/* GLIB - Library of useful routines for C programming * Copyright (C) 1995-1997 Peter Mattis, Spencer Kimball and Josh MacDonald * * GScanner: Flexible lexical scanner for general purpose. * Copyright (C) 1997, 1998 Tim Janik * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; either * version 2 of the License, or (at your option) any later version. * * This library is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public * License along with this library; if not, write to the * Free Software Foundation, Inc., 59 Temple Place - Suite 330, * Boston, MA 02111-1307, USA. */ /* * Modified by the GLib Team and others 1997-2000. See the AUTHORS * file for a list of people on the GLib Team. See the ChangeLog * files for a list of changes. These files are distributed with * GLib at ftp://ftp.gtk.org/pub/gtk/. */ /* * MT safe */ #include "config.h" #include #include #include #include #include #ifdef HAVE_UNISTD_H #include #endif #ifdef G_OS_WIN32 #include /* For _read() */ #endif #include "gscanner.h" #include "gprintfint.h" #include "gstrfuncs.h" #include "gstring.h" #include "gtestutils.h" /* --- defines --- */ #define to_lower(c) ( \ (guchar) ( \ ( (((guchar)(c))>='A' && ((guchar)(c))<='Z') * ('a'-'A') ) | \ ( (((guchar)(c))>=192 && ((guchar)(c))<=214) * (224-192) ) | \ ( (((guchar)(c))>=216 && ((guchar)(c))<=222) * (248-216) ) | \ ((guchar)(c)) \ ) \ ) #define READ_BUFFER_SIZE (4000) /* --- typedefs --- */ typedef struct _GScannerKey GScannerKey; struct _GScannerKey { guint scope_id; gchar *symbol; gpointer value; }; /* --- variables --- */ static const GScannerConfig g_scanner_config_template = { ( " \t\r\n" ) /* cset_skip_characters */, ( G_CSET_a_2_z "_" G_CSET_A_2_Z ) /* cset_identifier_first */, ( G_CSET_a_2_z "_" G_CSET_A_2_Z G_CSET_DIGITS G_CSET_LATINS G_CSET_LATINC ) /* cset_identifier_nth */, ( "#\n" ) /* cpair_comment_single */, FALSE /* case_sensitive */, TRUE /* skip_comment_multi */, TRUE /* skip_comment_single */, TRUE /* scan_comment_multi */, TRUE /* scan_identifier */, FALSE /* scan_identifier_1char */, FALSE /* scan_identifier_NULL */, TRUE /* scan_symbols */, FALSE /* scan_binary */, TRUE /* scan_octal */, TRUE /* scan_float */, TRUE /* scan_hex */, FALSE /* scan_hex_dollar */, TRUE /* scan_string_sq */, TRUE /* scan_string_dq */, TRUE /* numbers_2_int */, FALSE /* int_2_float */, FALSE /* identifier_2_string */, TRUE /* char_2_token */, FALSE /* symbol_2_token */, FALSE /* scope_0_fallback */, FALSE /* store_int64 */, }; /* --- prototypes --- */ static inline GScannerKey* g_scanner_lookup_internal (GScanner *scanner, guint scope_id, const gchar *symbol); static gboolean g_scanner_key_equal (gconstpointer v1, gconstpointer v2); static guint g_scanner_key_hash (gconstpointer v); static void g_scanner_get_token_ll (GScanner *scanner, GTokenType *token_p, GTokenValue *value_p, guint *line_p, guint *position_p); static void g_scanner_get_token_i (GScanner *scanner, GTokenType *token_p, GTokenValue *value_p, guint *line_p, guint *position_p); static guchar g_scanner_peek_next_char (GScanner *scanner); static guchar g_scanner_get_char (GScanner *scanner, guint *line_p, guint *position_p); static void g_scanner_msg_handler (GScanner *scanner, gchar *message, gboolean is_error); /* --- functions --- */ static inline gint g_scanner_char_2_num (guchar c, guchar base) { if (c >= '0' && c <= '9') c -= '0'; else if (c >= 'A' && c <= 'Z') c -= 'A' - 10; else if (c >= 'a' && c <= 'z') c -= 'a' - 10; else return -1; if (c < base) return c; return -1; } GScanner* g_scanner_new (const GScannerConfig *config_templ) { GScanner *scanner; if (!config_templ) config_templ = &g_scanner_config_template; scanner = g_new0 (GScanner, 1); scanner->user_data = NULL; scanner->max_parse_errors = 1; scanner->parse_errors = 0; scanner->input_name = NULL; g_datalist_init (&scanner->qdata); scanner->config = g_new0 (GScannerConfig, 1); scanner->config->case_sensitive = config_templ->case_sensitive; scanner->config->cset_skip_characters = config_templ->cset_skip_characters; if (!scanner->config->cset_skip_characters) scanner->config->cset_skip_characters = ""; scanner->config->cset_identifier_first = config_templ->cset_identifier_first; scanner->config->cset_identifier_nth = config_templ->cset_identifier_nth; scanner->config->cpair_comment_single = config_templ->cpair_comment_single; scanner->config->skip_comment_multi = config_templ->skip_comment_multi; scanner->config->skip_comment_single = config_templ->skip_comment_single; scanner->config->scan_comment_multi = config_templ->scan_comment_multi; scanner->config->scan_identifier = config_templ->scan_identifier; scanner->config->scan_identifier_1char = config_templ->scan_identifier_1char; scanner->config->scan_identifier_NULL = config_templ->scan_identifier_NULL; scanner->config->scan_symbols = config_templ->scan_symbols; scanner->config->scan_binary = config_templ->scan_binary; scanner->config->scan_octal = config_templ->scan_octal; scanner->config->scan_float = config_templ->scan_float; scanner->config->scan_hex = config_templ->scan_hex; scanner->config->scan_hex_dollar = config_templ->scan_hex_dollar; scanner->config->scan_string_sq = config_templ->scan_string_sq; scanner->config->scan_string_dq = config_templ->scan_string_dq; scanner->config->numbers_2_int = config_templ->numbers_2_int; scanner->config->int_2_float = config_templ->int_2_float; scanner->config->identifier_2_string = config_templ->identifier_2_string; scanner->config->char_2_token = config_templ->char_2_token; scanner->config->symbol_2_token = config_templ->symbol_2_token; scanner->config->scope_0_fallback = config_templ->scope_0_fallback; scanner->config->store_int64 = config_templ->store_int64; scanner->token = G_TOKEN_NONE; scanner->value.v_int64 = 0; scanner->line = 1; scanner->position = 0; scanner->next_token = G_TOKEN_NONE; scanner->next_value.v_int64 = 0; scanner->next_line = 1; scanner->next_position = 0; scanner->symbol_table = g_hash_table_new (g_scanner_key_hash, g_scanner_key_equal); scanner->input_fd = -1; scanner->text = NULL; scanner->text_end = NULL; scanner->buffer = NULL; scanner->scope_id = 0; scanner->msg_handler = g_scanner_msg_handler; return scanner; } static inline void g_scanner_free_value (GTokenType *token_p, GTokenValue *value_p) { switch (*token_p) { case G_TOKEN_STRING: case G_TOKEN_IDENTIFIER: case G_TOKEN_IDENTIFIER_NULL: case G_TOKEN_COMMENT_SINGLE: case G_TOKEN_COMMENT_MULTI: g_free (value_p->v_string); break; default: break; } *token_p = G_TOKEN_NONE; } static void g_scanner_destroy_symbol_table_entry (gpointer _key, gpointer _value, gpointer _data) { GScannerKey *key = _key; g_free (key->symbol); g_free (key); } void g_scanner_destroy (GScanner *scanner) { g_return_if_fail (scanner != NULL); g_datalist_clear (&scanner->qdata); g_hash_table_foreach (scanner->symbol_table, g_scanner_destroy_symbol_table_entry, NULL); g_hash_table_destroy (scanner->symbol_table); g_scanner_free_value (&scanner->token, &scanner->value); g_scanner_free_value (&scanner->next_token, &scanner->next_value); g_free (scanner->config); g_free (scanner->buffer); g_free (scanner); } static void g_scanner_msg_handler (GScanner *scanner, gchar *message, gboolean is_error) { g_return_if_fail (scanner != NULL); _g_fprintf (stderr, "%s:%d: ", scanner->input_name ? scanner->input_name : "", scanner->line); if (is_error) _g_fprintf (stderr, "error: "); _g_fprintf (stderr, "%s\n", message); } void g_scanner_error (GScanner *scanner, const gchar *format, ...) { g_return_if_fail (scanner != NULL); g_return_if_fail (format != NULL); scanner->parse_errors++; if (scanner->msg_handler) { va_list args; gchar *string; va_start (args, format); string = g_strdup_vprintf (format, args); va_end (args); scanner->msg_handler (scanner, string, TRUE); g_free (string); } } void g_scanner_warn (GScanner *scanner, const gchar *format, ...) { g_return_if_fail (scanner != NULL); g_return_if_fail (format != NULL); if (scanner->msg_handler) { va_list args; gchar *string; va_start (args, format); string = g_strdup_vprintf (format, args); va_end (args); scanner->msg_handler (scanner, string, FALSE); g_free (string); } } static gboolean g_scanner_key_equal (gconstpointer v1, gconstpointer v2) { const GScannerKey *key1 = v1; const GScannerKey *key2 = v2; return (key1->scope_id == key2->scope_id) && (strcmp (key1->symbol, key2->symbol) == 0); } static guint g_scanner_key_hash (gconstpointer v) { const GScannerKey *key = v; gchar *c; guint h; h = key->scope_id; for (c = key->symbol; *c; c++) h = (h << 5) - h + *c; return h; } static inline GScannerKey* g_scanner_lookup_internal (GScanner *scanner, guint scope_id, const gchar *symbol) { GScannerKey *key_p; GScannerKey key; key.scope_id = scope_id; if (!scanner->config->case_sensitive) { gchar *d; const gchar *c; key.symbol = g_new (gchar, strlen (symbol) + 1); for (d = key.symbol, c = symbol; *c; c++, d++) *d = to_lower (*c); *d = 0; key_p = g_hash_table_lookup (scanner->symbol_table, &key); g_free (key.symbol); } else { key.symbol = (gchar*) symbol; key_p = g_hash_table_lookup (scanner->symbol_table, &key); } return key_p; } void g_scanner_scope_add_symbol (GScanner *scanner, guint scope_id, const gchar *symbol, gpointer value) { GScannerKey *key; g_return_if_fail (scanner != NULL); g_return_if_fail (symbol != NULL); key = g_scanner_lookup_internal (scanner, scope_id, symbol); if (!key) { key = g_new (GScannerKey, 1); key->scope_id = scope_id; key->symbol = g_strdup (symbol); key->value = value; if (!scanner->config->case_sensitive) { gchar *c; c = key->symbol; while (*c != 0) { *c = to_lower (*c); c++; } } g_hash_table_insert (scanner->symbol_table, key, key); } else key->value = value; } void g_scanner_scope_remove_symbol (GScanner *scanner, guint scope_id, const gchar *symbol) { GScannerKey *key; g_return_if_fail (scanner != NULL); g_return_if_fail (symbol != NULL); key = g_scanner_lookup_internal (scanner, scope_id, symbol); if (key) { g_hash_table_remove (scanner->symbol_table, key); g_free (key->symbol); g_free (key); } } gpointer g_scanner_lookup_symbol (GScanner *scanner, const gchar *symbol) { GScannerKey *key; guint scope_id; g_return_val_if_fail (scanner != NULL, NULL); if (!symbol) return NULL; scope_id = scanner->scope_id; key = g_scanner_lookup_internal (scanner, scope_id, symbol); if (!key && scope_id && scanner->config->scope_0_fallback) key = g_scanner_lookup_internal (scanner, 0, symbol); if (key) return key->value; else return NULL; } gpointer g_scanner_scope_lookup_symbol (GScanner *scanner, guint scope_id, const gchar *symbol) { GScannerKey *key; g_return_val_if_fail (scanner != NULL, NULL); if (!symbol) return NULL; key = g_scanner_lookup_internal (scanner, scope_id, symbol); if (key) return key->value; else return NULL; } guint g_scanner_set_scope (GScanner *scanner, guint scope_id) { guint old_scope_id; g_return_val_if_fail (scanner != NULL, 0); old_scope_id = scanner->scope_id; scanner->scope_id = scope_id; return old_scope_id; } static void g_scanner_foreach_internal (gpointer _key, gpointer _value, gpointer _user_data) { GScannerKey *key; gpointer *d; GHFunc func; gpointer user_data; guint *scope_id; d = _user_data; func = (GHFunc) d[0]; user_data = d[1]; scope_id = d[2]; key = _value; if (key->scope_id == *scope_id) func (key->symbol, key->value, user_data); } void g_scanner_scope_foreach_symbol (GScanner *scanner, guint scope_id, GHFunc func, gpointer user_data) { gpointer d[3]; g_return_if_fail (scanner != NULL); d[0] = (gpointer) func; d[1] = user_data; d[2] = &scope_id; g_hash_table_foreach (scanner->symbol_table, g_scanner_foreach_internal, d); } GTokenType g_scanner_peek_next_token (GScanner *scanner) { g_return_val_if_fail (scanner != NULL, G_TOKEN_EOF); if (scanner->next_token == G_TOKEN_NONE) { scanner->next_line = scanner->line; scanner->next_position = scanner->position; g_scanner_get_token_i (scanner, &scanner->next_token, &scanner->next_value, &scanner->next_line, &scanner->next_position); } return scanner->next_token; } GTokenType g_scanner_get_next_token (GScanner *scanner) { g_return_val_if_fail (scanner != NULL, G_TOKEN_EOF); if (scanner->next_token != G_TOKEN_NONE) { g_scanner_free_value (&scanner->token, &scanner->value); scanner->token = scanner->next_token; scanner->value = scanner->next_value; scanner->line = scanner->next_line; scanner->position = scanner->next_position; scanner->next_token = G_TOKEN_NONE; } else g_scanner_get_token_i (scanner, &scanner->token, &scanner->value, &scanner->line, &scanner->position); return scanner->token; } GTokenType g_scanner_cur_token (GScanner *scanner) { g_return_val_if_fail (scanner != NULL, G_TOKEN_EOF); return scanner->token; } GTokenValue g_scanner_cur_value (GScanner *scanner) { GTokenValue v; v.v_int64 = 0; g_return_val_if_fail (scanner != NULL, v); /* MSC isn't capable of handling return scanner->value; ? */ v = scanner->value; return v; } guint g_scanner_cur_line (GScanner *scanner) { g_return_val_if_fail (scanner != NULL, 0); return scanner->line; } guint g_scanner_cur_position (GScanner *scanner) { g_return_val_if_fail (scanner != NULL, 0); return scanner->position; } gboolean g_scanner_eof (GScanner *scanner) { g_return_val_if_fail (scanner != NULL, TRUE); return scanner->token == G_TOKEN_EOF || scanner->token == G_TOKEN_ERROR; } void g_scanner_input_file (GScanner *scanner, gint input_fd) { g_return_if_fail (scanner != NULL); g_return_if_fail (input_fd >= 0); if (scanner->input_fd >= 0) g_scanner_sync_file_offset (scanner); scanner->token = G_TOKEN_NONE; scanner->value.v_int64 = 0; scanner->line = 1; scanner->position = 0; scanner->next_token = G_TOKEN_NONE; scanner->input_fd = input_fd; scanner->text = NULL; scanner->text_end = NULL; if (!scanner->buffer) scanner->buffer = g_new (gchar, READ_BUFFER_SIZE + 1); } void g_scanner_input_text (GScanner *scanner, const gchar *text, guint text_len) { g_return_if_fail (scanner != NULL); if (text_len) g_return_if_fail (text != NULL); else text = NULL; if (scanner->input_fd >= 0) g_scanner_sync_file_offset (scanner); scanner->token = G_TOKEN_NONE; scanner->value.v_int64 = 0; scanner->line = 1; scanner->position = 0; scanner->next_token = G_TOKEN_NONE; scanner->input_fd = -1; scanner->text = text; scanner->text_end = text + text_len; if (scanner->buffer) { g_free (scanner->buffer); scanner->buffer = NULL; } } static guchar g_scanner_peek_next_char (GScanner *scanner) { if (scanner->text < scanner->text_end) { return *scanner->text; } else if (scanner->input_fd >= 0) { gint count; gchar *buffer; buffer = scanner->buffer; do { count = read (scanner->input_fd, buffer, READ_BUFFER_SIZE); } while (count == -1 && (errno == EINTR || errno == EAGAIN)); if (count < 1) { scanner->input_fd = -1; return 0; } else { scanner->text = buffer; scanner->text_end = buffer + count; return *buffer; } } else return 0; } void g_scanner_sync_file_offset (GScanner *scanner) { g_return_if_fail (scanner != NULL); /* for file input, rewind the filedescriptor to the current * buffer position and blow the file read ahead buffer. useful * for third party uses of our file descriptor, which hooks * onto the current scanning position. */ if (scanner->input_fd >= 0 && scanner->text_end > scanner->text) { gint buffered; buffered = scanner->text_end - scanner->text; if (lseek (scanner->input_fd, - buffered, SEEK_CUR) >= 0) { /* we succeeded, blow our buffer's contents now */ scanner->text = NULL; scanner->text_end = NULL; } else errno = 0; } } static guchar g_scanner_get_char (GScanner *scanner, guint *line_p, guint *position_p) { guchar fchar; if (scanner->text < scanner->text_end) fchar = *(scanner->text++); else if (scanner->input_fd >= 0) { gint count; gchar *buffer; buffer = scanner->buffer; do { count = read (scanner->input_fd, buffer, READ_BUFFER_SIZE); } while (count == -1 && (errno == EINTR || errno == EAGAIN)); if (count < 1) { scanner->input_fd = -1; fchar = 0; } else { scanner->text = buffer + 1; scanner->text_end = buffer + count; fchar = *buffer; if (!fchar) { g_scanner_sync_file_offset (scanner); scanner->text_end = scanner->text; scanner->input_fd = -1; } } } else fchar = 0; if (fchar == '\n') { (*position_p) = 0; (*line_p)++; } else if (fchar) { (*position_p)++; } return fchar; } void g_scanner_unexp_token (GScanner *scanner, GTokenType expected_token, const gchar *identifier_spec, const gchar *symbol_spec, const gchar *symbol_name, const gchar *message, gint is_error) { gchar *token_string; guint token_string_len; gchar *expected_string; guint expected_string_len; gchar *message_prefix; gboolean print_unexp; void (*msg_handler) (GScanner*, const gchar*, ...); g_return_if_fail (scanner != NULL); if (is_error) msg_handler = g_scanner_error; else msg_handler = g_scanner_warn; if (!identifier_spec) identifier_spec = "identifier"; if (!symbol_spec) symbol_spec = "symbol"; token_string_len = 56; token_string = g_new (gchar, token_string_len + 1); expected_string_len = 64; expected_string = g_new (gchar, expected_string_len + 1); print_unexp = TRUE; switch (scanner->token) { case G_TOKEN_EOF: _g_snprintf (token_string, token_string_len, "end of file"); break; default: if (scanner->token >= 1 && scanner->token <= 255) { if ((scanner->token >= ' ' && scanner->token <= '~') || strchr (scanner->config->cset_identifier_first, scanner->token) || strchr (scanner->config->cset_identifier_nth, scanner->token)) _g_snprintf (token_string, token_string_len, "character `%c'", scanner->token); else _g_snprintf (token_string, token_string_len, "character `\\%o'", scanner->token); break; } else if (!scanner->config->symbol_2_token) { _g_snprintf (token_string, token_string_len, "(unknown) token <%d>", scanner->token); break; } /* fall through */ case G_TOKEN_SYMBOL: if (expected_token == G_TOKEN_SYMBOL || (scanner->config->symbol_2_token && expected_token > G_TOKEN_LAST)) print_unexp = FALSE; if (symbol_name) _g_snprintf (token_string, token_string_len, "%s%s `%s'", print_unexp ? "" : "invalid ", symbol_spec, symbol_name); else _g_snprintf (token_string, token_string_len, "%s%s", print_unexp ? "" : "invalid ", symbol_spec); break; case G_TOKEN_ERROR: print_unexp = FALSE; expected_token = G_TOKEN_NONE; switch (scanner->value.v_error) { case G_ERR_UNEXP_EOF: _g_snprintf (token_string, token_string_len, "scanner: unexpected end of file"); break; case G_ERR_UNEXP_EOF_IN_STRING: _g_snprintf (token_string, token_string_len, "scanner: unterminated string constant"); break; case G_ERR_UNEXP_EOF_IN_COMMENT: _g_snprintf (token_string, token_string_len, "scanner: unterminated comment"); break; case G_ERR_NON_DIGIT_IN_CONST: _g_snprintf (token_string, token_string_len, "scanner: non digit in constant"); break; case G_ERR_FLOAT_RADIX: _g_snprintf (token_string, token_string_len, "scanner: invalid radix for floating constant"); break; case G_ERR_FLOAT_MALFORMED: _g_snprintf (token_string, token_string_len, "scanner: malformed floating constant"); break; case G_ERR_DIGIT_RADIX: _g_snprintf (token_string, token_string_len, "scanner: digit is beyond radix"); break; case G_ERR_UNKNOWN: default: _g_snprintf (token_string, token_string_len, "scanner: unknown error"); break; } break; case G_TOKEN_CHAR: _g_snprintf (token_string, token_string_len, "character `%c'", scanner->value.v_char); break; case G_TOKEN_IDENTIFIER: case G_TOKEN_IDENTIFIER_NULL: if (expected_token == G_TOKEN_IDENTIFIER || expected_token == G_TOKEN_IDENTIFIER_NULL) print_unexp = FALSE; _g_snprintf (token_string, token_string_len, "%s%s `%s'", print_unexp ? "" : "invalid ", identifier_spec, scanner->token == G_TOKEN_IDENTIFIER ? scanner->value.v_string : "null"); break; case G_TOKEN_BINARY: case G_TOKEN_OCTAL: case G_TOKEN_INT: case G_TOKEN_HEX: if (scanner->config->store_int64) _g_snprintf (token_string, token_string_len, "number `%" G_GUINT64_FORMAT "'", scanner->value.v_int64); else _g_snprintf (token_string, token_string_len, "number `%lu'", scanner->value.v_int); break; case G_TOKEN_FLOAT: _g_snprintf (token_string, token_string_len, "number `%.3f'", scanner->value.v_float); break; case G_TOKEN_STRING: if (expected_token == G_TOKEN_STRING) print_unexp = FALSE; _g_snprintf (token_string, token_string_len, "%s%sstring constant \"%s\"", print_unexp ? "" : "invalid ", scanner->value.v_string[0] == 0 ? "empty " : "", scanner->value.v_string); token_string[token_string_len - 2] = '"'; token_string[token_string_len - 1] = 0; break; case G_TOKEN_COMMENT_SINGLE: case G_TOKEN_COMMENT_MULTI: _g_snprintf (token_string, token_string_len, "comment"); break; case G_TOKEN_NONE: /* somehow the user's parsing code is screwed, there isn't much * we can do about it. * Note, a common case to trigger this is * g_scanner_peek_next_token(); g_scanner_unexp_token(); * without an intermediate g_scanner_get_next_token(). */ g_assert_not_reached (); break; } switch (expected_token) { gboolean need_valid; gchar *tstring; case G_TOKEN_EOF: _g_snprintf (expected_string, expected_string_len, "end of file"); break; default: if (expected_token >= 1 && expected_token <= 255) { if ((expected_token >= ' ' && expected_token <= '~') || strchr (scanner->config->cset_identifier_first, expected_token) || strchr (scanner->config->cset_identifier_nth, expected_token)) _g_snprintf (expected_string, expected_string_len, "character `%c'", expected_token); else _g_snprintf (expected_string, expected_string_len, "character `\\%o'", expected_token); break; } else if (!scanner->config->symbol_2_token) { _g_snprintf (expected_string, expected_string_len, "(unknown) token <%d>", expected_token); break; } /* fall through */ case G_TOKEN_SYMBOL: need_valid = (scanner->token == G_TOKEN_SYMBOL || (scanner->config->symbol_2_token && scanner->token > G_TOKEN_LAST)); _g_snprintf (expected_string, expected_string_len, "%s%s", need_valid ? "valid " : "", symbol_spec); /* FIXME: should we attempt to lookup the symbol_name for symbol_2_token? */ break; case G_TOKEN_CHAR: _g_snprintf (expected_string, expected_string_len, "%scharacter", scanner->token == G_TOKEN_CHAR ? "valid " : ""); break; case G_TOKEN_BINARY: tstring = "binary"; _g_snprintf (expected_string, expected_string_len, "%snumber (%s)", scanner->token == expected_token ? "valid " : "", tstring); break; case G_TOKEN_OCTAL: tstring = "octal"; _g_snprintf (expected_string, expected_string_len, "%snumber (%s)", scanner->token == expected_token ? "valid " : "", tstring); break; case G_TOKEN_INT: tstring = "integer"; _g_snprintf (expected_string, expected_string_len, "%snumber (%s)", scanner->token == expected_token ? "valid " : "", tstring); break; case G_TOKEN_HEX: tstring = "hexadecimal"; _g_snprintf (expected_string, expected_string_len, "%snumber (%s)", scanner->token == expected_token ? "valid " : "", tstring); break; case G_TOKEN_FLOAT: tstring = "float"; _g_snprintf (expected_string, expected_string_len, "%snumber (%s)", scanner->token == expected_token ? "valid " : "", tstring); break; case G_TOKEN_STRING: _g_snprintf (expected_string, expected_string_len, "%sstring constant", scanner->token == G_TOKEN_STRING ? "valid " : ""); break; case G_TOKEN_IDENTIFIER: case G_TOKEN_IDENTIFIER_NULL: need_valid = (scanner->token == G_TOKEN_IDENTIFIER_NULL || scanner->token == G_TOKEN_IDENTIFIER); _g_snprintf (expected_string, expected_string_len, "%s%s", need_valid ? "valid " : "", identifier_spec); break; case G_TOKEN_COMMENT_SINGLE: tstring = "single-line"; _g_snprintf (expected_string, expected_string_len, "%scomment (%s)", scanner->token == expected_token ? "valid " : "", tstring); break; case G_TOKEN_COMMENT_MULTI: tstring = "multi-line"; _g_snprintf (expected_string, expected_string_len, "%scomment (%s)", scanner->token == expected_token ? "valid " : "", tstring); break; case G_TOKEN_NONE: case G_TOKEN_ERROR: /* this is handled upon printout */ break; } if (message && message[0] != 0) message_prefix = " - "; else { message_prefix = ""; message = ""; } if (expected_token == G_TOKEN_ERROR) { msg_handler (scanner, "failure around %s%s%s", token_string, message_prefix, message); } else if (expected_token == G_TOKEN_NONE) { if (print_unexp) msg_handler (scanner, "unexpected %s%s%s", token_string, message_prefix, message); else msg_handler (scanner, "%s%s%s", token_string, message_prefix, message); } else { if (print_unexp) msg_handler (scanner, "unexpected %s, expected %s%s%s", token_string, expected_string, message_prefix, message); else msg_handler (scanner, "%s, expected %s%s%s", token_string, expected_string, message_prefix, message); } g_free (token_string); g_free (expected_string); } static void g_scanner_get_token_i (GScanner *scanner, GTokenType *token_p, GTokenValue *value_p, guint *line_p, guint *position_p) { do { g_scanner_free_value (token_p, value_p); g_scanner_get_token_ll (scanner, token_p, value_p, line_p, position_p); } while (((*token_p > 0 && *token_p < 256) && strchr (scanner->config->cset_skip_characters, *token_p)) || (*token_p == G_TOKEN_CHAR && strchr (scanner->config->cset_skip_characters, value_p->v_char)) || (*token_p == G_TOKEN_COMMENT_MULTI && scanner->config->skip_comment_multi) || (*token_p == G_TOKEN_COMMENT_SINGLE && scanner->config->skip_comment_single)); switch (*token_p) { case G_TOKEN_IDENTIFIER: if (scanner->config->identifier_2_string) *token_p = G_TOKEN_STRING; break; case G_TOKEN_SYMBOL: if (scanner->config->symbol_2_token) *token_p = (GTokenType) value_p->v_symbol; break; case G_TOKEN_BINARY: case G_TOKEN_OCTAL: case G_TOKEN_HEX: if (scanner->config->numbers_2_int) *token_p = G_TOKEN_INT; break; default: break; } if (*token_p == G_TOKEN_INT && scanner->config->int_2_float) { *token_p = G_TOKEN_FLOAT; if (scanner->config->store_int64) { #ifdef _MSC_VER /* work around error C2520, see gvaluetransform.c */ value_p->v_float = (__int64)value_p->v_int64; #else value_p->v_float = value_p->v_int64; #endif } else value_p->v_float = value_p->v_int; } errno = 0; } static void g_scanner_get_token_ll (GScanner *scanner, GTokenType *token_p, GTokenValue *value_p, guint *line_p, guint *position_p) { GScannerConfig *config; GTokenType token; gboolean in_comment_multi; gboolean in_comment_single; gboolean in_string_sq; gboolean in_string_dq; GString *gstring; GTokenValue value; guchar ch; config = scanner->config; (*value_p).v_int64 = 0; if ((scanner->text >= scanner->text_end && scanner->input_fd < 0) || scanner->token == G_TOKEN_EOF) { *token_p = G_TOKEN_EOF; return; } in_comment_multi = FALSE; in_comment_single = FALSE; in_string_sq = FALSE; in_string_dq = FALSE; gstring = NULL; do /* while (ch != 0) */ { gboolean dotted_float = FALSE; ch = g_scanner_get_char (scanner, line_p, position_p); value.v_int64 = 0; token = G_TOKEN_NONE; /* this is *evil*, but needed ;( * we first check for identifier first character, because it * might interfere with other key chars like slashes or numbers */ if (config->scan_identifier && ch && strchr (config->cset_identifier_first, ch)) goto identifier_precedence; switch (ch) { case 0: token = G_TOKEN_EOF; (*position_p)++; /* ch = 0; */ break; case '/': if (!config->scan_comment_multi || g_scanner_peek_next_char (scanner) != '*') goto default_case; g_scanner_get_char (scanner, line_p, position_p); token = G_TOKEN_COMMENT_MULTI; in_comment_multi = TRUE; gstring = g_string_new (NULL); while ((ch = g_scanner_get_char (scanner, line_p, position_p)) != 0) { if (ch == '*' && g_scanner_peek_next_char (scanner) == '/') { g_scanner_get_char (scanner, line_p, position_p); in_comment_multi = FALSE; break; } else gstring = g_string_append_c (gstring, ch); } ch = 0; break; case '\'': if (!config->scan_string_sq) goto default_case; token = G_TOKEN_STRING; in_string_sq = TRUE; gstring = g_string_new (NULL); while ((ch = g_scanner_get_char (scanner, line_p, position_p)) != 0) { if (ch == '\'') { in_string_sq = FALSE; break; } else gstring = g_string_append_c (gstring, ch); } ch = 0; break; case '"': if (!config->scan_string_dq) goto default_case; token = G_TOKEN_STRING; in_string_dq = TRUE; gstring = g_string_new (NULL); while ((ch = g_scanner_get_char (scanner, line_p, position_p)) != 0) { if (ch == '"') { in_string_dq = FALSE; break; } else { if (ch == '\\') { ch = g_scanner_get_char (scanner, line_p, position_p); switch (ch) { guint i; guint fchar; case 0: break; case '\\': gstring = g_string_append_c (gstring, '\\'); break; case 'n': gstring = g_string_append_c (gstring, '\n'); break; case 't': gstring = g_string_append_c (gstring, '\t'); break; case 'r': gstring = g_string_append_c (gstring, '\r'); break; case 'b': gstring = g_string_append_c (gstring, '\b'); break; case 'f': gstring = g_string_append_c (gstring, '\f'); break; case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': i = ch - '0'; fchar = g_scanner_peek_next_char (scanner); if (fchar >= '0' && fchar <= '7') { ch = g_scanner_get_char (scanner, line_p, position_p); i = i * 8 + ch - '0'; fchar = g_scanner_peek_next_char (scanner); if (fchar >= '0' && fchar <= '7') { ch = g_scanner_get_char (scanner, line_p, position_p); i = i * 8 + ch - '0'; } } gstring = g_string_append_c (gstring, i); break; default: gstring = g_string_append_c (gstring, ch); break; } } else gstring = g_string_append_c (gstring, ch); } } ch = 0; break; case '.': if (!config->scan_float) goto default_case; token = G_TOKEN_FLOAT; dotted_float = TRUE; ch = g_scanner_get_char (scanner, line_p, position_p); goto number_parsing; case '$': if (!config->scan_hex_dollar) goto default_case; token = G_TOKEN_HEX; ch = g_scanner_get_char (scanner, line_p, position_p); goto number_parsing; case '0': if (config->scan_octal) token = G_TOKEN_OCTAL; else token = G_TOKEN_INT; ch = g_scanner_peek_next_char (scanner); if (config->scan_hex && (ch == 'x' || ch == 'X')) { token = G_TOKEN_HEX; g_scanner_get_char (scanner, line_p, position_p); ch = g_scanner_get_char (scanner, line_p, position_p); if (ch == 0) { token = G_TOKEN_ERROR; value.v_error = G_ERR_UNEXP_EOF; (*position_p)++; break; } if (g_scanner_char_2_num (ch, 16) < 0) { token = G_TOKEN_ERROR; value.v_error = G_ERR_DIGIT_RADIX; ch = 0; break; } } else if (config->scan_binary && (ch == 'b' || ch == 'B')) { token = G_TOKEN_BINARY; g_scanner_get_char (scanner, line_p, position_p); ch = g_scanner_get_char (scanner, line_p, position_p); if (ch == 0) { token = G_TOKEN_ERROR; value.v_error = G_ERR_UNEXP_EOF; (*position_p)++; break; } if (g_scanner_char_2_num (ch, 10) < 0) { token = G_TOKEN_ERROR; value.v_error = G_ERR_NON_DIGIT_IN_CONST; ch = 0; break; } } else ch = '0'; /* fall through */ case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': number_parsing: { gboolean in_number = TRUE; gchar *endptr; if (token == G_TOKEN_NONE) token = G_TOKEN_INT; gstring = g_string_new (dotted_float ? "0." : ""); gstring = g_string_append_c (gstring, ch); do /* while (in_number) */ { gboolean is_E; is_E = token == G_TOKEN_FLOAT && (ch == 'e' || ch == 'E'); ch = g_scanner_peek_next_char (scanner); if (g_scanner_char_2_num (ch, 36) >= 0 || (config->scan_float && ch == '.') || (is_E && (ch == '+' || ch == '-'))) { ch = g_scanner_get_char (scanner, line_p, position_p); switch (ch) { case '.': if (token != G_TOKEN_INT && token != G_TOKEN_OCTAL) { value.v_error = token == G_TOKEN_FLOAT ? G_ERR_FLOAT_MALFORMED : G_ERR_FLOAT_RADIX; token = G_TOKEN_ERROR; in_number = FALSE; } else { token = G_TOKEN_FLOAT; gstring = g_string_append_c (gstring, ch); } break; case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': gstring = g_string_append_c (gstring, ch); break; case '-': case '+': if (token != G_TOKEN_FLOAT) { token = G_TOKEN_ERROR; value.v_error = G_ERR_NON_DIGIT_IN_CONST; in_number = FALSE; } else gstring = g_string_append_c (gstring, ch); break; case 'e': case 'E': if ((token != G_TOKEN_HEX && !config->scan_float) || (token != G_TOKEN_HEX && token != G_TOKEN_OCTAL && token != G_TOKEN_FLOAT && token != G_TOKEN_INT)) { token = G_TOKEN_ERROR; value.v_error = G_ERR_NON_DIGIT_IN_CONST; in_number = FALSE; } else { if (token != G_TOKEN_HEX) token = G_TOKEN_FLOAT; gstring = g_string_append_c (gstring, ch); } break; default: if (token != G_TOKEN_HEX) { token = G_TOKEN_ERROR; value.v_error = G_ERR_NON_DIGIT_IN_CONST; in_number = FALSE; } else gstring = g_string_append_c (gstring, ch); break; } } else in_number = FALSE; } while (in_number); endptr = NULL; if (token == G_TOKEN_FLOAT) value.v_float = g_strtod (gstring->str, &endptr); else { guint64 ui64 = 0; switch (token) { case G_TOKEN_BINARY: ui64 = g_ascii_strtoull (gstring->str, &endptr, 2); break; case G_TOKEN_OCTAL: ui64 = g_ascii_strtoull (gstring->str, &endptr, 8); break; case G_TOKEN_INT: ui64 = g_ascii_strtoull (gstring->str, &endptr, 10); break; case G_TOKEN_HEX: ui64 = g_ascii_strtoull (gstring->str, &endptr, 16); break; default: ; } if (scanner->config->store_int64) value.v_int64 = ui64; else value.v_int = ui64; } if (endptr && *endptr) { token = G_TOKEN_ERROR; if (*endptr == 'e' || *endptr == 'E') value.v_error = G_ERR_NON_DIGIT_IN_CONST; else value.v_error = G_ERR_DIGIT_RADIX; } g_string_free (gstring, TRUE); gstring = NULL; ch = 0; } /* number_parsing:... */ break; default: default_case: { if (config->cpair_comment_single && ch == config->cpair_comment_single[0]) { token = G_TOKEN_COMMENT_SINGLE; in_comment_single = TRUE; gstring = g_string_new (NULL); ch = g_scanner_get_char (scanner, line_p, position_p); while (ch != 0) { if (ch == config->cpair_comment_single[1]) { in_comment_single = FALSE; ch = 0; break; } gstring = g_string_append_c (gstring, ch); ch = g_scanner_get_char (scanner, line_p, position_p); } /* ignore a missing newline at EOF for single line comments */ if (in_comment_single && config->cpair_comment_single[1] == '\n') in_comment_single = FALSE; } else if (config->scan_identifier && ch && strchr (config->cset_identifier_first, ch)) { identifier_precedence: if (config->cset_identifier_nth && ch && strchr (config->cset_identifier_nth, g_scanner_peek_next_char (scanner))) { token = G_TOKEN_IDENTIFIER; gstring = g_string_new (NULL); gstring = g_string_append_c (gstring, ch); do { ch = g_scanner_get_char (scanner, line_p, position_p); gstring = g_string_append_c (gstring, ch); ch = g_scanner_peek_next_char (scanner); } while (ch && strchr (config->cset_identifier_nth, ch)); ch = 0; } else if (config->scan_identifier_1char) { token = G_TOKEN_IDENTIFIER; value.v_identifier = g_new0 (gchar, 2); value.v_identifier[0] = ch; ch = 0; } } if (ch) { if (config->char_2_token) token = ch; else { token = G_TOKEN_CHAR; value.v_char = ch; } ch = 0; } } /* default_case:... */ break; } g_assert (ch == 0 && token != G_TOKEN_NONE); /* paranoid */ } while (ch != 0); if (in_comment_multi || in_comment_single || in_string_sq || in_string_dq) { token = G_TOKEN_ERROR; if (gstring) { g_string_free (gstring, TRUE); gstring = NULL; } (*position_p)++; if (in_comment_multi || in_comment_single) value.v_error = G_ERR_UNEXP_EOF_IN_COMMENT; else /* (in_string_sq || in_string_dq) */ value.v_error = G_ERR_UNEXP_EOF_IN_STRING; } if (gstring) { value.v_string = g_string_free (gstring, FALSE); gstring = NULL; } if (token == G_TOKEN_IDENTIFIER) { if (config->scan_symbols) { GScannerKey *key; guint scope_id; scope_id = scanner->scope_id; key = g_scanner_lookup_internal (scanner, scope_id, value.v_identifier); if (!key && scope_id && scanner->config->scope_0_fallback) key = g_scanner_lookup_internal (scanner, 0, value.v_identifier); if (key) { g_free (value.v_identifier); token = G_TOKEN_SYMBOL; value.v_symbol = key->value; } } if (token == G_TOKEN_IDENTIFIER && config->scan_identifier_NULL && strlen (value.v_identifier) == 4) { gchar *null_upper = "NULL"; gchar *null_lower = "null"; if (scanner->config->case_sensitive) { if (value.v_identifier[0] == null_upper[0] && value.v_identifier[1] == null_upper[1] && value.v_identifier[2] == null_upper[2] && value.v_identifier[3] == null_upper[3]) token = G_TOKEN_IDENTIFIER_NULL; } else { if ((value.v_identifier[0] == null_upper[0] || value.v_identifier[0] == null_lower[0]) && (value.v_identifier[1] == null_upper[1] || value.v_identifier[1] == null_lower[1]) && (value.v_identifier[2] == null_upper[2] || value.v_identifier[2] == null_lower[2]) && (value.v_identifier[3] == null_upper[3] || value.v_identifier[3] == null_lower[3])) token = G_TOKEN_IDENTIFIER_NULL; } } } *token_p = token; *value_p = value; }