ruby/yarp/regexp.c

581 строка
21 KiB
C

#include "yarp/regexp.h"
// This is the parser that is going to handle parsing regular expressions.
typedef struct {
const char *start;
const char *cursor;
const char *end;
yp_string_list_t *named_captures;
bool encoding_changed;
yp_encoding_t *encoding;
} yp_regexp_parser_t;
// This initializes a new parser with the given source.
static void
yp_regexp_parser_init(yp_regexp_parser_t *parser, const char *start, const char *end, yp_string_list_t *named_captures, bool encoding_changed, yp_encoding_t *encoding) {
*parser = (yp_regexp_parser_t) {
.start = start,
.cursor = start,
.end = end,
.named_captures = named_captures,
.encoding_changed = encoding_changed,
.encoding = encoding
};
}
// This appends a new string to the list of named captures.
static void
yp_regexp_parser_named_capture(yp_regexp_parser_t *parser, const char *start, const char *end) {
yp_string_t string;
yp_string_shared_init(&string, start, end);
yp_string_list_append(parser->named_captures, &string);
yp_string_free(&string);
}
// Returns true if the next character is the end of the source.
static inline bool
yp_regexp_char_is_eof(yp_regexp_parser_t *parser) {
return parser->cursor >= parser->end;
}
// Optionally accept a char and consume it if it exists.
static inline bool
yp_regexp_char_accept(yp_regexp_parser_t *parser, char value) {
if (!yp_regexp_char_is_eof(parser) && *parser->cursor == value) {
parser->cursor++;
return true;
}
return false;
}
// Expect a character to be present and consume it.
static inline bool
yp_regexp_char_expect(yp_regexp_parser_t *parser, char value) {
if (!yp_regexp_char_is_eof(parser) && *parser->cursor == value) {
parser->cursor++;
return true;
}
return false;
}
// This advances the current token to the next instance of the given character.
static bool
yp_regexp_char_find(yp_regexp_parser_t *parser, char value) {
if (yp_regexp_char_is_eof(parser)) {
return false;
}
const char *end = (const char *) yp_memchr(parser->cursor, value, (size_t) (parser->end - parser->cursor), parser->encoding_changed, parser->encoding);
if (end == NULL) {
return false;
}
parser->cursor = end + 1;
return true;
}
// Range quantifiers are a special class of quantifiers that look like
//
// * {digit}
// * {digit,}
// * {digit,digit}
// * {,digit}
//
// Unfortunately, if there are any spaces in between, then this just becomes a
// regular character match expression and we have to backtrack. So when this
// function first starts running, we'll create a "save" point and then attempt
// to parse the quantifier. If it fails, we'll restore the save point and
// return.
//
// The properly track everything, we're going to build a little state machine.
// It looks something like the following:
//
// ┌───────┐ ┌─────────┐ ────────────┐
// ──── lbrace ───> │ start │ ──── digit ───> │ minimum │ │
// └───────┘ └─────────┘ <─── digit ─┘
// │ │ │
// ┌───────┐ │ │ rbrace
// │ comma │ <───── comma ┌──── comma ───────┘ │
// └───────┘ V V
// │ ┌─────────┐ ┌─────────┐
// └── digit ──> │ maximum │ ── rbrace ──> │| final |│
// └─────────┘ └─────────┘
// │ ^
// └─ digit ─┘
//
// Note that by the time we've hit this function, the lbrace has already been
// consumed so we're in the start state.
static bool
yp_regexp_parse_range_quantifier(yp_regexp_parser_t *parser) {
const char *savepoint = parser->cursor;
enum {
YP_REGEXP_RANGE_QUANTIFIER_STATE_START,
YP_REGEXP_RANGE_QUANTIFIER_STATE_MINIMUM,
YP_REGEXP_RANGE_QUANTIFIER_STATE_MAXIMUM,
YP_REGEXP_RANGE_QUANTIFIER_STATE_COMMA
} state = YP_REGEXP_RANGE_QUANTIFIER_STATE_START;
while (1) {
switch (state) {
case YP_REGEXP_RANGE_QUANTIFIER_STATE_START:
switch (*parser->cursor) {
case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9':
parser->cursor++;
state = YP_REGEXP_RANGE_QUANTIFIER_STATE_MINIMUM;
break;
case ',':
parser->cursor++;
state = YP_REGEXP_RANGE_QUANTIFIER_STATE_COMMA;
break;
default:
parser->cursor = savepoint;
return true;
}
break;
case YP_REGEXP_RANGE_QUANTIFIER_STATE_MINIMUM:
switch (*parser->cursor) {
case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9':
parser->cursor++;
break;
case ',':
parser->cursor++;
state = YP_REGEXP_RANGE_QUANTIFIER_STATE_MAXIMUM;
break;
case '}':
parser->cursor++;
return true;
default:
parser->cursor = savepoint;
return true;
}
break;
case YP_REGEXP_RANGE_QUANTIFIER_STATE_COMMA:
switch (*parser->cursor) {
case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9':
parser->cursor++;
state = YP_REGEXP_RANGE_QUANTIFIER_STATE_MAXIMUM;
break;
default:
parser->cursor = savepoint;
return true;
}
break;
case YP_REGEXP_RANGE_QUANTIFIER_STATE_MAXIMUM:
switch (*parser->cursor) {
case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9':
parser->cursor++;
break;
case '}':
parser->cursor++;
return true;
default:
parser->cursor = savepoint;
return true;
}
break;
}
}
return true;
}
// quantifier : star-quantifier
// | plus-quantifier
// | optional-quantifier
// | range-quantifier
// | <empty>
// ;
static bool
yp_regexp_parse_quantifier(yp_regexp_parser_t *parser) {
switch (*parser->cursor) {
case '*':
case '+':
case '?':
parser->cursor++;
return true;
case '{':
parser->cursor++;
return yp_regexp_parse_range_quantifier(parser);
default:
// In this case there is no quantifier.
return true;
}
}
// match-posix-class : '[' '[' ':' '^'? CHAR+ ':' ']' ']'
// ;
static bool
yp_regexp_parse_posix_class(yp_regexp_parser_t *parser) {
if (!yp_regexp_char_expect(parser, ':')) {
return false;
}
yp_regexp_char_accept(parser, '^');
return (
yp_regexp_char_find(parser, ':') &&
yp_regexp_char_expect(parser, ']') &&
yp_regexp_char_expect(parser, ']')
);
}
// Forward declaration because character sets can be nested.
static bool
yp_regexp_parse_lbracket(yp_regexp_parser_t *parser);
// match-char-set : '[' '^'? (match-range | match-char)* ']'
// ;
static bool
yp_regexp_parse_character_set(yp_regexp_parser_t *parser) {
yp_regexp_char_accept(parser, '^');
while (!yp_regexp_char_is_eof(parser) && *parser->cursor != ']') {
switch (*parser->cursor++) {
case '[':
yp_regexp_parse_lbracket(parser);
break;
case '\\':
if (!yp_regexp_char_is_eof(parser)) {
parser->cursor++;
}
break;
default:
// do nothing, we've already advanced the cursor
break;
}
}
return yp_regexp_char_expect(parser, ']');
}
// A left bracket can either mean a POSIX class or a character set.
static bool
yp_regexp_parse_lbracket(yp_regexp_parser_t *parser) {
const char *reset = parser->cursor;
if ((parser->cursor + 2 < parser->end) && parser->cursor[0] == '[' && parser->cursor[1] == ':') {
parser->cursor++;
if (yp_regexp_parse_posix_class(parser)) return true;
parser->cursor = reset;
}
return yp_regexp_parse_character_set(parser);
}
// Forward declaration here since parsing groups needs to go back up the grammar
// to parse expressions within them.
static bool
yp_regexp_parse_expression(yp_regexp_parser_t *parser);
// These are the states of the options that are configurable on the regular
// expression (or from within a group).
typedef enum {
YP_REGEXP_OPTION_STATE_INVALID,
YP_REGEXP_OPTION_STATE_TOGGLEABLE,
YP_REGEXP_OPTION_STATE_ADDABLE,
YP_REGEXP_OPTION_STATE_ADDED,
YP_REGEXP_OPTION_STATE_REMOVED
} yp_regexp_option_state_t;
// These are the options that are configurable on the regular expression (or
// from within a group).
#define YP_REGEXP_OPTION_STATE_SLOT_MINIMUM 'a'
#define YP_REGEXP_OPTION_STATE_SLOT_MAXIMUM 'x'
#define YP_REGEXP_OPTION_STATE_SLOTS (YP_REGEXP_OPTION_STATE_SLOT_MAXIMUM - YP_REGEXP_OPTION_STATE_SLOT_MINIMUM + 1)
// This is the set of options that are configurable on the regular expression.
typedef struct {
unsigned char values[YP_REGEXP_OPTION_STATE_SLOTS];
} yp_regexp_options_t;
// Initialize a new set of options to their default values.
static void
yp_regexp_options_init(yp_regexp_options_t *options) {
memset(options, YP_REGEXP_OPTION_STATE_INVALID, sizeof(uint8_t) * YP_REGEXP_OPTION_STATE_SLOTS);
options->values['i' - YP_REGEXP_OPTION_STATE_SLOT_MINIMUM] = YP_REGEXP_OPTION_STATE_TOGGLEABLE;
options->values['m' - YP_REGEXP_OPTION_STATE_SLOT_MINIMUM] = YP_REGEXP_OPTION_STATE_TOGGLEABLE;
options->values['x' - YP_REGEXP_OPTION_STATE_SLOT_MINIMUM] = YP_REGEXP_OPTION_STATE_TOGGLEABLE;
options->values['d' - YP_REGEXP_OPTION_STATE_SLOT_MINIMUM] = YP_REGEXP_OPTION_STATE_ADDABLE;
options->values['a' - YP_REGEXP_OPTION_STATE_SLOT_MINIMUM] = YP_REGEXP_OPTION_STATE_ADDABLE;
options->values['u' - YP_REGEXP_OPTION_STATE_SLOT_MINIMUM] = YP_REGEXP_OPTION_STATE_ADDABLE;
}
// Attempt to add the given option to the set of options. Returns true if it was
// added, false if it was already present.
static bool
yp_regexp_options_add(yp_regexp_options_t *options, unsigned char key) {
if (key >= YP_REGEXP_OPTION_STATE_SLOT_MINIMUM && key <= YP_REGEXP_OPTION_STATE_SLOT_MAXIMUM) {
key = (unsigned char) (key - YP_REGEXP_OPTION_STATE_SLOT_MINIMUM);
switch (options->values[key]) {
case YP_REGEXP_OPTION_STATE_INVALID:
case YP_REGEXP_OPTION_STATE_REMOVED:
return false;
case YP_REGEXP_OPTION_STATE_TOGGLEABLE:
case YP_REGEXP_OPTION_STATE_ADDABLE:
options->values[key] = YP_REGEXP_OPTION_STATE_ADDED;
return true;
case YP_REGEXP_OPTION_STATE_ADDED:
return true;
}
}
return false;
}
// Attempt to remove the given option from the set of options. Returns true if
// it was removed, false if it was already absent.
static bool
yp_regexp_options_remove(yp_regexp_options_t *options, unsigned char key) {
if (key >= YP_REGEXP_OPTION_STATE_SLOT_MINIMUM && key <= YP_REGEXP_OPTION_STATE_SLOT_MAXIMUM) {
key = (unsigned char) (key - YP_REGEXP_OPTION_STATE_SLOT_MINIMUM);
switch (options->values[key]) {
case YP_REGEXP_OPTION_STATE_INVALID:
case YP_REGEXP_OPTION_STATE_ADDABLE:
return false;
case YP_REGEXP_OPTION_STATE_TOGGLEABLE:
case YP_REGEXP_OPTION_STATE_ADDED:
case YP_REGEXP_OPTION_STATE_REMOVED:
options->values[key] = YP_REGEXP_OPTION_STATE_REMOVED;
return true;
}
}
return false;
}
// Groups can have quite a few different patterns for syntax. They basically
// just wrap a set of expressions, but they can potentially have options after a
// question mark. If there _isn't_ a question mark, then it's just a set of
// expressions. If there _is_, then here are the options:
//
// * (?#...) - inline comments
// * (?:subexp) - non-capturing group
// * (?=subexp) - positive lookahead
// * (?!subexp) - negative lookahead
// * (?>subexp) - atomic group
// * (?~subexp) - absence operator
// * (?<=subexp) - positive lookbehind
// * (?<!subexp) - negative lookbehind
// * (?<name>subexp) - named capturing group
// * (?'name'subexp) - named capturing group
// * (?(cond)yes-subexp) - conditional expression
// * (?(cond)yes-subexp|no-subexp) - conditional expression
// * (?imxdau-imx) - turn on and off configuration
// * (?imxdau-imx:subexp) - turn on and off configuration for an expression
//
static bool
yp_regexp_parse_group(yp_regexp_parser_t *parser) {
// First, parse any options for the group.
if (yp_regexp_char_accept(parser, '?')) {
if (yp_regexp_char_is_eof(parser)) {
return false;
}
yp_regexp_options_t options;
yp_regexp_options_init(&options);
switch (*parser->cursor) {
case '#': { // inline comments
if (parser->encoding_changed && parser->encoding->multibyte) {
bool escaped = false;
// Here we're going to take a slow path and iterate through
// each multibyte character to find the close paren. We do
// this because \ can be a trailing byte in some encodings.
while (parser->cursor < parser->end) {
if (!escaped && *parser->cursor == ')') {
parser->cursor++;
return true;
}
size_t width = parser->encoding->char_width(parser->cursor, (ptrdiff_t) (parser->end - parser->cursor));
if (width == 0) return false;
escaped = (width == 1) && (*parser->cursor == '\\');
parser->cursor += width;
}
return false;
} else {
// Here we can take the fast path and use memchr to find the
// next ) because we are safe checking backward for \ since
// it cannot be a trailing character.
bool found = yp_regexp_char_find(parser, ')');
while (found && (parser->start <= parser->cursor - 2) && (*(parser->cursor - 2) == '\\')) {
found = yp_regexp_char_find(parser, ')');
}
return found;
}
}
case ':': // non-capturing group
case '=': // positive lookahead
case '!': // negative lookahead
case '>': // atomic group
case '~': // absence operator
parser->cursor++;
break;
case '<':
parser->cursor++;
if (yp_regexp_char_is_eof(parser)) {
return false;
}
switch (*parser->cursor) {
case '=': // positive lookbehind
case '!': // negative lookbehind
parser->cursor++;
break;
default: { // named capture group
const char *start = parser->cursor;
if (!yp_regexp_char_find(parser, '>')) {
return false;
}
yp_regexp_parser_named_capture(parser, start, parser->cursor - 1);
break;
}
}
break;
case '\'': { // named capture group
const char *start = ++parser->cursor;
if (!yp_regexp_char_find(parser, '\'')) {
return false;
}
yp_regexp_parser_named_capture(parser, start, parser->cursor - 1);
break;
}
case '(': // conditional expression
if (!yp_regexp_char_find(parser, ')')) {
return false;
}
break;
case 'i': case 'm': case 'x': case 'd': case 'a': case 'u': // options
while (!yp_regexp_char_is_eof(parser) && *parser->cursor != '-' && *parser->cursor != ':' && *parser->cursor != ')') {
if (!yp_regexp_options_add(&options, (unsigned char) *parser->cursor)) {
return false;
}
parser->cursor++;
}
if (yp_regexp_char_is_eof(parser)) {
return false;
}
// If we hit a -, then we're done parsing options.
if (*parser->cursor != '-') break;
// Otherwise, fallthrough to the - case.
/* fallthrough */
case '-':
parser->cursor++;
while (!yp_regexp_char_is_eof(parser) && *parser->cursor != ':' && *parser->cursor != ')') {
if (!yp_regexp_options_remove(&options, (unsigned char) *parser->cursor)) {
return false;
}
parser->cursor++;
}
if (yp_regexp_char_is_eof(parser)) {
return false;
}
break;
default:
return false;
}
}
// Now, parse the expressions within this group.
while (!yp_regexp_char_is_eof(parser) && *parser->cursor != ')') {
if (!yp_regexp_parse_expression(parser)) {
return false;
}
yp_regexp_char_accept(parser, '|');
}
// Finally, make sure we have a closing parenthesis.
return yp_regexp_char_expect(parser, ')');
}
// item : anchor
// | match-posix-class
// | match-char-set
// | match-char-class
// | match-char-prop
// | match-char
// | match-any
// | group
// | quantified
// ;
static bool
yp_regexp_parse_item(yp_regexp_parser_t *parser) {
switch (*parser->cursor++) {
case '^':
case '$':
return true;
case '\\':
if (!yp_regexp_char_is_eof(parser)) {
parser->cursor++;
}
return yp_regexp_parse_quantifier(parser);
case '(':
return yp_regexp_parse_group(parser) && yp_regexp_parse_quantifier(parser);
case '[':
return yp_regexp_parse_lbracket(parser) && yp_regexp_parse_quantifier(parser);
default:
return yp_regexp_parse_quantifier(parser);
}
}
// expression : item+
// ;
static bool
yp_regexp_parse_expression(yp_regexp_parser_t *parser) {
if (!yp_regexp_parse_item(parser)) {
return false;
}
while (!yp_regexp_char_is_eof(parser) && *parser->cursor != ')' && *parser->cursor != '|') {
if (!yp_regexp_parse_item(parser)) {
return false;
}
}
return true;
}
// pattern : EOF
// | expression EOF
// | expression '|' pattern
// ;
static bool
yp_regexp_parse_pattern(yp_regexp_parser_t *parser) {
return (
(
// Exit early if the pattern is empty.
yp_regexp_char_is_eof(parser) ||
// Parse the first expression in the pattern.
yp_regexp_parse_expression(parser)
) &&
(
// Return now if we've parsed the entire pattern.
yp_regexp_char_is_eof(parser) ||
// Otherwise, we should have a pipe character.
(yp_regexp_char_expect(parser, '|') && yp_regexp_parse_pattern(parser))
)
);
}
// Parse a regular expression and extract the names of all of the named capture
// groups.
YP_EXPORTED_FUNCTION bool
yp_regexp_named_capture_group_names(const char *source, size_t size, yp_string_list_t *named_captures, bool encoding_changed, yp_encoding_t *encoding) {
yp_regexp_parser_t parser;
yp_regexp_parser_init(&parser, source, source + size, named_captures, encoding_changed, encoding);
return yp_regexp_parse_pattern(&parser);
}