OVMS3/OVMS.V3/components/duktape/src-separate/duk_lexer.h

438 lines
18 KiB
C

/*
* Lexer defines.
*/
#if !defined(DUK_LEXER_H_INCLUDED)
#define DUK_LEXER_H_INCLUDED
typedef void (*duk_re_range_callback)(void *user, duk_codepoint_t r1, duk_codepoint_t r2, duk_bool_t direct);
/*
* A token is interpreted as any possible production of InputElementDiv
* and InputElementRegExp, see E5 Section 7 in its entirety. Note that
* the E5 "Token" production does not cover all actual tokens of the
* language (which is explicitly stated in the specification, Section 7.5).
* Null and boolean literals are defined as part of both ReservedWord
* (E5 Section 7.6.1) and Literal (E5 Section 7.8) productions. Here,
* null and boolean values have literal tokens, and are not reserved
* words.
*
* Decimal literal negative/positive sign is -not- part of DUK_TOK_NUMBER.
* The number tokens always have a non-negative value. The unary minus
* operator in "-1.0" is optimized during compilation to yield a single
* negative constant.
*
* Token numbering is free except that reserved words are required to be
* in a continuous range and in a particular order. See genstrings.py.
*/
#define DUK_LEXER_INITCTX(ctx) duk_lexer_initctx((ctx))
#define DUK_LEXER_SETPOINT(ctx,pt) duk_lexer_setpoint((ctx), (pt))
#define DUK_LEXER_GETPOINT(ctx,pt) duk_lexer_getpoint((ctx), (pt))
/* Currently 6 characters of lookup are actually needed (duk_lexer.c). */
#define DUK_LEXER_WINDOW_SIZE 6
#if defined(DUK_USE_LEXER_SLIDING_WINDOW)
#define DUK_LEXER_BUFFER_SIZE 64
#endif
#define DUK_TOK_MINVAL 0
/* returned after EOF (infinite amount) */
#define DUK_TOK_EOF 0
/* identifier names (E5 Section 7.6) */
#define DUK_TOK_IDENTIFIER 1
/* reserved words: keywords */
#define DUK_TOK_START_RESERVED 2
#define DUK_TOK_BREAK 2
#define DUK_TOK_CASE 3
#define DUK_TOK_CATCH 4
#define DUK_TOK_CONTINUE 5
#define DUK_TOK_DEBUGGER 6
#define DUK_TOK_DEFAULT 7
#define DUK_TOK_DELETE 8
#define DUK_TOK_DO 9
#define DUK_TOK_ELSE 10
#define DUK_TOK_FINALLY 11
#define DUK_TOK_FOR 12
#define DUK_TOK_FUNCTION 13
#define DUK_TOK_IF 14
#define DUK_TOK_IN 15
#define DUK_TOK_INSTANCEOF 16
#define DUK_TOK_NEW 17
#define DUK_TOK_RETURN 18
#define DUK_TOK_SWITCH 19
#define DUK_TOK_THIS 20
#define DUK_TOK_THROW 21
#define DUK_TOK_TRY 22
#define DUK_TOK_TYPEOF 23
#define DUK_TOK_VAR 24
#define DUK_TOK_CONST 25
#define DUK_TOK_VOID 26
#define DUK_TOK_WHILE 27
#define DUK_TOK_WITH 28
/* reserved words: future reserved words */
#define DUK_TOK_CLASS 29
#define DUK_TOK_ENUM 30
#define DUK_TOK_EXPORT 31
#define DUK_TOK_EXTENDS 32
#define DUK_TOK_IMPORT 33
#define DUK_TOK_SUPER 34
/* "null", "true", and "false" are always reserved words.
* Note that "get" and "set" are not!
*/
#define DUK_TOK_NULL 35
#define DUK_TOK_TRUE 36
#define DUK_TOK_FALSE 37
/* reserved words: additional future reserved words in strict mode */
#define DUK_TOK_START_STRICT_RESERVED 38 /* inclusive */
#define DUK_TOK_IMPLEMENTS 38
#define DUK_TOK_INTERFACE 39
#define DUK_TOK_LET 40
#define DUK_TOK_PACKAGE 41
#define DUK_TOK_PRIVATE 42
#define DUK_TOK_PROTECTED 43
#define DUK_TOK_PUBLIC 44
#define DUK_TOK_STATIC 45
#define DUK_TOK_YIELD 46
#define DUK_TOK_END_RESERVED 47 /* exclusive */
/* "get" and "set" are tokens but NOT ReservedWords. They are currently
* parsed and identifiers and these defines are actually now unused.
*/
#define DUK_TOK_GET 47
#define DUK_TOK_SET 48
/* punctuators (unlike the spec, also includes "/" and "/=") */
#define DUK_TOK_LCURLY 49
#define DUK_TOK_RCURLY 50
#define DUK_TOK_LBRACKET 51
#define DUK_TOK_RBRACKET 52
#define DUK_TOK_LPAREN 53
#define DUK_TOK_RPAREN 54
#define DUK_TOK_PERIOD 55
#define DUK_TOK_SEMICOLON 56
#define DUK_TOK_COMMA 57
#define DUK_TOK_LT 58
#define DUK_TOK_GT 59
#define DUK_TOK_LE 60
#define DUK_TOK_GE 61
#define DUK_TOK_EQ 62
#define DUK_TOK_NEQ 63
#define DUK_TOK_SEQ 64
#define DUK_TOK_SNEQ 65
#define DUK_TOK_ADD 66
#define DUK_TOK_SUB 67
#define DUK_TOK_MUL 68
#define DUK_TOK_DIV 69
#define DUK_TOK_MOD 70
#define DUK_TOK_EXP 71
#define DUK_TOK_INCREMENT 72
#define DUK_TOK_DECREMENT 73
#define DUK_TOK_ALSHIFT 74 /* named "arithmetic" because result is signed */
#define DUK_TOK_ARSHIFT 75
#define DUK_TOK_RSHIFT 76
#define DUK_TOK_BAND 77
#define DUK_TOK_BOR 78
#define DUK_TOK_BXOR 79
#define DUK_TOK_LNOT 80
#define DUK_TOK_BNOT 81
#define DUK_TOK_LAND 82
#define DUK_TOK_LOR 83
#define DUK_TOK_QUESTION 84
#define DUK_TOK_COLON 85
#define DUK_TOK_EQUALSIGN 86
#define DUK_TOK_ADD_EQ 87
#define DUK_TOK_SUB_EQ 88
#define DUK_TOK_MUL_EQ 89
#define DUK_TOK_DIV_EQ 90
#define DUK_TOK_MOD_EQ 91
#define DUK_TOK_EXP_EQ 92
#define DUK_TOK_ALSHIFT_EQ 93
#define DUK_TOK_ARSHIFT_EQ 94
#define DUK_TOK_RSHIFT_EQ 95
#define DUK_TOK_BAND_EQ 96
#define DUK_TOK_BOR_EQ 97
#define DUK_TOK_BXOR_EQ 98
/* literals (E5 Section 7.8), except null, true, false, which are treated
* like reserved words (above).
*/
#define DUK_TOK_NUMBER 99
#define DUK_TOK_STRING 100
#define DUK_TOK_REGEXP 101
#define DUK_TOK_MAXVAL 101 /* inclusive */
#define DUK_TOK_INVALID DUK_SMALL_UINT_MAX
/* Convert heap string index to a token (reserved words) */
#define DUK_STRIDX_TO_TOK(x) ((x) - DUK_STRIDX_START_RESERVED + DUK_TOK_START_RESERVED)
/* Sanity check */
#if (DUK_TOK_MAXVAL > 255)
#error DUK_TOK_MAXVAL too large, code assumes it fits into 8 bits
#endif
/* Sanity checks for string and token defines */
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_BREAK) != DUK_TOK_BREAK)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_CASE) != DUK_TOK_CASE)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_CATCH) != DUK_TOK_CATCH)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_CONTINUE) != DUK_TOK_CONTINUE)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_DEBUGGER) != DUK_TOK_DEBUGGER)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_DEFAULT) != DUK_TOK_DEFAULT)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_DELETE) != DUK_TOK_DELETE)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_DO) != DUK_TOK_DO)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_ELSE) != DUK_TOK_ELSE)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_FINALLY) != DUK_TOK_FINALLY)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_FOR) != DUK_TOK_FOR)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_LC_FUNCTION) != DUK_TOK_FUNCTION)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_IF) != DUK_TOK_IF)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_IN) != DUK_TOK_IN)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_INSTANCEOF) != DUK_TOK_INSTANCEOF)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_NEW) != DUK_TOK_NEW)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_RETURN) != DUK_TOK_RETURN)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_SWITCH) != DUK_TOK_SWITCH)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_THIS) != DUK_TOK_THIS)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_THROW) != DUK_TOK_THROW)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_TRY) != DUK_TOK_TRY)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_TYPEOF) != DUK_TOK_TYPEOF)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_VAR) != DUK_TOK_VAR)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_VOID) != DUK_TOK_VOID)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_WHILE) != DUK_TOK_WHILE)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_WITH) != DUK_TOK_WITH)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_CLASS) != DUK_TOK_CLASS)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_CONST) != DUK_TOK_CONST)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_ENUM) != DUK_TOK_ENUM)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_EXPORT) != DUK_TOK_EXPORT)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_EXTENDS) != DUK_TOK_EXTENDS)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_IMPORT) != DUK_TOK_IMPORT)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_SUPER) != DUK_TOK_SUPER)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_LC_NULL) != DUK_TOK_NULL)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_TRUE) != DUK_TOK_TRUE)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_FALSE) != DUK_TOK_FALSE)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_IMPLEMENTS) != DUK_TOK_IMPLEMENTS)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_INTERFACE) != DUK_TOK_INTERFACE)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_LET) != DUK_TOK_LET)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_PACKAGE) != DUK_TOK_PACKAGE)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_PRIVATE) != DUK_TOK_PRIVATE)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_PROTECTED) != DUK_TOK_PROTECTED)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_PUBLIC) != DUK_TOK_PUBLIC)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_STATIC) != DUK_TOK_STATIC)
#error mismatch in token defines
#endif
#if (DUK_STRIDX_TO_TOK(DUK_STRIDX_YIELD) != DUK_TOK_YIELD)
#error mismatch in token defines
#endif
/* Regexp tokens */
#define DUK_RETOK_EOF 0
#define DUK_RETOK_DISJUNCTION 1
#define DUK_RETOK_QUANTIFIER 2
#define DUK_RETOK_ASSERT_START 3
#define DUK_RETOK_ASSERT_END 4
#define DUK_RETOK_ASSERT_WORD_BOUNDARY 5
#define DUK_RETOK_ASSERT_NOT_WORD_BOUNDARY 6
#define DUK_RETOK_ASSERT_START_POS_LOOKAHEAD 7
#define DUK_RETOK_ASSERT_START_NEG_LOOKAHEAD 8
#define DUK_RETOK_ATOM_PERIOD 9
#define DUK_RETOK_ATOM_CHAR 10
#define DUK_RETOK_ATOM_DIGIT 11 /* assumptions in regexp compiler */
#define DUK_RETOK_ATOM_NOT_DIGIT 12 /* -""- */
#define DUK_RETOK_ATOM_WHITE 13 /* -""- */
#define DUK_RETOK_ATOM_NOT_WHITE 14 /* -""- */
#define DUK_RETOK_ATOM_WORD_CHAR 15 /* -""- */
#define DUK_RETOK_ATOM_NOT_WORD_CHAR 16 /* -""- */
#define DUK_RETOK_ATOM_BACKREFERENCE 17
#define DUK_RETOK_ATOM_START_CAPTURE_GROUP 18
#define DUK_RETOK_ATOM_START_NONCAPTURE_GROUP 19
#define DUK_RETOK_ATOM_START_CHARCLASS 20
#define DUK_RETOK_ATOM_START_CHARCLASS_INVERTED 21
#define DUK_RETOK_ATOM_END_GROUP 22
/* Constants for duk_lexer_ctx.buf. */
#define DUK_LEXER_TEMP_BUF_LIMIT 256
/* A token value. Can be memcpy()'d, but note that slot1/slot2 values are on the valstack.
* Some fields (like num, str1, str2) are only valid for specific token types and may have
* stale values otherwise.
*/
struct duk_token {
duk_small_uint_t t; /* token type (with reserved word identification) */
duk_small_uint_t t_nores; /* token type (with reserved words as DUK_TOK_IDENTIFER) */
duk_double_t num; /* numeric value of token */
duk_hstring *str1; /* string 1 of token (borrowed, stored to ctx->slot1_idx) */
duk_hstring *str2; /* string 2 of token (borrowed, stored to ctx->slot2_idx) */
duk_size_t start_offset; /* start byte offset of token in lexer input */
duk_int_t start_line; /* start line of token (first char) */
duk_int_t num_escapes; /* number of escapes and line continuations (for directive prologue) */
duk_bool_t lineterm; /* token was preceded by a lineterm */
duk_bool_t allow_auto_semi; /* token allows automatic semicolon insertion (eof or preceded by newline) */
};
#define DUK_RE_QUANTIFIER_INFINITE ((duk_uint32_t) 0xffffffffUL)
/* A regexp token value. */
struct duk_re_token {
duk_small_uint_t t; /* token type */
duk_small_uint_t greedy;
duk_uint32_t num; /* numeric value (character, count) */
duk_uint32_t qmin;
duk_uint32_t qmax;
};
/* A structure for 'snapshotting' a point for rewinding */
struct duk_lexer_point {
duk_size_t offset;
duk_int_t line;
};
/* Lexer codepoint with additional info like offset/line number */
struct duk_lexer_codepoint {
duk_codepoint_t codepoint;
duk_size_t offset;
duk_int_t line;
};
/* Lexer context. Same context is used for ECMAScript and Regexp parsing. */
struct duk_lexer_ctx {
#if defined(DUK_USE_LEXER_SLIDING_WINDOW)
duk_lexer_codepoint *window; /* unicode code points, window[0] is always next, points to 'buffer' */
duk_lexer_codepoint buffer[DUK_LEXER_BUFFER_SIZE];
#else
duk_lexer_codepoint window[DUK_LEXER_WINDOW_SIZE]; /* unicode code points, window[0] is always next */
#endif
duk_hthread *thr; /* thread; minimizes argument passing */
const duk_uint8_t *input; /* input string (may be a user pointer) */
duk_size_t input_length; /* input byte length */
duk_size_t input_offset; /* input offset for window leading edge (not window[0]) */
duk_int_t input_line; /* input linenumber at input_offset (not window[0]), init to 1 */
duk_idx_t slot1_idx; /* valstack slot for 1st token value */
duk_idx_t slot2_idx; /* valstack slot for 2nd token value */
duk_idx_t buf_idx; /* valstack slot for temp buffer */
duk_hbuffer_dynamic *buf; /* temp accumulation buffer */
duk_bufwriter_ctx bw; /* bufwriter for temp accumulation */
duk_int_t token_count; /* number of tokens parsed */
duk_int_t token_limit; /* maximum token count before error (sanity backstop) */
duk_small_uint_t flags; /* lexer flags, use compiler flag defines for now */
};
/*
* Prototypes
*/
DUK_INTERNAL_DECL void duk_lexer_initctx(duk_lexer_ctx *lex_ctx);
DUK_INTERNAL_DECL void duk_lexer_getpoint(duk_lexer_ctx *lex_ctx, duk_lexer_point *pt);
DUK_INTERNAL_DECL void duk_lexer_setpoint(duk_lexer_ctx *lex_ctx, duk_lexer_point *pt);
DUK_INTERNAL_DECL
void duk_lexer_parse_js_input_element(duk_lexer_ctx *lex_ctx,
duk_token *out_token,
duk_bool_t strict_mode,
duk_bool_t regexp_mode);
#if defined(DUK_USE_REGEXP_SUPPORT)
DUK_INTERNAL_DECL void duk_lexer_parse_re_token(duk_lexer_ctx *lex_ctx, duk_re_token *out_token);
DUK_INTERNAL_DECL void duk_lexer_parse_re_ranges(duk_lexer_ctx *lex_ctx, duk_re_range_callback gen_range, void *userdata);
#endif /* DUK_USE_REGEXP_SUPPORT */
#endif /* DUK_LEXER_H_INCLUDED */