A compiled (C) parser for the MoonScript language. This repo uses pgen to compile moonscript_parser/grammar.lua into a native C Lua module that can parse MoonScript code. This is the experimental new parser that is fully compatible and will replace the LPeg version in in the MoonScript repository.

luarocks install moonscript-parser

Building requires a C compiler and the Lua headers; the module has no other dependencies (no lpeg).

Recent versions of the moonscript rock can use this module in place of the LPeg parser. It is opt-in: set MOONSCRIPT_PARSER=pgen in the environment:

MOONSCRIPT_PARSER=pgen moonc ...

If the variable is set but this module isn't installed, parsing fails with a require error rather than silently using the LPeg parser.

local parse = require "moonscript_parser"

local tree, err = parse.string(moonscript_code)
if not tree then
  print(err) -- line/column message with a source excerpt
end

Benchmarked against the LPeg parser from moonscript 0.6.0 by parsing every .moon file from the lapis, lapis-community, and moonscript repositories: 405 files, 1.43 MiB, 58,296 lines. Times are the best of 5 passes over the full corpus, parse time only (source pre-loaded into memory, no compilation).

Measured on an AMD Ryzen 8840U with LPeg 1.1.0.

moonscript_parser.slow is the same grammar generated as a pure Lua output module by pgen for environments where the C module can't be built. It's substantially slower but requires no C modules.

The rock vendors the generated C module (moonscript_parser.c). To rebuild the parser, you will need pgen installed. (luarocks install pgen)

# regenerate moonscript_parser.c (and the vendored errors.lua) from the grammar
make generate

# run the test suite
make test

The test suite compares the parsed AST to a set of pre-built expected outputs. make test runs it twice: against the native module, then with TEST_SLOW=1 to cover the pure Lua parser. If you are changing the AST then you will need to regenerate the outputs

make build_test_outputs

The initial outputs were generated while the parser had full differential parity with the reference LPeg parser, so they encode its exact tree shapes.