Justfile grammar is a little weird. Because of the freeform
nature of recipe bodies, we don't tokenize them with the
same rules as the rest of the justfile. Instead the
tokenizer will emit a INDENT at the beginning of a recipe
body, one or more LINEs, which match everything after the
INDENT whitespace, and a DEDENT at the end.

Thus the lexer is context sensitive, which is a little
gross.

tokens:

NAME    = /[a-z]((_|-)?[a-z0-9])*/
EOL     = /\n|\r\n/
COMMENT = /#[^!].*/
COLON   = /:/
INDENT  = emitted when indentation increases
DEDENT  = emitted when indentation decreases
LINE    = /.*/ only emitted between INDENT/DEDENT pairs, doesn't include INDENT whitespace
EOF     = emitted at the end of input

grammar:

justfile = item* EOF

item = COMMENT
     | recipe
     | EOL

assignment = NAME EQUALS expression COMMENT? EOL

expression = STRING

recipe = NAME+ COLON NAME* EOL (INDENT LINE+ DEDENT)?