that range. Notice Number ENTER at 1:2 while no LEAVE is shown for Number, this means that we attempted to match Number but it failed since no LEAVE was shown afterwards.

To use LPegRex you need LPegLabel to be properly installed. If you have it already installed you can just copy the lpegrex.lua file.

If you can also install it using the LuaRocks package manager, with the following command:

luarocks install lpegrex

The library should work with Lua 5.x versions (and also LuaJIT).

A Lua 5.4 parser is defined in parsers/lua.lua, it servers as a good example on how to define a full language grammar in a single PEG that generates an AST suitable to be analyzed by a compiler, while also handling source syntax errors.

A Lua AST printer using it is available in examples/lua.lua You can run it to parse any Lua file and print its AST.

For example by doing lua examples/lua-ast.lua inputs/fact.lua you should get the following AST output:

Block
| FuncDecl
| | Id
| | | "fact"
| | -
| | | Id
| | | | "n"
| | Block
| | | If
| | | | BinaryOp
| | | | | Id
| | | | | | "n"
| | | | | "eq"
| | | | | Number
| | | | | | 0
| | | | Block
| | | | | Return
| | | | | | -
| | | | | | | Number
| | | | | | | | 1
| | | | Block
| | | | | Return
| | | | | | -
| | | | | | | BinaryOp
| | | | | | | | Id
| | | | | | | | | "n"
| | | | | | | | "mul"
| | | | | | | | Call
| | | | | | | | | -
| | | | | | | | | | BinaryOp
| | | | | | | | | | | Id
| | | | | | | | | | | | "n"
| | | | | | | | | | | "sub"
| | | | | | | | | | | Number
| | | | | | | | | | | | 1
| | | | | | | | | Id
| | | | | | | | | | "fact"
| Call
| | -
| | | Call
| | | | -
| | | | | Number
| | | | | | 10
| | | | Id
| | | | | "fact"
| | Id
| | | "print"

A complete C11 parser has been implemented and is available in parsers/c11.lua, it's experimental but it was verified to parse hundreds of prepossessed C file sources.

A C11 AST printer using it is available in examples/c11-ast.lua.

Note that the C file must be preprocessed, you can generate a preprocessed C file with GCC/Clang or running gcc -E file.c > file_preprocessed.c.

For example by doing lua examples/c11-ast.lua inputs/fact.c you should get the following AST output:

translation-unit
| declaration
| | type-declaration
| | | declaration-specifiers
| | | | storage-class-specifier
| | | | | "extern"
| | | | type-specifier
| | | | | "int"
| | | init-declarator-list
| | | | init-declarator
| | | | | declarator
| | | | | | declarator-parameters
| | | | | | | identifier
| | | | | | | | "printf"
| | | | | | | parameter-type-list
| | | | | | | | parameter-declaration
| | | | | | | | | declaration-specifiers
| | | | | | | | | | type-qualifier
| | | | | | | | | | | "const"
| | | | | | | | | | type-specifier
| | | | | | | | | | | "char"
| | | | | | | | | declarator
| | | | | | | | | | pointer
| | | | | | | | | | | identifier
| | | | | | | | | | | | "format"
| | | | | | | | parameter-varargs
| function-definition
| | declaration-specifiers
| | | storage-class-specifier
| | | | "static"
| | | type-specifier
| | | | "int"
| | declarator
| | | declarator-parameters
| | | | identifier
| | | | | "fact"
| | | | parameter-type-list
| | | | | parameter-declaration
| | | | | | declaration-specifiers
| | | | | | | type-specifier
| | | | | | | | "int"
| | | | | | declarator
| | | | | | | identifier
| | | | | | | | "n"
| | declaration-list
| | compound-statement
| | | if-statement
| | | | expression
| | | | | binary-op
| | | | | | identifier
| | | | | | | "n"
| | | | | | "=="
| | | | | | integer-constant
| | | | | | | "0"
| | | | return-statement
| | | | | expression
| | | | | | integer-constant
| | | | | | | "1"
| | | | return-statement
| | | | | expression
| | | | | | binary-op
| | | | | | | identifier
| | | | | | | | "n"
| | | | | | | "*"
| | | | | | | argument-expression
| | | | | | | | argument-expression-list
| | | | | | | | | binary-op
| | | | | | | | | | identifier
| | | | | | | | | | | "n"
| | | | | | | | | | "-"
| | | | | | | | | | integer-constant
| | | | | | | | | | | "1"
| | | | | | | | identifier
| | | | | | | | | "fact"
| function-definition
| | declaration-specifiers
| | | type-specifier
| | | | "int"
| | declarator
| | | declarator-parameters
| | | | identifier
| | | | | "main"
| | declaration-list
| | compound-statement
| | | expression-statement
| | | | expression
| | | | | argument-expression
| | | | | | argument-expression-list
| | | | | | | string-literal
| | | | | | | | "%d\\n"
| | | | | | | argument-expression
| | | | | | | | argument-expression-list
| | | | | | | | | integer-constant
| | | | | | | | | | "10"
| | | | | | | | identifier
| | | | | | | | | "fact"
| | | | | | identifier
| | | | | | | "printf"
| | | return-statement
| | | | expression
| | | | | integer-constant
| | | | | | "0"

LPegRex is successfully used as the parsing engine in the Nelua programming language compiler, you can see the complete syntax defined in a single PEG grammar in this file.

You can test and prototype grammars with LPegRex live in the browser using the cool lua-wasm-playground tool created by @mingodad. There are C11 and Lua parsers as examples there.

Most LPeg/LPegLabel tests where migrated into tests/lpegrex-test.lua and new tests for the addition extensions were added.

To run the tests just run lua tests/test.lua.

MIT, see LICENSE file.