diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-07 17:54:43 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-07 17:54:43 +0000 |
commit | e4283f6d48b98e764b988b43bbc86b9d52e6ec94 (patch) | |
tree | c8f7f7a6c2f5faa2942d27cefc6fd46cca492656 /src/st/croco/cr-statement.c | |
parent | Initial commit. (diff) | |
download | gnome-shell-e4283f6d48b98e764b988b43bbc86b9d52e6ec94.tar.xz gnome-shell-e4283f6d48b98e764b988b43bbc86b9d52e6ec94.zip |
Adding upstream version 43.9.upstream/43.9upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'src/st/croco/cr-statement.c')
-rw-r--r-- | src/st/croco/cr-statement.c | 2784 |
1 files changed, 2784 insertions, 0 deletions
diff --git a/src/st/croco/cr-statement.c b/src/st/croco/cr-statement.c new file mode 100644 index 0000000..eaeb49f --- /dev/null +++ b/src/st/croco/cr-statement.c @@ -0,0 +1,2784 @@ +/* -*- Mode: C; indent-tabs-mode: nil; c-basic-offset: 8 -*- */ + +/* + * This file is part of The Croco Library + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of version 2.1 of the GNU Lesser General Public + * License as published by the Free Software Foundation. + * + * This program 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 General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 + * USA + * + * Author: Dodji Seketeli. + * See COPYRIGHTS files for copyrights information. + */ + +#include <string.h> +#include "cr-statement.h" +#include "cr-parser.h" + +/** + *@file + *Definition of the #CRStatement class. + */ + +#define DECLARATION_INDENT_NB 2 + +static void cr_statement_clear (CRStatement * a_this); + +static void +parse_font_face_start_font_face_cb (CRDocHandler * a_this, + CRParsingLocation *a_location) +{ + CRStatement *stmt = NULL; + enum CRStatus status = CR_OK; + + stmt = cr_statement_new_at_font_face_rule (NULL, NULL); + g_return_if_fail (stmt); + + status = cr_doc_handler_set_ctxt (a_this, stmt); + g_return_if_fail (status == CR_OK); +} + +static void +parse_font_face_unrecoverable_error_cb (CRDocHandler * a_this) +{ + CRStatement *stmt = NULL; + CRStatement **stmtptr = NULL; + enum CRStatus status = CR_OK; + + g_return_if_fail (a_this); + + stmtptr = &stmt; + status = cr_doc_handler_get_ctxt (a_this, (gpointer *) stmtptr); + if (status != CR_OK) { + cr_utils_trace_info ("Couldn't get parsing context. " + "This may lead to some memory leaks."); + return; + } + if (stmt) { + cr_statement_destroy (stmt); + cr_doc_handler_set_ctxt (a_this, NULL); + return; + } +} + +static void +parse_font_face_property_cb (CRDocHandler * a_this, + CRString * a_name, + CRTerm * a_value, gboolean a_important) +{ + enum CRStatus status = CR_OK; + CRString *name = NULL; + CRDeclaration *decl = NULL; + CRStatement *stmt = NULL; + CRStatement **stmtptr = NULL; + + g_return_if_fail (a_this && a_name); + + stmtptr = &stmt; + status = cr_doc_handler_get_ctxt (a_this, (gpointer *) stmtptr); + g_return_if_fail (status == CR_OK && stmt); + g_return_if_fail (stmt->type == AT_FONT_FACE_RULE_STMT); + + name = cr_string_dup (a_name) ; + g_return_if_fail (name); + decl = cr_declaration_new (stmt, name, a_value); + if (!decl) { + cr_utils_trace_info ("cr_declaration_new () failed."); + goto error; + } + name = NULL; + + stmt->kind.font_face_rule->decl_list = + cr_declaration_append (stmt->kind.font_face_rule->decl_list, + decl); + if (!stmt->kind.font_face_rule->decl_list) + goto error; + decl = NULL; + + error: + if (decl) { + cr_declaration_unref (decl); + decl = NULL; + } + if (name) { + cr_string_destroy (name); + name = NULL; + } +} + +static void +parse_font_face_end_font_face_cb (CRDocHandler * a_this) +{ + CRStatement *result = NULL; + CRStatement **resultptr = NULL; + enum CRStatus status = CR_OK; + + g_return_if_fail (a_this); + + resultptr = &result; + status = cr_doc_handler_get_ctxt (a_this, (gpointer *) resultptr); + g_return_if_fail (status == CR_OK && result); + g_return_if_fail (result->type == AT_FONT_FACE_RULE_STMT); + + status = cr_doc_handler_set_result (a_this, result); + g_return_if_fail (status == CR_OK); +} + +static void +parse_page_start_page_cb (CRDocHandler * a_this, + CRString * a_name, + CRString * a_pseudo_page, + CRParsingLocation *a_location) +{ + CRStatement *stmt = NULL; + enum CRStatus status = CR_OK; + CRString *page_name = NULL, *pseudo_name = NULL ; + + if (a_name) + page_name = cr_string_dup (a_name) ; + if (a_pseudo_page) + pseudo_name = cr_string_dup (a_pseudo_page) ; + + stmt = cr_statement_new_at_page_rule (NULL, NULL, + page_name, + pseudo_name); + page_name = NULL ; + pseudo_name = NULL ; + g_return_if_fail (stmt); + status = cr_doc_handler_set_ctxt (a_this, stmt); + g_return_if_fail (status == CR_OK); +} + +static void +parse_page_unrecoverable_error_cb (CRDocHandler * a_this) +{ + CRStatement *stmt = NULL; + CRStatement **stmtptr = NULL; + enum CRStatus status = CR_OK; + + g_return_if_fail (a_this); + + stmtptr = &stmt; + status = cr_doc_handler_get_ctxt (a_this, (gpointer *) stmtptr); + if (status != CR_OK) { + cr_utils_trace_info ("Couldn't get parsing context. " + "This may lead to some memory leaks."); + return; + } + if (stmt) { + cr_statement_destroy (stmt); + stmt = NULL; + cr_doc_handler_set_ctxt (a_this, NULL); + } +} + +static void +parse_page_property_cb (CRDocHandler * a_this, + CRString * a_name, + CRTerm * a_expression, gboolean a_important) +{ + CRString *name = NULL; + CRStatement *stmt = NULL; + CRStatement **stmtptr = NULL; + CRDeclaration *decl = NULL; + enum CRStatus status = CR_OK; + + stmtptr = &stmt; + status = cr_doc_handler_get_ctxt (a_this, (gpointer *) stmtptr); + g_return_if_fail (status == CR_OK && stmt->type == AT_PAGE_RULE_STMT); + + name = cr_string_dup (a_name); + g_return_if_fail (name); + + decl = cr_declaration_new (stmt, name, a_expression); + g_return_if_fail (decl); + decl->important = a_important; + stmt->kind.page_rule->decl_list = + cr_declaration_append (stmt->kind.page_rule->decl_list, decl); + g_return_if_fail (stmt->kind.page_rule->decl_list); +} + +static void +parse_page_end_page_cb (CRDocHandler * a_this, + CRString * a_name, + CRString * a_pseudo_page) +{ + enum CRStatus status = CR_OK; + CRStatement *stmt = NULL; + CRStatement **stmtptr = NULL; + + stmtptr = &stmt; + status = cr_doc_handler_get_ctxt (a_this, (gpointer *) stmtptr); + g_return_if_fail (status == CR_OK && stmt); + g_return_if_fail (stmt->type == AT_PAGE_RULE_STMT); + + status = cr_doc_handler_set_result (a_this, stmt); + g_return_if_fail (status == CR_OK); +} + +static void +parse_at_media_start_media_cb (CRDocHandler * a_this, + GList * a_media_list, + CRParsingLocation *a_location) +{ + enum CRStatus status = CR_OK; + CRStatement *at_media = NULL; + GList *media_list = NULL; + + g_return_if_fail (a_this && a_this->priv); + + if (a_media_list) { + /*duplicate media list */ + media_list = cr_utils_dup_glist_of_cr_string + (a_media_list); + } + + g_return_if_fail (media_list); + + /*make sure cr_statement_new_at_media_rule works in this case. */ + at_media = cr_statement_new_at_media_rule (NULL, NULL, media_list); + + status = cr_doc_handler_set_ctxt (a_this, at_media); + g_return_if_fail (status == CR_OK); + status = cr_doc_handler_set_result (a_this, at_media); + g_return_if_fail (status == CR_OK); +} + +static void +parse_at_media_unrecoverable_error_cb (CRDocHandler * a_this) +{ + enum CRStatus status = CR_OK; + CRStatement *stmt = NULL; + CRStatement **stmtptr = NULL; + + g_return_if_fail (a_this); + + stmtptr = &stmt; + status = cr_doc_handler_get_result (a_this, (gpointer *) stmtptr); + if (status != CR_OK) { + cr_utils_trace_info ("Couldn't get parsing context. " + "This may lead to some memory leaks."); + return; + } + if (stmt) { + cr_statement_destroy (stmt); + stmt = NULL; + cr_doc_handler_set_ctxt (a_this, NULL); + cr_doc_handler_set_result (a_this, NULL); + } +} + +static void +parse_at_media_start_selector_cb (CRDocHandler * a_this, + CRSelector * a_sellist) +{ + enum CRStatus status = CR_OK; + CRStatement *at_media = NULL; + CRStatement **at_media_ptr = NULL; + CRStatement *ruleset = NULL; + + g_return_if_fail (a_this && a_this->priv && a_sellist); + + at_media_ptr = &at_media; + status = cr_doc_handler_get_ctxt (a_this, (gpointer *) at_media_ptr); + g_return_if_fail (status == CR_OK && at_media); + g_return_if_fail (at_media->type == AT_MEDIA_RULE_STMT); + ruleset = cr_statement_new_ruleset (NULL, a_sellist, NULL, at_media); + g_return_if_fail (ruleset); + status = cr_doc_handler_set_ctxt (a_this, ruleset); + g_return_if_fail (status == CR_OK); +} + +static void +parse_at_media_property_cb (CRDocHandler * a_this, + CRString * a_name, CRTerm * a_value, + gboolean a_important) +{ + enum CRStatus status = CR_OK; + + /* + *the current ruleset stmt, child of the + *current at-media being parsed. + */ + CRStatement *stmt = NULL; + CRStatement **stmtptr = NULL; + CRDeclaration *decl = NULL; + CRString *name = NULL; + + g_return_if_fail (a_this && a_name); + + name = cr_string_dup (a_name) ; + g_return_if_fail (name); + + stmtptr = &stmt; + status = cr_doc_handler_get_ctxt (a_this, + (gpointer *) stmtptr); + g_return_if_fail (status == CR_OK && stmt); + g_return_if_fail (stmt->type == RULESET_STMT); + + decl = cr_declaration_new (stmt, name, a_value); + g_return_if_fail (decl); + decl->important = a_important; + status = cr_statement_ruleset_append_decl (stmt, decl); + g_return_if_fail (status == CR_OK); +} + +static void +parse_at_media_end_selector_cb (CRDocHandler * a_this, + CRSelector * a_sellist) +{ + enum CRStatus status = CR_OK; + + /* + *the current ruleset stmt, child of the + *current at-media being parsed. + */ + CRStatement *stmt = NULL; + CRStatement **stmtptr = NULL; + + g_return_if_fail (a_this && a_sellist); + + stmtptr = &stmt; + status = cr_doc_handler_get_ctxt (a_this, (gpointer *) stmtptr); + g_return_if_fail (status == CR_OK && stmt + && stmt->type == RULESET_STMT); + g_return_if_fail (stmt->kind.ruleset->parent_media_rule); + + status = cr_doc_handler_set_ctxt + (a_this, stmt->kind.ruleset->parent_media_rule); + g_return_if_fail (status == CR_OK); +} + +static void +parse_at_media_end_media_cb (CRDocHandler * a_this, + GList * a_media_list) +{ + enum CRStatus status = CR_OK; + CRStatement *at_media = NULL; + CRStatement **at_media_ptr = NULL; + + g_return_if_fail (a_this && a_this->priv); + + at_media_ptr = &at_media; + status = cr_doc_handler_get_ctxt (a_this, + (gpointer *) at_media_ptr); + g_return_if_fail (status == CR_OK && at_media); + status = cr_doc_handler_set_result (a_this, at_media); +} + +static void +parse_ruleset_start_selector_cb (CRDocHandler * a_this, + CRSelector * a_sellist) +{ + CRStatement *ruleset = NULL; + + g_return_if_fail (a_this && a_this->priv && a_sellist); + + ruleset = cr_statement_new_ruleset (NULL, a_sellist, NULL, NULL); + g_return_if_fail (ruleset); + + cr_doc_handler_set_result (a_this, ruleset); +} + +static void +parse_ruleset_unrecoverable_error_cb (CRDocHandler * a_this) +{ + CRStatement *stmt = NULL; + CRStatement **stmtptr = NULL; + enum CRStatus status = CR_OK; + + stmtptr = &stmt; + status = cr_doc_handler_get_result (a_this, (gpointer *) stmtptr); + if (status != CR_OK) { + cr_utils_trace_info ("Couldn't get parsing context. " + "This may lead to some memory leaks."); + return; + } + if (stmt) { + cr_statement_destroy (stmt); + stmt = NULL; + cr_doc_handler_set_result (a_this, NULL); + } +} + +static void +parse_ruleset_property_cb (CRDocHandler * a_this, + CRString * a_name, + CRTerm * a_value, gboolean a_important) +{ + enum CRStatus status = CR_OK; + CRStatement *ruleset = NULL; + CRStatement **rulesetptr = NULL; + CRDeclaration *decl = NULL; + CRString *stringue = NULL; + + g_return_if_fail (a_this && a_this->priv && a_name); + + stringue = cr_string_dup (a_name); + g_return_if_fail (stringue); + + rulesetptr = &ruleset; + status = cr_doc_handler_get_result (a_this, (gpointer *) rulesetptr); + g_return_if_fail (status == CR_OK + && ruleset + && ruleset->type == RULESET_STMT); + + decl = cr_declaration_new (ruleset, stringue, a_value); + g_return_if_fail (decl); + decl->important = a_important; + status = cr_statement_ruleset_append_decl (ruleset, decl); + g_return_if_fail (status == CR_OK); +} + +static void +parse_ruleset_end_selector_cb (CRDocHandler * a_this, + CRSelector * a_sellist) +{ + CRStatement *result = NULL; + CRStatement **resultptr = NULL; + enum CRStatus status = CR_OK; + + g_return_if_fail (a_this && a_sellist); + + resultptr = &result; + status = cr_doc_handler_get_result (a_this, (gpointer *) resultptr); + + g_return_if_fail (status == CR_OK + && result + && result->type == RULESET_STMT); +} + +static void +cr_statement_clear (CRStatement * a_this) +{ + g_return_if_fail (a_this); + + switch (a_this->type) { + case AT_RULE_STMT: + break; + case RULESET_STMT: + if (!a_this->kind.ruleset) + return; + if (a_this->kind.ruleset->sel_list) { + cr_selector_unref (a_this->kind.ruleset->sel_list); + a_this->kind.ruleset->sel_list = NULL; + } + if (a_this->kind.ruleset->decl_list) { + cr_declaration_destroy + (a_this->kind.ruleset->decl_list); + a_this->kind.ruleset->decl_list = NULL; + } + g_free (a_this->kind.ruleset); + a_this->kind.ruleset = NULL; + break; + + case AT_IMPORT_RULE_STMT: + if (!a_this->kind.import_rule) + return; + if (a_this->kind.import_rule->url) { + cr_string_destroy + (a_this->kind.import_rule->url) ; + a_this->kind.import_rule->url = NULL; + } + g_free (a_this->kind.import_rule); + a_this->kind.import_rule = NULL; + break; + + case AT_MEDIA_RULE_STMT: + if (!a_this->kind.media_rule) + return; + if (a_this->kind.media_rule->rulesets) { + cr_statement_destroy + (a_this->kind.media_rule->rulesets); + a_this->kind.media_rule->rulesets = NULL; + } + if (a_this->kind.media_rule->media_list) { + GList *cur = NULL; + + for (cur = a_this->kind.media_rule->media_list; + cur; cur = cur->next) { + if (cur->data) { + cr_string_destroy ((CRString *) cur->data); + cur->data = NULL; + } + + } + g_list_free (a_this->kind.media_rule->media_list); + a_this->kind.media_rule->media_list = NULL; + } + g_free (a_this->kind.media_rule); + a_this->kind.media_rule = NULL; + break; + + case AT_PAGE_RULE_STMT: + if (!a_this->kind.page_rule) + return; + + if (a_this->kind.page_rule->decl_list) { + cr_declaration_destroy + (a_this->kind.page_rule->decl_list); + a_this->kind.page_rule->decl_list = NULL; + } + if (a_this->kind.page_rule->name) { + cr_string_destroy + (a_this->kind.page_rule->name); + a_this->kind.page_rule->name = NULL; + } + if (a_this->kind.page_rule->pseudo) { + cr_string_destroy + (a_this->kind.page_rule->pseudo); + a_this->kind.page_rule->pseudo = NULL; + } + g_free (a_this->kind.page_rule); + a_this->kind.page_rule = NULL; + break; + + case AT_CHARSET_RULE_STMT: + if (!a_this->kind.charset_rule) + return; + + if (a_this->kind.charset_rule->charset) { + cr_string_destroy + (a_this->kind.charset_rule->charset); + a_this->kind.charset_rule->charset = NULL; + } + g_free (a_this->kind.charset_rule); + a_this->kind.charset_rule = NULL; + break; + + case AT_FONT_FACE_RULE_STMT: + if (!a_this->kind.font_face_rule) + return; + + if (a_this->kind.font_face_rule->decl_list) { + cr_declaration_unref + (a_this->kind.font_face_rule->decl_list); + a_this->kind.font_face_rule->decl_list = NULL; + } + g_free (a_this->kind.font_face_rule); + a_this->kind.font_face_rule = NULL; + break; + + default: + break; + } +} + +/** + * cr_statement_ruleset_to_string: + * + *@a_this: the current instance of #CRStatement + *@a_indent: the number of whitespace to use for indentation + * + *Serializes the ruleset statement into a string + * + *Returns the newly allocated serialised string. Must be freed + *by the caller, using g_free(). + */ +static gchar * +cr_statement_ruleset_to_string (CRStatement const * a_this, glong a_indent) +{ + GString *stringue = NULL; + gchar *tmp_str = NULL, + *result = NULL; + + g_return_val_if_fail (a_this && a_this->type == RULESET_STMT, NULL); + + stringue = g_string_new (NULL); + + if (a_this->kind.ruleset->sel_list) { + if (a_indent) + cr_utils_dump_n_chars2 (' ', stringue, a_indent); + + tmp_str = + (gchar *) cr_selector_to_string (a_this->kind.ruleset-> + sel_list); + if (tmp_str) { + g_string_append (stringue, tmp_str); + g_free (tmp_str); + tmp_str = NULL; + } + } + g_string_append (stringue, " {\n"); + if (a_this->kind.ruleset->decl_list) { + tmp_str = (gchar *) cr_declaration_list_to_string2 + (a_this->kind.ruleset->decl_list, + a_indent + DECLARATION_INDENT_NB, TRUE); + if (tmp_str) { + g_string_append (stringue, tmp_str); + g_free (tmp_str); + tmp_str = NULL; + } + g_string_append (stringue, "\n"); + cr_utils_dump_n_chars2 (' ', stringue, a_indent); + } + g_string_append (stringue, "}"); + result = g_string_free (stringue, FALSE); + + if (tmp_str) { + g_free (tmp_str); + tmp_str = NULL; + } + return result; +} + + +/** + * cr_statement_font_face_rule_to_string: + * + *@a_this: the current instance of #CRStatement to consider + *It must be a font face rule statement. + *@a_indent: the number of white spaces of indentation. + * + *Serializes a font face rule statement into a string. + * + *Returns the serialized string. Must be deallocated by the caller + *using g_free(). + */ +static gchar * +cr_statement_font_face_rule_to_string (CRStatement const * a_this, + glong a_indent) +{ + gchar *result = NULL, *tmp_str = NULL ; + GString *stringue = NULL ; + + g_return_val_if_fail (a_this + && a_this->type == AT_FONT_FACE_RULE_STMT, + NULL); + + if (a_this->kind.font_face_rule->decl_list) { + stringue = g_string_new (NULL) ; + g_return_val_if_fail (stringue, NULL) ; + if (a_indent) + cr_utils_dump_n_chars2 (' ', stringue, + a_indent); + g_string_append (stringue, "@font-face {\n"); + tmp_str = (gchar *) cr_declaration_list_to_string2 + (a_this->kind.font_face_rule->decl_list, + a_indent + DECLARATION_INDENT_NB, TRUE) ; + if (tmp_str) { + g_string_append (stringue, + tmp_str) ; + g_free (tmp_str) ; + tmp_str = NULL ; + } + g_string_append (stringue, "\n}"); + } + if (stringue) { + result = g_string_free (stringue, FALSE); + stringue = NULL ; + } + return result ; +} + + +/** + * cr_statement_charset_to_string: + * + *Serialises an \@charset statement into a string. + *@a_this: the statement to serialize. + *@a_indent: the number of indentation spaces + * + *Returns the serialized charset statement. Must be + *freed by the caller using g_free(). + */ +static gchar * +cr_statement_charset_to_string (CRStatement const *a_this, + gulong a_indent) +{ + gchar *str = NULL ; + GString *stringue = NULL ; + + g_return_val_if_fail (a_this + && a_this->type == AT_CHARSET_RULE_STMT, + NULL) ; + + if (a_this->kind.charset_rule + && a_this->kind.charset_rule->charset + && a_this->kind.charset_rule->charset->stryng + && a_this->kind.charset_rule->charset->stryng->str) { + str = g_strndup (a_this->kind.charset_rule->charset->stryng->str, + a_this->kind.charset_rule->charset->stryng->len); + g_return_val_if_fail (str, NULL); + stringue = g_string_new (NULL) ; + g_return_val_if_fail (stringue, NULL) ; + cr_utils_dump_n_chars2 (' ', stringue, a_indent); + g_string_append_printf (stringue, + "@charset \"%s\" ;", str); + if (str) { + g_free (str); + str = NULL; + } + } + if (stringue) { + str = g_string_free (stringue, FALSE); + } + return str ; +} + + +/** + * cr_statement_at_page_rule_to_string: + * + *Serialises the at page rule statement into a string + *@a_this: the current instance of #CRStatement. Must + *be an "\@page" rule statement. + * + *Returns the serialized string. Must be freed by the caller + */ +static gchar * +cr_statement_at_page_rule_to_string (CRStatement const *a_this, + gulong a_indent) +{ + GString *stringue = NULL; + gchar *result = NULL ; + + stringue = g_string_new (NULL) ; + + cr_utils_dump_n_chars2 (' ', stringue, a_indent) ; + g_string_append (stringue, "@page"); + if (a_this->kind.page_rule->name + && a_this->kind.page_rule->name->stryng) { + g_string_append_printf + (stringue, " %s", + a_this->kind.page_rule->name->stryng->str) ; + } else { + g_string_append (stringue, " "); + } + if (a_this->kind.page_rule->pseudo + && a_this->kind.page_rule->pseudo->stryng) { + g_string_append_printf + (stringue, " :%s", + a_this->kind.page_rule->pseudo->stryng->str) ; + } + if (a_this->kind.page_rule->decl_list) { + gchar *str = NULL ; + g_string_append (stringue, " {\n"); + str = (gchar *) cr_declaration_list_to_string2 + (a_this->kind.page_rule->decl_list, + a_indent + DECLARATION_INDENT_NB, TRUE) ; + if (str) { + g_string_append (stringue, str) ; + g_free (str) ; + str = NULL ; + } + g_string_append (stringue, "\n}\n"); + } + result = g_string_free (stringue, FALSE) ; + stringue = NULL ; + return result ; +} + + +/** + *Serializes an \@media statement. + *@param a_this the current instance of #CRStatement + *@param a_indent the number of spaces of indentation. + *@return the serialized \@media statement. Must be freed + *by the caller using g_free(). + */ +static gchar * +cr_statement_media_rule_to_string (CRStatement const *a_this, + gulong a_indent) +{ + gchar *str = NULL ; + GString *stringue = NULL ; + GList const *cur = NULL; + + g_return_val_if_fail (a_this->type == AT_MEDIA_RULE_STMT, + NULL); + + if (a_this->kind.media_rule) { + stringue = g_string_new (NULL) ; + cr_utils_dump_n_chars2 (' ', stringue, a_indent); + g_string_append (stringue, "@media"); + + for (cur = a_this->kind.media_rule->media_list; cur; + cur = cur->next) { + if (cur->data) { + gchar *str2 = cr_string_dup2 + ((CRString const *) cur->data); + + if (str2) { + if (cur->prev) { + g_string_append + (stringue, + ","); + } + g_string_append_printf + (stringue, + " %s", str2); + g_free (str2); + str2 = NULL; + } + } + } + g_string_append (stringue, " {\n"); + str = cr_statement_list_to_string + (a_this->kind.media_rule->rulesets, + a_indent + DECLARATION_INDENT_NB) ; + if (str) { + g_string_append (stringue, str) ; + g_free (str) ; + str = NULL ; + } + g_string_append (stringue, "\n}"); + } + if (stringue) { + str = g_string_free (stringue, FALSE) ; + } + return str ; +} + + +static gchar * +cr_statement_import_rule_to_string (CRStatement const *a_this, + gulong a_indent) +{ + GString *stringue = NULL ; + gchar *str = NULL; + + g_return_val_if_fail (a_this + && a_this->type == AT_IMPORT_RULE_STMT + && a_this->kind.import_rule, + NULL) ; + + if (a_this->kind.import_rule->url + && a_this->kind.import_rule->url->stryng) { + stringue = g_string_new (NULL) ; + g_return_val_if_fail (stringue, NULL) ; + str = g_strndup (a_this->kind.import_rule->url->stryng->str, + a_this->kind.import_rule->url->stryng->len); + cr_utils_dump_n_chars2 (' ', stringue, a_indent); + if (str) { + g_string_append_printf (stringue, + "@import url(\"%s\")", + str); + g_free (str); + str = NULL ; + } else /*there is no url, so no import rule, get out! */ + return NULL; + + if (a_this->kind.import_rule->media_list) { + GList const *cur = NULL; + + for (cur = a_this->kind.import_rule->media_list; + cur; cur = cur->next) { + if (cur->data) { + CRString const *crstr = cur->data; + + if (cur->prev) { + g_string_append + (stringue, ", "); + } + if (crstr + && crstr->stryng + && crstr->stryng->str) { + g_string_append_len + (stringue, + crstr->stryng->str, + crstr->stryng->len) ; + } + } + } + } + g_string_append (stringue, " ;"); + } + if (stringue) { + str = g_string_free (stringue, FALSE) ; + stringue = NULL ; + } + return str ; +} + + +/******************* + *public functions + ******************/ + +/** + * cr_statement_does_buf_parses_against_core: + * + *@a_buf: the buffer to parse. + *@a_encoding: the character encoding of a_buf. + * + *Tries to parse a buffer and says whether if the content of the buffer + *is a css statement as defined by the "Core CSS Grammar" (chapter 4 of the + *css spec) or not. + * + *Returns TRUE if the buffer parses against the core grammar, false otherwise. + */ +gboolean +cr_statement_does_buf_parses_against_core (const guchar * a_buf, + enum CREncoding a_encoding) +{ + CRParser *parser = NULL; + enum CRStatus status = CR_OK; + gboolean result = FALSE; + + parser = cr_parser_new_from_buf ((guchar*)a_buf, strlen ((const char *) a_buf), + a_encoding, FALSE); + g_return_val_if_fail (parser, FALSE); + + status = cr_parser_set_use_core_grammar (parser, TRUE); + if (status != CR_OK) { + goto cleanup; + } + + status = cr_parser_parse_statement_core (parser); + if (status == CR_OK) { + result = TRUE; + } + + cleanup: + if (parser) { + cr_parser_destroy (parser); + } + + return result; +} + +/** + * cr_statement_parse_from_buf: + * + *@a_buf: the buffer to parse. + *@a_encoding: the character encoding of a_buf. + * + *Parses a buffer that contains a css statement and returns + *an instance of #CRStatement in case of successful parsing. + *TODO: at support of "\@import" rules. + * + *Returns the newly built instance of #CRStatement in case + *of successful parsing, NULL otherwise. + */ +CRStatement * +cr_statement_parse_from_buf (const guchar * a_buf, enum CREncoding a_encoding) +{ + CRStatement *result = NULL; + + /* + *The strategy of this function is "brute force". + *It tries to parse all the types of CRStatement it knows about. + *I could do this a smarter way but I don't have the time now. + *I think I will revisit this when time of performances and + *pull based incremental parsing comes. + */ + + result = cr_statement_ruleset_parse_from_buf (a_buf, a_encoding); + if (!result) { + result = cr_statement_at_charset_rule_parse_from_buf + (a_buf, a_encoding); + } else { + goto out; + } + + if (!result) { + result = cr_statement_at_media_rule_parse_from_buf + (a_buf, a_encoding); + } else { + goto out; + } + + if (!result) { + result = cr_statement_at_charset_rule_parse_from_buf + (a_buf, a_encoding); + } else { + goto out; + } + + if (!result) { + result = cr_statement_font_face_rule_parse_from_buf + (a_buf, a_encoding); + + } else { + goto out; + } + + if (!result) { + result = cr_statement_at_page_rule_parse_from_buf + (a_buf, a_encoding); + } else { + goto out; + } + + if (!result) { + result = cr_statement_at_import_rule_parse_from_buf + (a_buf, a_encoding); + } else { + goto out; + } + + out: + return result; +} + +/** + * cr_statement_ruleset_parse_from_buf: + * + *@a_buf: the buffer to parse. + *@a_enc: the character encoding of a_buf. + * + *Parses a buffer that contains a ruleset statement an instantiates + *a #CRStatement of type RULESET_STMT. + * + *Returns the newly built instance of #CRStatement in case of successful parsing, + *NULL otherwise. + */ +CRStatement * +cr_statement_ruleset_parse_from_buf (const guchar * a_buf, + enum CREncoding a_enc) +{ + enum CRStatus status = CR_OK; + CRStatement *result = NULL; + CRStatement **resultptr = NULL; + CRParser *parser = NULL; + CRDocHandler *sac_handler = NULL; + + g_return_val_if_fail (a_buf, NULL); + + parser = cr_parser_new_from_buf ((guchar*)a_buf, strlen ((const char *) a_buf), + a_enc, FALSE); + + g_return_val_if_fail (parser, NULL); + + sac_handler = cr_doc_handler_new (); + g_return_val_if_fail (parser, NULL); + + sac_handler->start_selector = parse_ruleset_start_selector_cb; + sac_handler->end_selector = parse_ruleset_end_selector_cb; + sac_handler->property = parse_ruleset_property_cb; + sac_handler->unrecoverable_error = + parse_ruleset_unrecoverable_error_cb; + + cr_parser_set_sac_handler (parser, sac_handler); + cr_parser_try_to_skip_spaces_and_comments (parser); + status = cr_parser_parse_ruleset (parser); + if (status != CR_OK) { + goto cleanup; + } + + resultptr = &result; + status = cr_doc_handler_get_result (sac_handler, + (gpointer *) resultptr); + if (!((status == CR_OK) && result)) { + if (result) { + cr_statement_destroy (result); + result = NULL; + } + } + + cleanup: + if (parser) { + cr_parser_destroy (parser); + parser = NULL; + sac_handler = NULL ; + } + if (sac_handler) { + cr_doc_handler_unref (sac_handler); + sac_handler = NULL; + } + return result; +} + +/** + * cr_statement_new_ruleset: + * + *@a_sel_list: the list of #CRSimpleSel (selectors) + *the rule applies to. + *@a_decl_list: the list of instances of #CRDeclaration + *that composes the ruleset. + *@a_media_types: a list of instances of GString that + *describe the media list this ruleset applies to. + * + *Creates a new instance of #CRStatement of type + *#CRRulSet. + * + *Returns the new instance of #CRStatement or NULL if something + *went wrong. + */ +CRStatement * +cr_statement_new_ruleset (CRStyleSheet * a_sheet, + CRSelector * a_sel_list, + CRDeclaration * a_decl_list, + CRStatement * a_parent_media_rule) +{ + CRStatement *result = NULL; + + g_return_val_if_fail (a_sel_list, NULL); + + if (a_parent_media_rule) { + g_return_val_if_fail + (a_parent_media_rule->type == AT_MEDIA_RULE_STMT, + NULL); + g_return_val_if_fail (a_parent_media_rule->kind.media_rule, + NULL); + } + + result = g_try_malloc (sizeof (CRStatement)); + + if (!result) { + cr_utils_trace_info ("Out of memory"); + return NULL; + } + + memset (result, 0, sizeof (CRStatement)); + result->type = RULESET_STMT; + result->kind.ruleset = g_try_malloc (sizeof (CRRuleSet)); + + if (!result->kind.ruleset) { + cr_utils_trace_info ("Out of memory"); + if (result) + g_free (result); + return NULL; + } + + memset (result->kind.ruleset, 0, sizeof (CRRuleSet)); + result->kind.ruleset->sel_list = a_sel_list; + if (a_sel_list) + cr_selector_ref (a_sel_list); + result->kind.ruleset->decl_list = a_decl_list; + + if (a_parent_media_rule) { + result->kind.ruleset->parent_media_rule = a_parent_media_rule; + a_parent_media_rule->kind.media_rule->rulesets = + cr_statement_append + (a_parent_media_rule->kind.media_rule->rulesets, + result); + } + + cr_statement_set_parent_sheet (result, a_sheet); + + return result; +} + +/** + * cr_statement_at_media_rule_parse_from_buf: + * + *@a_buf: the input to parse. + *@a_enc: the encoding of the buffer. + * + *Parses a buffer that contains an "\@media" declaration + *and builds an \@media css statement. + * + *Returns the \@media statement, or NULL if the buffer could not + *be successfully parsed. + */ +CRStatement * +cr_statement_at_media_rule_parse_from_buf (const guchar * a_buf, + enum CREncoding a_enc) +{ + CRParser *parser = NULL; + CRStatement *result = NULL; + CRStatement **resultptr = NULL; + CRDocHandler *sac_handler = NULL; + enum CRStatus status = CR_OK; + + parser = cr_parser_new_from_buf ((guchar*)a_buf, strlen ((const char *) a_buf), + a_enc, FALSE); + if (!parser) { + cr_utils_trace_info ("Instantiation of the parser failed"); + goto cleanup; + } + + sac_handler = cr_doc_handler_new (); + if (!sac_handler) { + cr_utils_trace_info + ("Instantiation of the sac handler failed"); + goto cleanup; + } + + sac_handler->start_media = parse_at_media_start_media_cb; + sac_handler->start_selector = parse_at_media_start_selector_cb; + sac_handler->property = parse_at_media_property_cb; + sac_handler->end_selector = parse_at_media_end_selector_cb; + sac_handler->end_media = parse_at_media_end_media_cb; + sac_handler->unrecoverable_error = + parse_at_media_unrecoverable_error_cb; + + status = cr_parser_set_sac_handler (parser, sac_handler); + if (status != CR_OK) + goto cleanup; + + status = cr_parser_try_to_skip_spaces_and_comments (parser); + if (status != CR_OK) + goto cleanup; + + status = cr_parser_parse_media (parser); + if (status != CR_OK) + goto cleanup; + + resultptr = &result; + status = cr_doc_handler_get_result (sac_handler, + (gpointer *) resultptr); + if (status != CR_OK) + goto cleanup; + + cleanup: + + if (parser) { + cr_parser_destroy (parser); + parser = NULL; + sac_handler = NULL ; + } + if (sac_handler) { + cr_doc_handler_unref (sac_handler); + sac_handler = NULL; + } + + return result; +} + +/** + * cr_statement_new_at_media_rule: + * + *@a_ruleset: the ruleset statements contained + *in the \@media rule. + *@a_media: the media string list. A list of GString pointers. + * + *Instantiates an instance of #CRStatement of type + *AT_MEDIA_RULE_STMT (\@media ruleset). + * + */ +CRStatement * +cr_statement_new_at_media_rule (CRStyleSheet * a_sheet, + CRStatement * a_rulesets, GList * a_media) +{ + CRStatement *result = NULL, + *cur = NULL; + + if (a_rulesets) + g_return_val_if_fail (a_rulesets->type == RULESET_STMT, NULL); + + result = g_try_malloc (sizeof (CRStatement)); + + if (!result) { + cr_utils_trace_info ("Out of memory"); + return NULL; + } + + memset (result, 0, sizeof (CRStatement)); + result->type = AT_MEDIA_RULE_STMT; + + result->kind.media_rule = g_try_malloc (sizeof (CRAtMediaRule)); + if (!result->kind.media_rule) { + cr_utils_trace_info ("Out of memory"); + g_free (result); + return NULL; + } + memset (result->kind.media_rule, 0, sizeof (CRAtMediaRule)); + result->kind.media_rule->rulesets = a_rulesets; + for (cur = a_rulesets; cur; cur = cur->next) { + if (cur->type != RULESET_STMT || !cur->kind.ruleset) { + cr_utils_trace_info ("Bad parameter a_rulesets. " + "It should be a list of " + "correct ruleset statement only !"); + goto error; + } + cur->kind.ruleset->parent_media_rule = result; + } + + result->kind.media_rule->media_list = a_media; + if (a_sheet) { + cr_statement_set_parent_sheet (result, a_sheet); + } + + return result; + + error: + return NULL; +} + +/** + * cr_statement_new_at_import_rule: + * + *@a_url: the url to connect to the get the file + *to be imported. + *@a_sheet: the imported parsed stylesheet. + * + *Creates a new instance of #CRStatment of type + *#CRAtImportRule. + * + *Returns the newly built instance of #CRStatement. + */ +CRStatement * +cr_statement_new_at_import_rule (CRStyleSheet * a_container_sheet, + CRString * a_url, + GList * a_media_list, + CRStyleSheet * a_imported_sheet) +{ + CRStatement *result = NULL; + + result = g_try_malloc (sizeof (CRStatement)); + + if (!result) { + cr_utils_trace_info ("Out of memory"); + return NULL; + } + + memset (result, 0, sizeof (CRStatement)); + result->type = AT_IMPORT_RULE_STMT; + + result->kind.import_rule = g_try_malloc (sizeof (CRAtImportRule)); + + if (!result->kind.import_rule) { + cr_utils_trace_info ("Out of memory"); + g_free (result); + return NULL; + } + + memset (result->kind.import_rule, 0, sizeof (CRAtImportRule)); + result->kind.import_rule->url = a_url; + result->kind.import_rule->media_list = a_media_list; + result->kind.import_rule->sheet = a_imported_sheet; + if (a_container_sheet) + cr_statement_set_parent_sheet (result, a_container_sheet); + + return result; +} + +/** + * cr_statement_at_import_rule_parse_from_buf: + * + *@a_buf: the buffer to parse. + *@a_encoding: the encoding of a_buf. + * + *Parses a buffer that contains an "\@import" rule and + *instantiate a #CRStatement of type AT_IMPORT_RULE_STMT + * + *Returns the newly built instance of #CRStatement in case of + *a successful parsing, NULL otherwise. + */ +CRStatement * +cr_statement_at_import_rule_parse_from_buf (const guchar * a_buf, + enum CREncoding a_encoding) +{ + enum CRStatus status = CR_OK; + CRParser *parser = NULL; + CRStatement *result = NULL; + GList *media_list = NULL; + CRString *import_string = NULL; + CRParsingLocation location = {0} ; + + parser = cr_parser_new_from_buf ((guchar*)a_buf, strlen ((const char *) a_buf), + a_encoding, FALSE); + if (!parser) { + cr_utils_trace_info ("Instantiation of parser failed."); + goto cleanup; + } + + status = cr_parser_try_to_skip_spaces_and_comments (parser); + if (status != CR_OK) + goto cleanup; + + status = cr_parser_parse_import (parser, + &media_list, + &import_string, + &location); + if (status != CR_OK || !import_string) + goto cleanup; + + result = cr_statement_new_at_import_rule (NULL, import_string, + media_list, NULL); + if (result) { + cr_parsing_location_copy (&result->location, + &location) ; + import_string = NULL; + media_list = NULL; + } + + cleanup: + if (parser) { + cr_parser_destroy (parser); + parser = NULL; + } + if (media_list) { + for (; media_list; + media_list = g_list_next (media_list)) { + if (media_list->data) { + cr_string_destroy ((CRString*)media_list->data); + media_list->data = NULL; + } + } + g_list_free (media_list); + media_list = NULL; + } + if (import_string) { + cr_string_destroy (import_string); + import_string = NULL; + } + + return result; +} + +/** + * cr_statement_new_at_page_rule: + * + *@a_decl_list: a list of instances of #CRDeclarations + *which is actually the list of declarations that applies to + *this page rule. + *@a_selector: the page rule selector. + * + *Creates a new instance of #CRStatement of type + *#CRAtPageRule. + * + *Returns the newly built instance of #CRStatement or NULL + *in case of error. + */ +CRStatement * +cr_statement_new_at_page_rule (CRStyleSheet * a_sheet, + CRDeclaration * a_decl_list, + CRString * a_name, CRString * a_pseudo) +{ + CRStatement *result = NULL; + + result = g_try_malloc (sizeof (CRStatement)); + + if (!result) { + cr_utils_trace_info ("Out of memory"); + return NULL; + } + + memset (result, 0, sizeof (CRStatement)); + result->type = AT_PAGE_RULE_STMT; + + result->kind.page_rule = g_try_malloc (sizeof (CRAtPageRule)); + + if (!result->kind.page_rule) { + cr_utils_trace_info ("Out of memory"); + g_free (result); + return NULL; + } + + memset (result->kind.page_rule, 0, sizeof (CRAtPageRule)); + if (a_decl_list) { + result->kind.page_rule->decl_list = a_decl_list; + cr_declaration_ref (a_decl_list); + } + result->kind.page_rule->name = a_name; + result->kind.page_rule->pseudo = a_pseudo; + if (a_sheet) + cr_statement_set_parent_sheet (result, a_sheet); + + return result; +} + +/** + * cr_statement_at_page_rule_parse_from_buf: + * + *@a_buf: the character buffer to parse. + *@a_encoding: the character encoding of a_buf. + * + *Parses a buffer that contains an "\@page" production and, + *if the parsing succeeds, builds the page statement. + * + *Returns the newly built at page statement in case of successful parsing, + *NULL otherwise. + */ +CRStatement * +cr_statement_at_page_rule_parse_from_buf (const guchar * a_buf, + enum CREncoding a_encoding) +{ + enum CRStatus status = CR_OK; + CRParser *parser = NULL; + CRDocHandler *sac_handler = NULL; + CRStatement *result = NULL; + CRStatement **resultptr = NULL; + + g_return_val_if_fail (a_buf, NULL); + + parser = cr_parser_new_from_buf ((guchar*)a_buf, strlen ((const char *) a_buf), + a_encoding, FALSE); + if (!parser) { + cr_utils_trace_info ("Instantiation of the parser failed."); + goto cleanup; + } + + sac_handler = cr_doc_handler_new (); + if (!sac_handler) { + cr_utils_trace_info + ("Instantiation of the sac handler failed."); + goto cleanup; + } + + sac_handler->start_page = parse_page_start_page_cb; + sac_handler->property = parse_page_property_cb; + sac_handler->end_page = parse_page_end_page_cb; + sac_handler->unrecoverable_error = parse_page_unrecoverable_error_cb; + + status = cr_parser_set_sac_handler (parser, sac_handler); + if (status != CR_OK) + goto cleanup; + + /*Now, invoke the parser to parse the "@page production" */ + cr_parser_try_to_skip_spaces_and_comments (parser); + if (status != CR_OK) + goto cleanup; + status = cr_parser_parse_page (parser); + if (status != CR_OK) + goto cleanup; + + resultptr = &result; + status = cr_doc_handler_get_result (sac_handler, + (gpointer *) resultptr); + + cleanup: + + if (parser) { + cr_parser_destroy (parser); + parser = NULL; + sac_handler = NULL ; + } + if (sac_handler) { + cr_doc_handler_unref (sac_handler); + sac_handler = NULL; + } + return result; +} + +/** + * cr_statement_new_at_charset_rule: + * + *@a_charset: the string representing the charset. + *Note that the newly built instance of #CRStatement becomes + *the owner of a_charset. The caller must not free a_charset !!!. + * + *Creates a new instance of #CRStatement of type + *#CRAtCharsetRule. + * + *Returns the newly built instance of #CRStatement or NULL + *if an error arises. + */ +CRStatement * +cr_statement_new_at_charset_rule (CRStyleSheet * a_sheet, + CRString * a_charset) +{ + CRStatement *result = NULL; + + g_return_val_if_fail (a_charset, NULL); + + result = g_try_malloc (sizeof (CRStatement)); + + if (!result) { + cr_utils_trace_info ("Out of memory"); + return NULL; + } + + memset (result, 0, sizeof (CRStatement)); + result->type = AT_CHARSET_RULE_STMT; + + result->kind.charset_rule = g_try_malloc (sizeof (CRAtCharsetRule)); + + if (!result->kind.charset_rule) { + cr_utils_trace_info ("Out of memory"); + g_free (result); + return NULL; + } + memset (result->kind.charset_rule, 0, sizeof (CRAtCharsetRule)); + result->kind.charset_rule->charset = a_charset; + cr_statement_set_parent_sheet (result, a_sheet); + + return result; +} + +/** + * cr_statement_at_charset_rule_parse_from_buf: + * + *@a_buf: the buffer to parse. + *@a_encoding: the character encoding of the buffer. + * + *Parses a buffer that contains an '\@charset' rule and + *creates an instance of #CRStatement of type AT_CHARSET_RULE_STMT. + * + *Returns the newly built instance of #CRStatement. + */ +CRStatement * +cr_statement_at_charset_rule_parse_from_buf (const guchar * a_buf, + enum CREncoding a_encoding) +{ + enum CRStatus status = CR_OK; + CRParser *parser = NULL; + CRStatement *result = NULL; + CRString *charset = NULL; + + g_return_val_if_fail (a_buf, NULL); + + parser = cr_parser_new_from_buf ((guchar*)a_buf, strlen ((const char *) a_buf), + a_encoding, FALSE); + if (!parser) { + cr_utils_trace_info ("Instantiation of the parser failed."); + goto cleanup; + } + + /*Now, invoke the parser to parse the "@charset production" */ + cr_parser_try_to_skip_spaces_and_comments (parser); + if (status != CR_OK) + goto cleanup; + status = cr_parser_parse_charset (parser, &charset, NULL); + if (status != CR_OK || !charset) + goto cleanup; + + result = cr_statement_new_at_charset_rule (NULL, charset); + if (result) + charset = NULL; + + cleanup: + + if (parser) { + cr_parser_destroy (parser); + parser = NULL; + } + if (charset) { + cr_string_destroy (charset); + } + + return result; +} + +/** + * cr_statement_new_at_font_face_rule: + * + *@a_font_decls: a list of instances of #CRDeclaration. Each declaration + *is actually a font declaration. + * + *Creates an instance of #CRStatement of type #CRAtFontFaceRule. + * + *Returns the newly built instance of #CRStatement. + */ +CRStatement * +cr_statement_new_at_font_face_rule (CRStyleSheet * a_sheet, + CRDeclaration * a_font_decls) +{ + CRStatement *result = NULL; + + result = g_try_malloc (sizeof (CRStatement)); + + if (!result) { + cr_utils_trace_info ("Out of memory"); + return NULL; + } + memset (result, 0, sizeof (CRStatement)); + result->type = AT_FONT_FACE_RULE_STMT; + + result->kind.font_face_rule = g_try_malloc + (sizeof (CRAtFontFaceRule)); + + if (!result->kind.font_face_rule) { + cr_utils_trace_info ("Out of memory"); + g_free (result); + return NULL; + } + memset (result->kind.font_face_rule, 0, sizeof (CRAtFontFaceRule)); + + result->kind.font_face_rule->decl_list = a_font_decls; + if (a_sheet) + cr_statement_set_parent_sheet (result, a_sheet); + + return result; +} + +/** + * cr_statement_font_face_rule_parse_from_buf: + * + * + *@a_buf: the buffer to parse. + *@a_encoding: the character encoding of a_buf. + * + *Parses a buffer that contains an "\@font-face" rule and builds + *an instance of #CRStatement of type AT_FONT_FACE_RULE_STMT out of it. + * + *Returns the newly built instance of #CRStatement in case of successufull + *parsing, NULL otherwise. + */ +CRStatement * +cr_statement_font_face_rule_parse_from_buf (const guchar * a_buf, + enum CREncoding a_encoding) +{ + CRStatement *result = NULL; + CRStatement **resultptr = NULL; + CRParser *parser = NULL; + CRDocHandler *sac_handler = NULL; + enum CRStatus status = CR_OK; + + parser = cr_parser_new_from_buf ((guchar*)a_buf, strlen ((const char *) a_buf), + a_encoding, FALSE); + if (!parser) + goto cleanup; + + sac_handler = cr_doc_handler_new (); + if (!sac_handler) + goto cleanup; + + /* + *set sac callbacks here + */ + sac_handler->start_font_face = parse_font_face_start_font_face_cb; + sac_handler->property = parse_font_face_property_cb; + sac_handler->end_font_face = parse_font_face_end_font_face_cb; + sac_handler->unrecoverable_error = + parse_font_face_unrecoverable_error_cb; + + status = cr_parser_set_sac_handler (parser, sac_handler); + if (status != CR_OK) + goto cleanup; + + /* + *cleanup spaces of comment that may be there before the real + *"@font-face" thing. + */ + status = cr_parser_try_to_skip_spaces_and_comments (parser); + if (status != CR_OK) + goto cleanup; + + status = cr_parser_parse_font_face (parser); + if (status != CR_OK) + goto cleanup; + + resultptr = &result; + status = cr_doc_handler_get_result (sac_handler, + (gpointer *) resultptr); + if (status != CR_OK || !result) + goto cleanup; + + cleanup: + if (parser) { + cr_parser_destroy (parser); + parser = NULL; + sac_handler = NULL ; + } + if (sac_handler) { + cr_doc_handler_unref (sac_handler); + sac_handler = NULL; + } + return result; +} + +/** + * cr_statement_set_parent_sheet: + * + *@a_this: the current instance of #CRStatement. + *@a_sheet: the sheet that contains the current statement. + * + *Sets the container stylesheet. + * + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_set_parent_sheet (CRStatement * a_this, CRStyleSheet * a_sheet) +{ + g_return_val_if_fail (a_this, CR_BAD_PARAM_ERROR); + a_this->parent_sheet = a_sheet; + return CR_OK; +} + +/** + * cr_statement_get_parent_sheet: + * + *@a_this: the current #CRStatement. + *@a_sheet: out parameter. A pointer to the sheets that + * + *Gets the sheets that contains the current statement. + * + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_get_parent_sheet (CRStatement * a_this, CRStyleSheet ** a_sheet) +{ + g_return_val_if_fail (a_this && a_sheet, CR_BAD_PARAM_ERROR); + *a_sheet = a_this->parent_sheet; + return CR_OK; +} + +/** + * cr_statement_append: + * + *@a_this: the current instance of the statement list. + *@a_new: a_new the new instance of #CRStatement to append. + * + *Appends a new statement to the statement list. + * + *Returns the new list statement list, or NULL in cas of failure. + */ +CRStatement * +cr_statement_append (CRStatement * a_this, CRStatement * a_new) +{ + CRStatement *cur = NULL; + + g_return_val_if_fail (a_new, NULL); + + if (!a_this) { + return a_new; + } + + /*walk forward in the current list to find the tail list element */ + for (cur = a_this; cur && cur->next; cur = cur->next) ; + + cur->next = a_new; + a_new->prev = cur; + + return a_this; +} + +/** + * cr_statement_prepend: + * + *@a_this: the current instance of #CRStatement. + *@a_new: the new statement to prepend. + * + *Prepends the an instance of #CRStatement to + *the current statement list. + * + *Returns the new list with the new statement prepended, + *or NULL in case of an error. + */ +CRStatement * +cr_statement_prepend (CRStatement * a_this, CRStatement * a_new) +{ + CRStatement *cur = NULL; + + g_return_val_if_fail (a_new, NULL); + + if (!a_this) + return a_new; + + a_new->next = a_this; + a_this->prev = a_new; + + /*walk backward in the prepended list to find the head list element */ + for (cur = a_new; cur && cur->prev; cur = cur->prev) ; + + return cur; +} + +/** + * cr_statement_unlink: + * + *@a_this: the current statements list. + *@a_to_unlink: the statement to unlink from the list. + * + *Unlinks a statement from the statements list. + * + *Returns the new list where a_to_unlink has been unlinked + *from, or NULL in case of error. + */ +CRStatement * +cr_statement_unlink (CRStatement * a_stmt) +{ + CRStatement *result = a_stmt; + + g_return_val_if_fail (result, NULL); + + /** + *Some sanity checks first + */ + if (a_stmt->next) { + g_return_val_if_fail (a_stmt->next->prev == a_stmt, NULL); + } + if (a_stmt->prev) { + g_return_val_if_fail (a_stmt->prev->next == a_stmt, NULL); + } + + /** + *Now, the real unlinking job. + */ + if (a_stmt->next) { + a_stmt->next->prev = a_stmt->prev; + } + if (a_stmt->prev) { + a_stmt->prev->next = a_stmt->next; + } + + if (a_stmt->parent_sheet + && a_stmt->parent_sheet->statements == a_stmt) { + a_stmt->parent_sheet->statements = + a_stmt->parent_sheet->statements->next; + } + + a_stmt->next = NULL; + a_stmt->prev = NULL; + a_stmt->parent_sheet = NULL; + + return result; +} + +/** + * cr_statement_nr_rules: + * + *@a_this: the current instance of #CRStatement. + * + *Gets the number of rules in the statement list; + * + *Returns number of rules in the statement list. + */ +gint +cr_statement_nr_rules (CRStatement const * a_this) +{ + CRStatement const *cur = NULL; + int nr = 0; + + g_return_val_if_fail (a_this, -1); + + for (cur = a_this; cur; cur = cur->next) + nr++; + return nr; +} + +/** + * cr_statement_get_from_list: + * + *@a_this: the current instance of #CRStatement. + *@itemnr: the index into the statement list. + * + *Use an index to get a CRStatement from the statement list. + * + *Returns CRStatement at position itemnr, if itemnr > number of statements - 1, + *it will return NULL. + */ +CRStatement * +cr_statement_get_from_list (CRStatement * a_this, int itemnr) +{ + CRStatement *cur = NULL; + int nr = 0; + + g_return_val_if_fail (a_this, NULL); + + for (cur = a_this; cur; cur = cur->next) + if (nr++ == itemnr) + return cur; + return NULL; +} + +/** + * cr_statement_ruleset_set_sel_list: + * + *@a_this: the current ruleset statement. + *@a_sel_list: the selector list to set. Note + *that this function increments the ref count of a_sel_list. + *The sel list will be destroyed at the destruction of the + *current instance of #CRStatement. + * + *Sets a selector list to a ruleset statement. + * + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_ruleset_set_sel_list (CRStatement * a_this, + CRSelector * a_sel_list) +{ + g_return_val_if_fail (a_this && a_this->type == RULESET_STMT, + CR_BAD_PARAM_ERROR); + + if (a_this->kind.ruleset->sel_list) + cr_selector_unref (a_this->kind.ruleset->sel_list); + + a_this->kind.ruleset->sel_list = a_sel_list; + + if (a_sel_list) + cr_selector_ref (a_sel_list); + + return CR_OK; +} + +/** + * cr_statement_ruleset_get_declarations: + * + *@a_this: the current instance of #CRStatement. + *@a_decl_list: out parameter. A pointer to the the returned + *list of declaration. Must not be NULL. + * + *Gets a pointer to the list of declaration contained + *in the ruleset statement. + * + *Returns CR_OK upon successful completion, an error code if something + *bad happened. + */ +enum CRStatus +cr_statement_ruleset_get_declarations (CRStatement * a_this, + CRDeclaration ** a_decl_list) +{ + g_return_val_if_fail (a_this + && a_this->type == RULESET_STMT + && a_this->kind.ruleset + && a_decl_list, CR_BAD_PARAM_ERROR); + + *a_decl_list = a_this->kind.ruleset->decl_list; + + return CR_OK; +} + +/** + * cr_statement_ruleset_get_sel_list: + * + *@a_this: the current ruleset statement. + *@a_list: out parameter. The returned selector list, + *if and only if the function returned CR_OK. + * + *Gets a pointer to the selector list contained in + *the current ruleset statement. + * + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_ruleset_get_sel_list (CRStatement const * a_this, CRSelector ** a_list) +{ + g_return_val_if_fail (a_this && a_this->type == RULESET_STMT + && a_this->kind.ruleset, CR_BAD_PARAM_ERROR); + + *a_list = a_this->kind.ruleset->sel_list; + + return CR_OK; +} + +/** + * cr_statement_ruleset_set_decl_list: + * + *@a_this: the current ruleset statement. + *@a_list: the declaration list to be added to the current + *ruleset statement. + * + *Sets a declaration list to the current ruleset statement. + * + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_ruleset_set_decl_list (CRStatement * a_this, + CRDeclaration * a_list) +{ + g_return_val_if_fail (a_this && a_this->type == RULESET_STMT + && a_this->kind.ruleset, CR_BAD_PARAM_ERROR); + + if (a_this->kind.ruleset->decl_list == a_list) + return CR_OK; + + if (a_this->kind.ruleset->sel_list) { + cr_declaration_destroy (a_this->kind.ruleset->decl_list); + } + + a_this->kind.ruleset->sel_list = NULL; + + return CR_OK; +} + +/** + * cr_statement_ruleset_append_decl2: + * + *@a_this: the current statement. + *@a_prop: the property of the declaration. + *@a_value: the value of the declaration. + * + *Appends a declaration to the current ruleset statement. + * + *Returns CR_OK upon successful completion, an error code + *otherwise. + */ +enum CRStatus +cr_statement_ruleset_append_decl2 (CRStatement * a_this, + CRString * a_prop, + CRTerm * a_value) +{ + CRDeclaration *new_decls = NULL; + + g_return_val_if_fail (a_this && a_this->type == RULESET_STMT + && a_this->kind.ruleset, CR_BAD_PARAM_ERROR); + + new_decls = cr_declaration_append2 + (a_this->kind.ruleset->decl_list, + a_prop, a_value); + g_return_val_if_fail (new_decls, CR_ERROR); + a_this->kind.ruleset->decl_list = new_decls; + + return CR_OK; +} + +/** + * cr_statement_ruleset_append_decl: + * + *Appends a declaration to the current statement. + * + *@a_this: the current statement. + *@a_declaration: the declaration to append. + * + *Returns CR_OK upon successful completion, an error code + *otherwise. + */ +enum CRStatus +cr_statement_ruleset_append_decl (CRStatement * a_this, + CRDeclaration * a_decl) +{ + CRDeclaration *new_decls = NULL; + + g_return_val_if_fail (a_this && a_this->type == RULESET_STMT + && a_this->kind.ruleset, CR_BAD_PARAM_ERROR); + + new_decls = cr_declaration_append + (a_this->kind.ruleset->decl_list, a_decl); + g_return_val_if_fail (new_decls, CR_ERROR); + a_this->kind.ruleset->decl_list = new_decls; + + return CR_OK; +} + +/** + * cr_statement_at_import_rule_set_imported_sheet: + * + *Sets a stylesheet to the current \@import rule. + *@a_this: the current \@import rule. + *@a_sheet: the stylesheet. The stylesheet is owned + *by the current instance of #CRStatement, that is, the + *stylesheet will be destroyed when the current instance + *of #CRStatement is destroyed. + * + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_at_import_rule_set_imported_sheet (CRStatement * a_this, + CRStyleSheet * a_sheet) +{ + g_return_val_if_fail (a_this + && a_this->type == AT_IMPORT_RULE_STMT + && a_this->kind.import_rule, + CR_BAD_PARAM_ERROR); + + a_this->kind.import_rule->sheet = a_sheet; + + return CR_OK; +} + +/** + * cr_statement_at_import_rule_get_imported_sheet: + * + *@a_this: the current \@import rule statement. + *@a_sheet: out parameter. The returned stylesheet if and + *only if the function returns CR_OK. + * + *Gets the stylesheet contained by the \@import rule statement. + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_at_import_rule_get_imported_sheet (CRStatement * a_this, + CRStyleSheet ** a_sheet) +{ + g_return_val_if_fail (a_this + && a_this->type == AT_IMPORT_RULE_STMT + && a_this->kind.import_rule, + CR_BAD_PARAM_ERROR); + *a_sheet = a_this->kind.import_rule->sheet; + return CR_OK; + +} + +/** + * cr_statement_at_import_rule_set_url: + * + *@a_this: the current \@import rule statement. + *@a_url: the url to set. + * + *Sets an url to the current \@import rule statement. + * + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_at_import_rule_set_url (CRStatement * a_this, + CRString * a_url) +{ + g_return_val_if_fail (a_this + && a_this->type == AT_IMPORT_RULE_STMT + && a_this->kind.import_rule, + CR_BAD_PARAM_ERROR); + + if (a_this->kind.import_rule->url) { + cr_string_destroy (a_this->kind.import_rule->url); + } + + a_this->kind.import_rule->url = a_url; + + return CR_OK; +} + +/** + * cr_statement_at_import_rule_get_url: + * + *@a_this: the current \@import rule statement. + *@a_url: out parameter. The returned url if + *and only if the function returned CR_OK. + * + *Gets the url of the \@import rule statement. + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_at_import_rule_get_url (CRStatement const * a_this, + CRString ** a_url) +{ + g_return_val_if_fail (a_this + && a_this->type == AT_IMPORT_RULE_STMT + && a_this->kind.import_rule, + CR_BAD_PARAM_ERROR); + + *a_url = a_this->kind.import_rule->url; + + return CR_OK; +} + +/** + * cr_statement_at_media_nr_rules: + * + *@a_this: the current instance of #CRStatement. + * + *Returns the number of rules in the media rule; + */ +int +cr_statement_at_media_nr_rules (CRStatement const * a_this) +{ + g_return_val_if_fail (a_this + && a_this->type == AT_MEDIA_RULE_STMT + && a_this->kind.media_rule, CR_BAD_PARAM_ERROR); + + return cr_statement_nr_rules (a_this->kind.media_rule->rulesets); +} + +/** + * cr_statement_at_media_get_from_list: + * + *@a_this: the current instance of #CRStatement. + *@itemnr: the index into the media rule list of rules. + * + *Use an index to get a CRStatement from the media rule list of rules. + * + *Returns CRStatement at position itemnr, if itemnr > number of rules - 1, + *it will return NULL. + */ +CRStatement * +cr_statement_at_media_get_from_list (CRStatement * a_this, int itemnr) +{ + g_return_val_if_fail (a_this + && a_this->type == AT_MEDIA_RULE_STMT + && a_this->kind.media_rule, NULL); + + return cr_statement_get_from_list (a_this->kind.media_rule->rulesets, + itemnr); +} + +/** + * cr_statement_at_page_rule_set_declarations: + * + *@a_this: the current \@page rule statement. + *@a_decl_list: the declaration list to add. Will be freed + *by the current instance of #CRStatement when it is destroyed. + * + *Sets a declaration list to the current \@page rule statement. + * + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_at_page_rule_set_declarations (CRStatement * a_this, + CRDeclaration * a_decl_list) +{ + g_return_val_if_fail (a_this + && a_this->type == AT_PAGE_RULE_STMT + && a_this->kind.page_rule, CR_BAD_PARAM_ERROR); + + if (a_this->kind.page_rule->decl_list) { + cr_declaration_unref (a_this->kind.page_rule->decl_list); + } + + a_this->kind.page_rule->decl_list = a_decl_list; + + if (a_decl_list) { + cr_declaration_ref (a_decl_list); + } + + return CR_OK; +} + +/** + * cr_statement_at_page_rule_get_declarations: + * + *@a_this: the current \@page rule statement. + *@a_decl_list: out parameter. The returned declaration list. + * + *Gets the declaration list associated to the current \@page rule + *statement. + * + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_at_page_rule_get_declarations (CRStatement * a_this, + CRDeclaration ** a_decl_list) +{ + g_return_val_if_fail (a_this + && a_this->type == AT_PAGE_RULE_STMT + && a_this->kind.page_rule, CR_BAD_PARAM_ERROR); + + *a_decl_list = a_this->kind.page_rule->decl_list; + + return CR_OK; +} + +/** + * cr_statement_at_charset_rule_set_charset: + * + * + *@a_this: the current \@charset rule statement. + *@a_charset: the charset to set. + * + *Sets the charset of the current \@charset rule statement. + * + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_at_charset_rule_set_charset (CRStatement * a_this, + CRString * a_charset) +{ + g_return_val_if_fail (a_this + && a_this->type == AT_CHARSET_RULE_STMT + && a_this->kind.charset_rule, + CR_BAD_PARAM_ERROR); + + if (a_this->kind.charset_rule->charset) { + cr_string_destroy (a_this->kind.charset_rule->charset); + } + a_this->kind.charset_rule->charset = a_charset; + return CR_OK; +} + +/** + * cr_statement_at_charset_rule_get_charset: + *@a_this: the current \@charset rule statement. + *@a_charset: out parameter. The returned charset string if + *and only if the function returned CR_OK. + * + *Gets the charset string associated to the current + *\@charset rule statement. + * + * Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_at_charset_rule_get_charset (CRStatement const * a_this, + CRString ** a_charset) +{ + g_return_val_if_fail (a_this + && a_this->type == AT_CHARSET_RULE_STMT + && a_this->kind.charset_rule, + CR_BAD_PARAM_ERROR); + + *a_charset = a_this->kind.charset_rule->charset; + + return CR_OK; +} + +/** + * cr_statement_at_font_face_rule_set_decls: + * + *@a_this: the current \@font-face rule statement. + *@a_decls: the declarations list to set. + * + *Sets a declaration list to the current \@font-face rule statement. + * + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_at_font_face_rule_set_decls (CRStatement * a_this, + CRDeclaration * a_decls) +{ + g_return_val_if_fail (a_this + && a_this->type == AT_FONT_FACE_RULE_STMT + && a_this->kind.font_face_rule, + CR_BAD_PARAM_ERROR); + + if (a_this->kind.font_face_rule->decl_list) { + cr_declaration_unref (a_this->kind.font_face_rule->decl_list); + } + + a_this->kind.font_face_rule->decl_list = a_decls; + cr_declaration_ref (a_decls); + + return CR_OK; +} + +/** + * cr_statement_at_font_face_rule_get_decls: + * + *@a_this: the current \@font-face rule statement. + *@a_decls: out parameter. The returned declaration list if + *and only if this function returns CR_OK. + * + *Gets the declaration list associated to the current instance + *of \@font-face rule statement. + * + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_at_font_face_rule_get_decls (CRStatement * a_this, + CRDeclaration ** a_decls) +{ + g_return_val_if_fail (a_this + && a_this->type == AT_FONT_FACE_RULE_STMT + && a_this->kind.font_face_rule, + CR_BAD_PARAM_ERROR); + + *a_decls = a_this->kind.font_face_rule->decl_list; + + return CR_OK; +} + +/** + * cr_statement_at_font_face_rule_add_decl: + * + *@a_this: the current \@font-face rule statement. + *@a_prop: the property of the declaration. + *@a_value: the value of the declaration. + * + *Adds a declaration to the current \@font-face rule + *statement. + * + *Returns CR_OK upon successful completion, an error code otherwise. + */ +enum CRStatus +cr_statement_at_font_face_rule_add_decl (CRStatement * a_this, + CRString * a_prop, CRTerm * a_value) +{ + CRDeclaration *decls = NULL; + + g_return_val_if_fail (a_this + && a_this->type == AT_FONT_FACE_RULE_STMT + && a_this->kind.font_face_rule, + CR_BAD_PARAM_ERROR); + + decls = cr_declaration_append2 + (a_this->kind.font_face_rule->decl_list, + a_prop, a_value); + + g_return_val_if_fail (decls, CR_ERROR); + + if (a_this->kind.font_face_rule->decl_list == NULL) + cr_declaration_ref (decls); + + a_this->kind.font_face_rule->decl_list = decls; + + return CR_OK; +} + + +/** + * cr_statement_to_string: + * + *@a_this: the current statement to serialize + *@a_indent: the number of white space of indentation. + * + *Serializes a css statement into a string + * + *Returns the serialized statement. Must be freed by the caller + *using g_free(). + */ +gchar * +cr_statement_to_string (CRStatement const * a_this, gulong a_indent) +{ + gchar *str = NULL ; + + if (!a_this) + return NULL; + + switch (a_this->type) { + case RULESET_STMT: + str = cr_statement_ruleset_to_string + (a_this, a_indent); + break; + + case AT_FONT_FACE_RULE_STMT: + str = cr_statement_font_face_rule_to_string + (a_this, a_indent) ; + break; + + case AT_CHARSET_RULE_STMT: + str = cr_statement_charset_to_string + (a_this, a_indent); + break; + + case AT_PAGE_RULE_STMT: + str = cr_statement_at_page_rule_to_string + (a_this, a_indent); + break; + + case AT_MEDIA_RULE_STMT: + str = cr_statement_media_rule_to_string + (a_this, a_indent); + break; + + case AT_IMPORT_RULE_STMT: + str = cr_statement_import_rule_to_string + (a_this, a_indent); + break; + + default: + cr_utils_trace_info ("Statement unrecognized"); + break; + } + return str ; +} + +gchar* +cr_statement_list_to_string (CRStatement const *a_this, gulong a_indent) +{ + CRStatement const *cur_stmt = NULL ; + GString *stringue = NULL ; + gchar *str = NULL ; + + g_return_val_if_fail (a_this, NULL) ; + + stringue = g_string_new (NULL) ; + if (!stringue) { + cr_utils_trace_info ("Out of memory") ; + return NULL ; + } + for (cur_stmt = a_this ; cur_stmt; + cur_stmt = cur_stmt->next) { + str = cr_statement_to_string (cur_stmt, a_indent) ; + if (str) { + if (!cur_stmt->prev) { + g_string_append (stringue, str) ; + } else { + g_string_append_printf + (stringue, "\n%s", str) ; + } + g_free (str) ; + str = NULL ; + } + } + str = g_string_free (stringue, FALSE) ; + return str ; +} + +/** + * cr_statement_dump: + * + *@a_this: the current css2 statement. + *@a_fp: the destination file pointer. + *@a_indent: the number of white space indentation characters. + * + *Dumps the css2 statement to a file. + */ +void +cr_statement_dump (CRStatement const * a_this, FILE * a_fp, gulong a_indent) +{ + gchar *str = NULL ; + + if (!a_this) + return; + + str = cr_statement_to_string (a_this, a_indent) ; + if (str) { + fprintf (a_fp, "%s",str) ; + g_free (str) ; + str = NULL ; + } +} + +/** + * cr_statement_dump_ruleset: + * + *@a_this: the current instance of #CRStatement. + *@a_fp: the destination file pointer. + *@a_indent: the number of indentation white spaces to add. + * + *Dumps a ruleset statement to a file. + */ +void +cr_statement_dump_ruleset (CRStatement const * a_this, FILE * a_fp, glong a_indent) +{ + gchar *str = NULL; + + g_return_if_fail (a_fp && a_this); + str = cr_statement_ruleset_to_string (a_this, a_indent); + if (str) { + fprintf (a_fp, "%s", str); + g_free (str); + str = NULL; + } +} + +/** + * cr_statement_dump_font_face_rule: + * + *@a_this: the current instance of font face rule statement. + *@a_fp: the destination file pointer. + *@a_indent: the number of white space indentation. + * + *Dumps a font face rule statement to a file. + */ +void +cr_statement_dump_font_face_rule (CRStatement const * a_this, FILE * a_fp, + glong a_indent) +{ + gchar *str = NULL ; + g_return_if_fail (a_this + && a_this->type == AT_FONT_FACE_RULE_STMT); + + str = cr_statement_font_face_rule_to_string (a_this, + a_indent) ; + if (str) { + fprintf (a_fp, "%s", str) ; + g_free (str) ; + str = NULL ; + } +} + +/** + * cr_statement_dump_charset: + * + *@a_this: the current instance of the \@charset rule statement. + *@a_fp: the destination file pointer. + *@a_indent: the number of indentation white spaces. + * + *Dumps an \@charset rule statement to a file. + */ +void +cr_statement_dump_charset (CRStatement const * a_this, FILE * a_fp, gulong a_indent) +{ + gchar *str = NULL; + + g_return_if_fail (a_this && a_this->type == AT_CHARSET_RULE_STMT); + + str = cr_statement_charset_to_string (a_this, + a_indent) ; + if (str) { + fprintf (a_fp, "%s", str) ; + g_free (str) ; + str = NULL ; + } +} + + +/** + * cr_statement_dump_page: + * + *@a_this: the statement to dump on stdout. + *@a_fp: the destination file pointer. + *@a_indent: the number of indentation white spaces. + * + *Dumps an \@page rule statement on stdout. + */ +void +cr_statement_dump_page (CRStatement const * a_this, FILE * a_fp, gulong a_indent) +{ + gchar *str = NULL; + + g_return_if_fail (a_this + && a_this->type == AT_PAGE_RULE_STMT + && a_this->kind.page_rule); + + str = cr_statement_at_page_rule_to_string (a_this, a_indent) ; + if (str) { + fprintf (a_fp, "%s", str); + g_free (str) ; + str = NULL ; + } +} + + +/** + * cr_statement_dump_media_rule: + * + *@a_this: the statement to dump. + *@a_fp: the destination file pointer + *@a_indent: the number of white spaces indentation. + * + *Dumps an \@media rule statement to a file. + */ +void +cr_statement_dump_media_rule (CRStatement const * a_this, + FILE * a_fp, + gulong a_indent) +{ + gchar *str = NULL ; + g_return_if_fail (a_this->type == AT_MEDIA_RULE_STMT); + + str = cr_statement_media_rule_to_string (a_this, a_indent) ; + if (str) { + fprintf (a_fp, "%s", str) ; + g_free (str) ; + str = NULL ; + } +} + +/** + * cr_statement_dump_import_rule: + * + *@a_fp: the destination file pointer. + *@a_indent: the number of white space indentations. + * + *Dumps an \@import rule statement to a file. + */ +void +cr_statement_dump_import_rule (CRStatement const * a_this, FILE * a_fp, + gulong a_indent) +{ + gchar *str = NULL ; + g_return_if_fail (a_this + && a_this->type == AT_IMPORT_RULE_STMT + && a_fp + && a_this->kind.import_rule); + + str = cr_statement_import_rule_to_string (a_this, a_indent) ; + if (str) { + fprintf (a_fp, "%s", str) ; + g_free (str) ; + str = NULL ; + } +} + +/** + * cr_statement_destroy: + * + * @a_this: the current instance of #CRStatement. + * + *Destructor of #CRStatement. + */ +void +cr_statement_destroy (CRStatement * a_this) +{ + CRStatement *cur = NULL; + + g_return_if_fail (a_this); + + /*go get the tail of the list */ + for (cur = a_this; cur && cur->next; cur = cur->next) { + cr_statement_clear (cur); + } + + if (cur) + cr_statement_clear (cur); + + if (cur->prev == NULL) { + g_free (a_this); + return; + } + + /*walk backward and free next element */ + for (cur = cur->prev; cur && cur->prev; cur = cur->prev) { + if (cur->next) { + g_free (cur->next); + cur->next = NULL; + } + } + + if (!cur) + return; + + /*free the one remaining list */ + if (cur->next) { + g_free (cur->next); + cur->next = NULL; + } + + g_free (cur); + cur = NULL; +} |