72 lines
2.5 KiB
Lua
72 lines
2.5 KiB
Lua
|
-- Copyright 2006-2019 Mitchell mitchell.att.foicica.com. See License.txt.
|
||
|
-- Objective C LPeg lexer.
|
||
|
|
||
|
local lexer = require('lexer')
|
||
|
local token, word_match = lexer.token, lexer.word_match
|
||
|
local P, R, S = lpeg.P, lpeg.R, lpeg.S
|
||
|
|
||
|
local lex = lexer.new('objective_c')
|
||
|
|
||
|
-- Whitespace.
|
||
|
lex:add_rule('whitespace', token(lexer.WHITESPACE, lexer.space^1))
|
||
|
|
||
|
-- Keywords.
|
||
|
lex:add_rule('keyword', token(lexer.KEYWORD, word_match[[
|
||
|
-- From C.
|
||
|
asm auto break case const continue default do else extern false for goto if
|
||
|
inline register return sizeof static switch true typedef void volatile while
|
||
|
restrict _Bool _Complex _Pragma _Imaginary
|
||
|
-- Objective C.
|
||
|
oneway in out inout bycopy byref self super
|
||
|
-- Preprocessor directives.
|
||
|
@interface @implementation @protocol @end @private @protected @public @class
|
||
|
@selector @encode @defs @synchronized @try @throw @catch @finally
|
||
|
-- Constants.
|
||
|
TRUE FALSE YES NO NULL nil Nil METHOD_NULL
|
||
|
]]))
|
||
|
|
||
|
-- Types.
|
||
|
lex:add_rule('type', token(lexer.TYPE, word_match[[
|
||
|
apply_t id Class MetaClass Object Protocol retval_t SEL STR IMP BOOL
|
||
|
TypedStream
|
||
|
]]))
|
||
|
|
||
|
-- Strings.
|
||
|
local sq_str = P('L')^-1 * lexer.delimited_range("'", true)
|
||
|
local dq_str = P('L')^-1 * lexer.delimited_range('"', true)
|
||
|
lex:add_rule('string', token(lexer.STRING, sq_str + dq_str))
|
||
|
|
||
|
-- Identifiers.
|
||
|
lex:add_rule('identifier', token(lexer.IDENTIFIER, lexer.word))
|
||
|
|
||
|
-- Comments.
|
||
|
local line_comment = '//' * lexer.nonnewline_esc^0
|
||
|
local block_comment = '/*' * (lexer.any - '*/')^0 * P('*/')^-1
|
||
|
lex:add_rule('comment', token(lexer.COMMENT, line_comment + block_comment))
|
||
|
|
||
|
-- Numbers.
|
||
|
lex:add_rule('number', token(lexer.NUMBER, lexer.float + lexer.integer))
|
||
|
|
||
|
-- Preprocessor.
|
||
|
local preproc_word = word_match[[
|
||
|
define elif else endif error if ifdef ifndef import include line pragma undef
|
||
|
warning
|
||
|
]]
|
||
|
lex:add_rule('preprocessor', #lexer.starts_line('#') *
|
||
|
token(lexer.PREPROCESSOR, '#' * S('\t ')^0 *
|
||
|
preproc_word))
|
||
|
|
||
|
-- Operators.
|
||
|
lex:add_rule('operator', token(lexer.OPERATOR, S('+-/*%<>!=^&|?~:;.()[]{}')))
|
||
|
|
||
|
-- Fold symbols.
|
||
|
lex:add_fold_point(lexer.PREPROCESSOR, 'region', 'endregion')
|
||
|
lex:add_fold_point(lexer.PREPROCESSOR, 'if', 'endif')
|
||
|
lex:add_fold_point(lexer.PREPROCESSOR, 'ifdef', 'endif')
|
||
|
lex:add_fold_point(lexer.PREPROCESSOR, 'ifndef', 'endif')
|
||
|
lex:add_fold_point(lexer.OPERATOR, '{', '}')
|
||
|
lex:add_fold_point(lexer.COMMENT, '/*', '*/')
|
||
|
lex:add_fold_point(lexer.COMMENT, '//', lexer.fold_line_comments('//'))
|
||
|
|
||
|
return lex
|