netsurf/render/html.c

2315 lines
54 KiB
C

/*
* Copyright 2007 James Bursa <bursa@users.sourceforge.net>
* Copyright 2010 Michael Drake <tlsa@netsurf-browser.org>
*
* This file is part of NetSurf, http://www.netsurf-browser.org/
*
* NetSurf is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; version 2 of the License.
*
* NetSurf 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 General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
/** \file
* Content for text/html (implementation).
*/
#include <assert.h>
#include <ctype.h>
#include <stdint.h>
#include <string.h>
#include <strings.h>
#include <stdlib.h>
#include "utils/config.h"
#include "utils/corestrings.h"
#include "utils/http.h"
#include "utils/libdom.h"
#include "utils/log.h"
#include "utils/messages.h"
#include "utils/talloc.h"
#include "utils/utf8.h"
#include "utils/nsoption.h"
#include "content/content_protected.h"
#include "content/fetch.h"
#include "content/hlcache.h"
#include "desktop/selection.h"
#include "desktop/scrollbar.h"
#include "desktop/textarea.h"
#include "image/bitmap.h"
#include "javascript/js.h"
#include "desktop/browser.h"
#include "desktop/font.h"
#include "desktop/gui_utf8.h"
#include "desktop/gui_internal.h"
#include "render/box.h"
#include "render/form_internal.h"
#include "render/html_internal.h"
#include "render/imagemap.h"
#include "render/layout.h"
#include "render/search.h"
#define CHUNK 4096
/* Change these to 1 to cause a dump to stderr of the frameset or box
* when the trees have been built.
*/
#define ALWAYS_DUMP_FRAMESET 0
#define ALWAYS_DUMP_BOX 0
static const char *html_types[] = {
"application/xhtml+xml",
"text/html"
};
/**
* Perform post-box-creation conversion of a document
*
* \param c HTML content to complete conversion of
* \param success Whether box tree construction was successful
*/
static void html_box_convert_done(html_content *c, bool success)
{
nserror err;
dom_exception exc; /* returned by libdom functions */
dom_node *html;
LOG(("Done XML to box (%p)", c));
/* Clean up and report error if unsuccessful or aborted */
if ((success == false) || (c->aborted)) {
html_object_free_objects(c);
if (success == false) {
content_broadcast_errorcode(&c->base, NSERROR_BOX_CONVERT);
} else {
content_broadcast_errorcode(&c->base, NSERROR_STOPPED);
}
content_set_error(&c->base);
return;
}
#if ALWAYS_DUMP_BOX
box_dump(stderr, c->layout->children, 0, true);
#endif
#if ALWAYS_DUMP_FRAMESET
if (c->frameset)
html_dump_frameset(c->frameset, 0);
#endif
exc = dom_document_get_document_element(c->document, (void *) &html);
if ((exc != DOM_NO_ERR) || (html == NULL)) {
/** @todo should this call html_object_free_objects(c);
* like the other error paths
*/
LOG(("error retrieving html element from dom"));
content_broadcast_errorcode(&c->base, NSERROR_DOM);
content_set_error(&c->base);
return;
}
/* extract image maps - can't do this sensibly in dom_to_box */
err = imagemap_extract(c);
if (err != NSERROR_OK) {
LOG(("imagemap extraction failed"));
html_object_free_objects(c);
content_broadcast_errorcode(&c->base, err);
content_set_error(&c->base);
dom_node_unref(html);
return;
}
/*imagemap_dump(c);*/
/* Destroy the parser binding */
dom_hubbub_parser_destroy(c->parser);
c->parser = NULL;
content_set_ready(&c->base);
if (c->base.active == 0) {
content_set_done(&c->base);
}
html_set_status(c, "");
dom_node_unref(html);
}
/** process link node */
static bool html_process_link(html_content *c, dom_node *node)
{
struct content_rfc5988_link link; /* the link added to the content */
dom_exception exc; /* returned by libdom functions */
dom_string *atr_string;
nserror error;
memset(&link, 0, sizeof(struct content_rfc5988_link));
/* check that the relation exists - w3c spec says must be present */
exc = dom_element_get_attribute(node, corestring_dom_rel, &atr_string);
if ((exc != DOM_NO_ERR) || (atr_string == NULL)) {
return false;
}
/* get a lwc string containing the link relation */
exc = dom_string_intern(atr_string, &link.rel);
dom_string_unref(atr_string);
if (exc != DOM_NO_ERR) {
return false;
}
/* check that the href exists - w3c spec says must be present */
exc = dom_element_get_attribute(node, corestring_dom_href, &atr_string);
if ((exc != DOM_NO_ERR) || (atr_string == NULL)) {
lwc_string_unref(link.rel);
return false;
}
/* get nsurl */
error = nsurl_join(c->base_url, dom_string_data(atr_string),
&link.href);
dom_string_unref(atr_string);
if (error != NSERROR_OK) {
lwc_string_unref(link.rel);
return false;
}
/* look for optional properties -- we don't care if internment fails */
exc = dom_element_get_attribute(node,
corestring_dom_hreflang, &atr_string);
if ((exc == DOM_NO_ERR) && (atr_string != NULL)) {
/* get a lwc string containing the href lang */
exc = dom_string_intern(atr_string, &link.hreflang);
dom_string_unref(atr_string);
}
exc = dom_element_get_attribute(node,
corestring_dom_type, &atr_string);
if ((exc == DOM_NO_ERR) && (atr_string != NULL)) {
/* get a lwc string containing the type */
exc = dom_string_intern(atr_string, &link.type);
dom_string_unref(atr_string);
}
exc = dom_element_get_attribute(node,
corestring_dom_media, &atr_string);
if ((exc == DOM_NO_ERR) && (atr_string != NULL)) {
/* get a lwc string containing the media */
exc = dom_string_intern(atr_string, &link.media);
dom_string_unref(atr_string);
}
exc = dom_element_get_attribute(node,
corestring_dom_sizes, &atr_string);
if ((exc == DOM_NO_ERR) && (atr_string != NULL)) {
/* get a lwc string containing the sizes */
exc = dom_string_intern(atr_string, &link.sizes);
dom_string_unref(atr_string);
}
/* add to content */
content__add_rfc5988_link(&c->base, &link);
if (link.sizes != NULL)
lwc_string_unref(link.sizes);
if (link.media != NULL)
lwc_string_unref(link.media);
if (link.type != NULL)
lwc_string_unref(link.type);
if (link.hreflang != NULL)
lwc_string_unref(link.hreflang);
nsurl_unref(link.href);
lwc_string_unref(link.rel);
return true;
}
/** process title node */
static bool html_process_title(html_content *c, dom_node *node)
{
dom_exception exc; /* returned by libdom functions */
dom_string *title;
char *title_str;
bool success;
exc = dom_node_get_text_content(node, &title);
if ((exc != DOM_NO_ERR) || (title == NULL)) {
return false;
}
title_str = squash_whitespace(dom_string_data(title));
dom_string_unref(title);
if (title_str == NULL) {
return false;
}
success = content__set_title(&c->base, title_str);
free(title_str);
return success;
}
static bool html_process_base(html_content *c, dom_node *node)
{
dom_exception exc; /* returned by libdom functions */
dom_string *atr_string;
/* get href attribute if present */
exc = dom_element_get_attribute(node,
corestring_dom_href, &atr_string);
if ((exc == DOM_NO_ERR) && (atr_string != NULL)) {
nsurl *url;
nserror error;
/* get url from string */
error = nsurl_create(dom_string_data(atr_string), &url);
dom_string_unref(atr_string);
if (error == NSERROR_OK) {
if (c->base_url != NULL)
nsurl_unref(c->base_url);
c->base_url = url;
}
}
/* get target attribute if present and not already set */
if (c->base_target != NULL) {
return true;
}
exc = dom_element_get_attribute(node,
corestring_dom_target, &atr_string);
if ((exc == DOM_NO_ERR) && (atr_string != NULL)) {
/* Validation rules from the HTML5 spec for the base element:
* The target must be one of _blank, _self, _parent, or
* _top or any identifier which does not begin with an
* underscore
*/
if (*dom_string_data(atr_string) != '_' ||
dom_string_caseless_lwc_isequal(atr_string,
corestring_lwc__blank) ||
dom_string_caseless_lwc_isequal(atr_string,
corestring_lwc__self) ||
dom_string_caseless_lwc_isequal(atr_string,
corestring_lwc__parent) ||
dom_string_caseless_lwc_isequal(atr_string,
corestring_lwc__top)) {
c->base_target = strdup(dom_string_data(atr_string));
}
dom_string_unref(atr_string);
}
return true;
}
static nserror html_meta_refresh_process_element(html_content *c, dom_node *n)
{
union content_msg_data msg_data;
const char *url, *end, *refresh = NULL;
char *new_url;
char quote = '\0';
dom_string *equiv, *content;
dom_exception exc;
nsurl *nsurl;
nserror error = NSERROR_OK;
exc = dom_element_get_attribute(n, corestring_dom_http_equiv, &equiv);
if (exc != DOM_NO_ERR) {
return NSERROR_DOM;
}
if (equiv == NULL) {
return NSERROR_OK;
}
if (!dom_string_caseless_lwc_isequal(equiv, corestring_lwc_refresh)) {
dom_string_unref(equiv);
return NSERROR_OK;
}
dom_string_unref(equiv);
exc = dom_element_get_attribute(n, corestring_dom_content, &content);
if (exc != DOM_NO_ERR) {
return NSERROR_DOM;
}
if (content == NULL) {
return NSERROR_OK;
}
end = dom_string_data(content) + dom_string_byte_length(content);
/* content := *LWS intpart fracpart? *LWS [';' *LWS *1url *LWS]
* intpart := 1*DIGIT
* fracpart := 1*('.' | DIGIT)
* url := "url" *LWS '=' *LWS (url-nq | url-sq | url-dq)
* url-nq := *urlchar
* url-sq := "'" *(urlchar | '"') "'"
* url-dq := '"' *(urlchar | "'") '"'
* urlchar := [#x9#x21#x23-#x26#x28-#x7E] | nonascii
* nonascii := [#x80-#xD7FF#xE000-#xFFFD#x10000-#x10FFFF]
*/
url = dom_string_data(content);
/* *LWS */
while (url < end && isspace(*url)) {
url++;
}
/* intpart */
if (url == end || (*url < '0' || '9' < *url)) {
/* Empty content, or invalid timeval */
dom_string_unref(content);
return NSERROR_OK;
}
msg_data.delay = (int) strtol(url, &new_url, 10);
/* a very small delay and self-referencing URL can cause a loop
* that grinds machines to a halt. To prevent this we set a
* minimum refresh delay of 1s. */
if (msg_data.delay < 1) {
msg_data.delay = 1;
}
url = new_url;
/* fracpart? (ignored, as delay is integer only) */
while (url < end && (('0' <= *url && *url <= '9') ||
*url == '.')) {
url++;
}
/* *LWS */
while (url < end && isspace(*url)) {
url++;
}
/* ';' */
if (url < end && *url == ';')
url++;
/* *LWS */
while (url < end && isspace(*url)) {
url++;
}
if (url == end) {
/* Just delay specified, so refresh current page */
dom_string_unref(content);
c->base.refresh = nsurl_ref(
content_get_url(&c->base));
content_broadcast(&c->base, CONTENT_MSG_REFRESH, msg_data);
return NSERROR_OK;
}
/* "url" */
if (url <= end - 3) {
if (strncasecmp(url, "url", 3) == 0) {
url += 3;
} else {
/* Unexpected input, ignore this header */
dom_string_unref(content);
return NSERROR_OK;
}
} else {
/* Insufficient input, ignore this header */
dom_string_unref(content);
return NSERROR_OK;
}
/* *LWS */
while (url < end && isspace(*url)) {
url++;
}
/* '=' */
if (url < end) {
if (*url == '=') {
url++;
} else {
/* Unexpected input, ignore this header */
dom_string_unref(content);
return NSERROR_OK;
}
} else {
/* Insufficient input, ignore this header */
dom_string_unref(content);
return NSERROR_OK;
}
/* *LWS */
while (url < end && isspace(*url)) {
url++;
}
/* '"' or "'" */
if (url < end && (*url == '"' || *url == '\'')) {
quote = *url;
url++;
}
/* Start of URL */
refresh = url;
if (quote != 0) {
/* url-sq | url-dq */
while (url < end && *url != quote)
url++;
} else {
/* url-nq */
while (url < end && !isspace(*url))
url++;
}
/* '"' or "'" or *LWS (we don't care) */
if (url > refresh) {
/* There's a URL */
new_url = strndup(refresh, url - refresh);
if (new_url == NULL) {
dom_string_unref(content);
return NSERROR_NOMEM;
}
error = nsurl_join(c->base_url, new_url, &nsurl);
if (error == NSERROR_OK) {
/* broadcast valid refresh url */
c->base.refresh = nsurl;
content_broadcast(&c->base, CONTENT_MSG_REFRESH, msg_data);
c->refresh = true;
}
free(new_url);
}
dom_string_unref(content);
return error;
}
static bool html_process_img(html_content *c, dom_node *node)
{
dom_string *src;
nsurl *url;
nserror err;
dom_exception exc;
bool success;
/* Do nothing if foreground images are disabled */
if (nsoption_bool(foreground_images) == false) {
return true;
}
exc = dom_element_get_attribute(node, corestring_dom_src, &src);
if (exc != DOM_NO_ERR || src == NULL) {
return true;
}
err = nsurl_join(c->base_url, dom_string_data(src), &url);
if (err != NSERROR_OK) {
dom_string_unref(src);
return false;
}
dom_string_unref(src);
/* Speculatively fetch the image */
success = html_fetch_object(c, url, NULL, CONTENT_IMAGE, 0, 0, false);
nsurl_unref(url);
return success;
}
/* exported function documented in render/html_internal.h */
void html_finish_conversion(html_content *htmlc)
{
union content_msg_data msg_data;
dom_exception exc; /* returned by libdom functions */
dom_node *html;
nserror error;
/* Bail out if we've been aborted */
if (htmlc->aborted) {
content_broadcast_errorcode(&htmlc->base, NSERROR_STOPPED);
content_set_error(&htmlc->base);
return;
}
/* create new css selection context */
error = html_css_new_selection_context(htmlc, &htmlc->select_ctx);
if (error != NSERROR_OK) {
content_broadcast_errorcode(&htmlc->base, error);
content_set_error(&htmlc->base);
return;
}
/* fire a simple event named load at the Document's Window
* object, but with its target set to the Document object (and
* the currentTarget set to the Window object)
*/
js_fire_event(htmlc->jscontext, "load", htmlc->document, NULL);
/* convert dom tree to box tree */
LOG(("DOM to box (%p)", htmlc));
content_set_status(&htmlc->base, messages_get("Processing"));
msg_data.explicit_status_text = NULL;
content_broadcast(&htmlc->base, CONTENT_MSG_STATUS, msg_data);
exc = dom_document_get_document_element(htmlc->document, (void *) &html);
if ((exc != DOM_NO_ERR) || (html == NULL)) {
LOG(("error retrieving html element from dom"));
content_broadcast_errorcode(&htmlc->base, NSERROR_DOM);
content_set_error(&htmlc->base);
return;
}
error = dom_to_box(html, htmlc, html_box_convert_done);
if (error != NSERROR_OK) {
LOG(("box conversion failed"));
dom_node_unref(html);
html_object_free_objects(htmlc);
content_broadcast_errorcode(&htmlc->base, error);
content_set_error(&htmlc->base);
return;
}
dom_node_unref(html);
}
/* callback for DOMNodeInserted end type */
static void
dom_default_action_DOMNodeInserted_cb(struct dom_event *evt, void *pw)
{
dom_event_target *node;
dom_node_type type;
dom_string *name;
dom_exception exc;
html_content *htmlc = pw;
exc = dom_event_get_target(evt, &node);
if ((exc == DOM_NO_ERR) && (node != NULL)) {
exc = dom_node_get_node_type(node, &type);
if ((exc == DOM_NO_ERR) && (type == DOM_ELEMENT_NODE)) {
/* an element node has been inserted */
exc = dom_node_get_node_name(node, &name);
if ((exc == DOM_NO_ERR) && (name != NULL)) {
if (dom_string_caseless_isequal(name,
corestring_dom_link)) {
/* Handle stylesheet loading */
html_css_process_link(htmlc,
(dom_node *)node);
/* Generic link handling */
html_process_link(htmlc,
(dom_node *)node);
} else if (dom_string_caseless_lwc_isequal(name,
corestring_lwc_meta) &&
htmlc->refresh == false) {
html_meta_refresh_process_element(htmlc,
(dom_node *)node);
} else if (dom_string_caseless_lwc_isequal(
name, corestring_lwc_base)) {
html_process_base(htmlc,
(dom_node *)node);
} else if (dom_string_caseless_lwc_isequal(
name, corestring_lwc_title) &&
htmlc->title == NULL) {
htmlc->title = dom_node_ref(node);
} else if (dom_string_caseless_lwc_isequal(
name, corestring_lwc_img)) {
html_process_img(htmlc,
(dom_node *) node);
}
dom_string_unref(name);
}
}
dom_node_unref(node);
}
}
/* callback for DOMNodeInserted end type */
static void
dom_default_action_DOMSubtreeModified_cb(struct dom_event *evt, void *pw)
{
dom_event_target *node;
dom_node_type type;
dom_string *name;
dom_exception exc;
html_content *htmlc = pw;
exc = dom_event_get_target(evt, &node);
if ((exc == DOM_NO_ERR) && (node != NULL)) {
if (htmlc->title == (dom_node *)node) {
/* Node is our title node */
html_process_title(htmlc, (dom_node *)node);
dom_node_unref(node);
return;
}
exc = dom_node_get_node_type(node, &type);
if ((exc == DOM_NO_ERR) && (type == DOM_ELEMENT_NODE)) {
/* an element node has been modified */
exc = dom_node_get_node_name(node, &name);
if ((exc == DOM_NO_ERR) && (name != NULL)) {
if (dom_string_caseless_isequal(name,
corestring_dom_style)) {
html_css_update_style(htmlc,
(dom_node *)node);
}
dom_string_unref(name);
}
}
dom_node_unref(node);
}
}
/* callback function selector
*
* selects a callback function for libdom to call based on the type and phase.
* dom_default_action_phase from events/document_event.h
*
* The principle events are:
* DOMSubtreeModified
* DOMAttrModified
* DOMNodeInserted
* DOMNodeInsertedIntoDocument
*
* @return callback function pointer or NULL for none
*/
static dom_default_action_callback
dom_event_fetcher(dom_string *type,
dom_default_action_phase phase,
void **pw)
{
//LOG(("type:%s", dom_string_data(type)));
if (phase == DOM_DEFAULT_ACTION_END) {
if (dom_string_isequal(type, corestring_dom_DOMNodeInserted)) {
return dom_default_action_DOMNodeInserted_cb;
} else if (dom_string_isequal(type, corestring_dom_DOMSubtreeModified)) {
return dom_default_action_DOMSubtreeModified_cb;
}
}
return NULL;
}
static nserror
html_create_html_data(html_content *c, const http_parameter *params)
{
lwc_string *charset;
nserror nerror;
dom_hubbub_parser_params parse_params;
dom_hubbub_error error;
c->parser = NULL;
c->parse_completed = false;
c->document = NULL;
c->quirks = DOM_DOCUMENT_QUIRKS_MODE_NONE;
c->encoding = NULL;
c->base_url = nsurl_ref(content_get_url(&c->base));
c->base_target = NULL;
c->aborted = false;
c->refresh = false;
c->reflowing = false;
c->title = NULL;
c->bctx = NULL;
c->layout = NULL;
c->background_colour = NS_TRANSPARENT;
c->stylesheet_count = 0;
c->stylesheets = NULL;
c->select_ctx = NULL;
c->universal = NULL;
c->num_objects = 0;
c->object_list = NULL;
c->forms = NULL;
c->imagemaps = NULL;
c->bw = NULL;
c->frameset = NULL;
c->iframe = NULL;
c->page = NULL;
c->font_func = &nsfont;
c->drag_type = HTML_DRAG_NONE;
c->drag_owner.no_owner = true;
c->selection_type = HTML_SELECTION_NONE;
c->selection_owner.none = true;
c->focus_type = HTML_FOCUS_SELF;
c->focus_owner.self = true;
c->search = NULL;
c->search_string = NULL;
c->scripts_count = 0;
c->scripts = NULL;
c->jscontext = NULL;
c->base.active = 1; /* The html content itself is active */
if (lwc_intern_string("*", SLEN("*"), &c->universal) != lwc_error_ok) {
return NSERROR_NOMEM;
}
selection_prepare(&c->sel, (struct content *)c, true);
nerror = http_parameter_list_find_item(params, corestring_lwc_charset, &charset);
if (nerror == NSERROR_OK) {
c->encoding = strdup(lwc_string_data(charset));
lwc_string_unref(charset);
if (c->encoding == NULL) {
lwc_string_unref(c->universal);
c->universal = NULL;
return NSERROR_NOMEM;
}
c->encoding_source = DOM_HUBBUB_ENCODING_SOURCE_HEADER;
}
/* Create the parser binding */
parse_params.enc = c->encoding;
parse_params.fix_enc = true;
parse_params.enable_script = nsoption_bool(enable_javascript);
parse_params.msg = NULL;
parse_params.script = html_process_script;
parse_params.ctx = c;
parse_params.daf = dom_event_fetcher;
error = dom_hubbub_parser_create(&parse_params,
&c->parser,
&c->document);
if ((error != DOM_HUBBUB_OK) && (c->encoding != NULL)) {
/* Ok, we don't support the declared encoding. Bailing out
* isn't exactly user-friendly, so fall back to autodetect */
free(c->encoding);
c->encoding = NULL;
parse_params.enc = c->encoding;
error = dom_hubbub_parser_create(&parse_params,
&c->parser,
&c->document);
}
if (error != DOM_HUBBUB_OK) {
nsurl_unref(c->base_url);
c->base_url = NULL;
lwc_string_unref(c->universal);
c->universal = NULL;
return libdom_hubbub_error_to_nserror(error);
}
return NSERROR_OK;
}
/**
* Create a CONTENT_HTML.
*
* The content_html_data structure is initialized and the HTML parser is
* created.
*/
static nserror
html_create(const content_handler *handler,
lwc_string *imime_type,
const http_parameter *params,
llcache_handle *llcache,
const char *fallback_charset,
bool quirks,
struct content **c)
{
html_content *html;
nserror error;
html = calloc(1, sizeof(html_content));
if (html == NULL)
return NSERROR_NOMEM;
error = content__init(&html->base, handler, imime_type, params,
llcache, fallback_charset, quirks);
if (error != NSERROR_OK) {
free(html);
return error;
}
error = html_create_html_data(html, params);
if (error != NSERROR_OK) {
content_broadcast_errorcode(&html->base, error);
free(html);
return error;
}
error = html_css_new_stylesheets(html);
if (error != NSERROR_OK) {
content_broadcast_errorcode(&html->base, error);
free(html);
return error;
}
*c = (struct content *) html;
return NSERROR_OK;
}
static nserror
html_process_encoding_change(struct content *c,
const char *data,
unsigned int size)
{
html_content *html = (html_content *) c;
dom_hubbub_parser_params parse_params;
dom_hubbub_error error;
const char *encoding;
const char *source_data;
unsigned long source_size;
/* Retrieve new encoding */
encoding = dom_hubbub_parser_get_encoding(html->parser,
&html->encoding_source);
if (encoding == NULL) {
return NSERROR_NOMEM;
}
if (html->encoding != NULL) {
free(html->encoding);
html->encoding = NULL;
}
html->encoding = strdup(encoding);
if (html->encoding == NULL) {
return NSERROR_NOMEM;
}
/* Destroy binding */
dom_hubbub_parser_destroy(html->parser);
html->parser = NULL;
if (html->document != NULL) {
dom_node_unref(html->document);
}
parse_params.enc = html->encoding;
parse_params.fix_enc = true;
parse_params.enable_script = nsoption_bool(enable_javascript);
parse_params.msg = NULL;
parse_params.script = html_process_script;
parse_params.ctx = html;
parse_params.daf = dom_event_fetcher;
/* Create new binding, using the new encoding */
error = dom_hubbub_parser_create(&parse_params,
&html->parser,
&html->document);
if (error != DOM_HUBBUB_OK) {
/* Ok, we don't support the declared encoding. Bailing out
* isn't exactly user-friendly, so fall back to Windows-1252 */
free(html->encoding);
html->encoding = strdup("Windows-1252");
if (html->encoding == NULL) {
return NSERROR_NOMEM;
}
parse_params.enc = html->encoding;
error = dom_hubbub_parser_create(&parse_params,
&html->parser,
&html->document);
if (error != DOM_HUBBUB_OK) {
return libdom_hubbub_error_to_nserror(error);
}
}
source_data = content__get_source_data(c, &source_size);
/* Reprocess all the data. This is safe because
* the encoding is now specified at parser start which means
* it cannot be changed again.
*/
error = dom_hubbub_parser_parse_chunk(html->parser,
(const uint8_t *)source_data,
source_size);
return libdom_hubbub_error_to_nserror(error);
}
/**
* Process data for CONTENT_HTML.
*/
static bool
html_process_data(struct content *c, const char *data, unsigned int size)
{
html_content *html = (html_content *) c;
dom_hubbub_error dom_ret;
nserror err = NSERROR_OK; /* assume its all going to be ok */
dom_ret = dom_hubbub_parser_parse_chunk(html->parser,
(const uint8_t *) data,
size);
err = libdom_hubbub_error_to_nserror(dom_ret);
/* deal with encoding change */
if (err == NSERROR_ENCODING_CHANGE) {
err = html_process_encoding_change(c, data, size);
}
/* broadcast the error if necessary */
if (err != NSERROR_OK) {
content_broadcast_errorcode(c, err);
return false;
}
return true;
}
/**
* Convert a CONTENT_HTML for display.
*
* The following steps are carried out in order:
*
* - parsing to an XML tree is completed
* - stylesheets are fetched
* - the XML tree is converted to a box tree and object fetches are started
*
* On exit, the content status will be either CONTENT_STATUS_DONE if the
* document is completely loaded or CONTENT_STATUS_READY if objects are still
* being fetched.
*/
static bool html_convert(struct content *c)
{
html_content *htmlc = (html_content *) c;
dom_exception exc; /* returned by libdom functions */
/* The quirk check and associated stylesheet fetch is "safe"
* once the root node has been inserted into the document
* which must have happened by this point in the parse.
*
* faliure to retrive the quirk mode or to start the
* stylesheet fetch is non fatal as this "only" affects the
* render and it would annoy the user to fail the entire
* render for want of a quirks stylesheet.
*/
exc = dom_document_get_quirks_mode(htmlc->document, &htmlc->quirks);
if (exc == DOM_NO_ERR) {
html_css_quirks_stylesheets(htmlc);
LOG(("quirks set to %d", htmlc->quirks));
}
htmlc->base.active--; /* the html fetch is no longer active */
LOG(("%d fetches active", htmlc->base.active));
/* The parse cannot be completed here because it may be paused
* untill all the resources being fetched have completed.
*/
/* if there are no active fetches in progress no scripts are
* being fetched or they completed already.
*/
if (html_can_begin_conversion(htmlc)) {
return html_begin_conversion(htmlc);
}
return true;
}
/* Exported interface documented in html_internal.h */
bool html_can_begin_conversion(html_content *htmlc)
{
unsigned int i;
if (htmlc->base.active != 0)
return false;
for (i = 0; i != htmlc->stylesheet_count; i++) {
if (htmlc->stylesheets[i].modified)
return false;
}
return true;
}
bool
html_begin_conversion(html_content *htmlc)
{
dom_node *html;
nserror ns_error;
struct form *f;
dom_exception exc; /* returned by libdom functions */
dom_string *node_name = NULL;
dom_hubbub_error error;
/* The act of completing the parse can result in additional data
* being flushed through the parser. This may result in new style or
* script nodes, upon which the conversion depends. Thus, once we
* have completed the parse, we must check again to see if we can
* begin the conversion. If we can't, we must stop and wait for the
* new styles/scripts to be processed. Once they have been processed,
* we will be called again to begin the conversion for real. Thus,
* we must also ensure that we don't attempt to complete the parse
* multiple times, so store a flag to indicate that parsing is
* complete to avoid repeating the completion pointlessly.
*/
if (htmlc->parse_completed == false) {
LOG(("Completing parse"));
/* complete parsing */
error = dom_hubbub_parser_completed(htmlc->parser);
if (error != DOM_HUBBUB_OK) {
LOG(("Parsing failed"));
content_broadcast_errorcode(&htmlc->base,
libdom_hubbub_error_to_nserror(error));
return false;
}
htmlc->parse_completed = true;
}
if (html_can_begin_conversion(htmlc) == false) {
/* We can't proceed (see commentary above) */
return true;
}
/* Give up processing if we've been aborted */
if (htmlc->aborted) {
content_broadcast_errorcode(&htmlc->base, NSERROR_STOPPED);
return false;
}
/* complete script execution */
html_scripts_exec(htmlc);
/* fire a simple event that bubbles named DOMContentLoaded at
* the Document.
*/
/* get encoding */
if (htmlc->encoding == NULL) {
const char *encoding;
encoding = dom_hubbub_parser_get_encoding(htmlc->parser,
&htmlc->encoding_source);
if (encoding == NULL) {
content_broadcast_errorcode(&htmlc->base,
NSERROR_NOMEM);
return false;
}
htmlc->encoding = strdup(encoding);
if (htmlc->encoding == NULL) {
content_broadcast_errorcode(&htmlc->base,
NSERROR_NOMEM);
return false;
}
}
/* locate root element and ensure it is html */
exc = dom_document_get_document_element(htmlc->document, (void *) &html);
if ((exc != DOM_NO_ERR) || (html == NULL)) {
LOG(("error retrieving html element from dom"));
content_broadcast_errorcode(&htmlc->base, NSERROR_DOM);
return false;
}
exc = dom_node_get_node_name(html, &node_name);
if ((exc != DOM_NO_ERR) ||
(node_name == NULL) ||
(!dom_string_caseless_lwc_isequal(node_name,
corestring_lwc_html))) {
LOG(("root element not html"));
content_broadcast_errorcode(&htmlc->base, NSERROR_DOM);
dom_node_unref(html);
return false;
}
dom_string_unref(node_name);
/* Retrieve forms from parser */
htmlc->forms = html_forms_get_forms(htmlc->encoding,
(dom_html_document *) htmlc->document);
for (f = htmlc->forms; f != NULL; f = f->prev) {
nsurl *action;
/* Make all actions absolute */
if (f->action == NULL || f->action[0] == '\0') {
/* HTML5 4.10.22.3 step 9 */
nsurl *doc_addr = content_get_url(&htmlc->base);
ns_error = nsurl_join(htmlc->base_url,
nsurl_access(doc_addr),
&action);
} else {
ns_error = nsurl_join(htmlc->base_url,
f->action,
&action);
}
if (ns_error != NSERROR_OK) {
content_broadcast_errorcode(&htmlc->base, ns_error);
dom_node_unref(html);
return false;
}
free(f->action);
f->action = strdup(nsurl_access(action));
nsurl_unref(action);
if (f->action == NULL) {
content_broadcast_errorcode(&htmlc->base,
NSERROR_NOMEM);
dom_node_unref(html);
return false;
}
/* Ensure each form has a document encoding */
if (f->document_charset == NULL) {
f->document_charset = strdup(htmlc->encoding);
if (f->document_charset == NULL) {
content_broadcast_errorcode(&htmlc->base,
NSERROR_NOMEM);
dom_node_unref(html);
return false;
}
}
}
dom_node_unref(html);
if (htmlc->base.active == 0) {
html_finish_conversion(htmlc);
}
return true;
}
/**
* Stop loading a CONTENT_HTML.
*/
static void html_stop(struct content *c)
{
html_content *htmlc = (html_content *) c;
switch (c->status) {
case CONTENT_STATUS_LOADING:
/* Still loading; simply flag that we've been aborted
* html_convert/html_finish_conversion will do the rest */
htmlc->aborted = true;
break;
case CONTENT_STATUS_READY:
html_object_abort_objects(htmlc);
/* If there are no further active fetches and we're still
* in the READY state, transition to the DONE state. */
if (c->status == CONTENT_STATUS_READY && c->active == 0) {
html_set_status(htmlc, "");
content_set_done(c);
}
break;
case CONTENT_STATUS_DONE:
/* Nothing to do */
break;
default:
LOG(("Unexpected status %d", c->status));
assert(0);
}
}
/**
* Reformat a CONTENT_HTML to a new width.
*/
static void html_reformat(struct content *c, int width, int height)
{
html_content *htmlc = (html_content *) c;
struct box *layout;
unsigned int time_before, time_taken;
time_before = wallclock();
htmlc->reflowing = true;
layout_document(htmlc, width, height);
layout = htmlc->layout;
/* width and height are at least margin box of document */
c->width = layout->x + layout->padding[LEFT] + layout->width +
layout->padding[RIGHT] + layout->border[RIGHT].width +
layout->margin[RIGHT];
c->height = layout->y + layout->padding[TOP] + layout->height +
layout->padding[BOTTOM] + layout->border[BOTTOM].width +
layout->margin[BOTTOM];
/* if boxes overflow right or bottom edge, expand to contain it */
if (c->width < layout->x + layout->descendant_x1)
c->width = layout->x + layout->descendant_x1;
if (c->height < layout->y + layout->descendant_y1)
c->height = layout->y + layout->descendant_y1;
selection_reinit(&htmlc->sel, htmlc->layout);
htmlc->reflowing = false;
time_taken = wallclock() - time_before;
c->reformat_time = wallclock() +
((time_taken * 3 < nsoption_uint(min_reflow_period) ?
nsoption_uint(min_reflow_period) : time_taken * 3));
}
/**
* Redraw a box.
*
* \param h content containing the box, of type CONTENT_HTML
* \param box box to redraw
*/
void html_redraw_a_box(hlcache_handle *h, struct box *box)
{
int x, y;
box_coords(box, &x, &y);
content_request_redraw(h, x, y,
box->padding[LEFT] + box->width + box->padding[RIGHT],
box->padding[TOP] + box->height + box->padding[BOTTOM]);
}
/**
* Redraw a box.
*
* \param html content containing the box, of type CONTENT_HTML
* \param box box to redraw.
*/
void html__redraw_a_box(struct html_content *html, struct box *box)
{
int x, y;
box_coords(box, &x, &y);
content__request_redraw((struct content *)html, x, y,
box->padding[LEFT] + box->width + box->padding[RIGHT],
box->padding[TOP] + box->height + box->padding[BOTTOM]);
}
static void html_destroy_frameset(struct content_html_frames *frameset)
{
int i;
if (frameset->name) {
talloc_free(frameset->name);
frameset->name = NULL;
}
if (frameset->url) {
talloc_free(frameset->url);
frameset->url = NULL;
}
if (frameset->children) {
for (i = 0; i < (frameset->rows * frameset->cols); i++) {
if (frameset->children[i].name) {
talloc_free(frameset->children[i].name);
frameset->children[i].name = NULL;
}
if (frameset->children[i].url) {
nsurl_unref(frameset->children[i].url);
frameset->children[i].url = NULL;
}
if (frameset->children[i].children)
html_destroy_frameset(&frameset->children[i]);
}
talloc_free(frameset->children);
frameset->children = NULL;
}
}
static void html_destroy_iframe(struct content_html_iframe *iframe)
{
struct content_html_iframe *next;
next = iframe;
while ((iframe = next) != NULL) {
next = iframe->next;
if (iframe->name)
talloc_free(iframe->name);
if (iframe->url) {
nsurl_unref(iframe->url);
iframe->url = NULL;
}
talloc_free(iframe);
}
}
static void html_free_layout(html_content *htmlc)
{
if (htmlc->bctx != NULL) {
/* freeing talloc context should let the entire box
* set be destroyed
*/
talloc_free(htmlc->bctx);
}
}
/**
* Destroy a CONTENT_HTML and free all resources it owns.
*/
static void html_destroy(struct content *c)
{
html_content *html = (html_content *) c;
struct form *f, *g;
LOG(("content %p", c));
/* Destroy forms */
for (f = html->forms; f != NULL; f = g) {
g = f->prev;
form_free(f);
}
imagemap_destroy(html);
if (c->refresh)
nsurl_unref(c->refresh);
if (html->base_url)
nsurl_unref(html->base_url);
if (html->parser != NULL) {
dom_hubbub_parser_destroy(html->parser);
html->parser = NULL;
}
if (html->document != NULL) {
dom_node_unref(html->document);
html->document = NULL;
}
if (html->title != NULL) {
dom_node_unref(html->title);
html->title = NULL;
}
/* Free encoding */
if (html->encoding != NULL) {
free(html->encoding);
html->encoding = NULL;
}
/* Free base target */
if (html->base_target != NULL) {
free(html->base_target);
html->base_target = NULL;
}
/* Free frameset */
if (html->frameset != NULL) {
html_destroy_frameset(html->frameset);
talloc_free(html->frameset);
html->frameset = NULL;
}
/* Free iframes */
if (html->iframe != NULL) {
html_destroy_iframe(html->iframe);
html->iframe = NULL;
}
/* Destroy selection context */
if (html->select_ctx != NULL) {
css_select_ctx_destroy(html->select_ctx);
html->select_ctx = NULL;
}
if (html->universal != NULL) {
lwc_string_unref(html->universal);
html->universal = NULL;
}
/* Free stylesheets */
html_css_free_stylesheets(html);
/* Free scripts */
html_free_scripts(html);
/* Free objects */
html_object_free_objects(html);
/* free layout */
html_free_layout(html);
}
static nserror html_clone(const struct content *old, struct content **newc)
{
/** \todo Clone HTML specifics */
/* In the meantime, we should never be called, as HTML contents
* cannot be shared and we're not intending to fix printing's
* cloning of documents. */
assert(0 && "html_clone should never be called");
return true;
}
/**
* Set the content status.
*/
void html_set_status(html_content *c, const char *extra)
{
content_set_status(&c->base, extra);
}
/**
* Handle a window containing a CONTENT_HTML being opened.
*/
static void
html_open(struct content *c,
struct browser_window *bw,
struct content *page,
struct object_params *params)
{
html_content *html = (html_content *) c;
html->bw = bw;
html->page = (html_content *) page;
html->drag_type = HTML_DRAG_NONE;
html->drag_owner.no_owner = true;
/* text selection */
selection_init(&html->sel, html->layout);
html->selection_type = HTML_SELECTION_NONE;
html->selection_owner.none = true;
html_object_open_objects(html, bw);
}
/**
* Handle a window containing a CONTENT_HTML being closed.
*/
static void html_close(struct content *c)
{
html_content *html = (html_content *) c;
selection_clear(&html->sel, false);
if (html->search != NULL)
search_destroy_context(html->search);
html->bw = NULL;
html_object_close_objects(html);
}
/**
* Return an HTML content's selection context
*/
static void html_clear_selection(struct content *c)
{
html_content *html = (html_content *) c;
switch (html->selection_type) {
case HTML_SELECTION_NONE:
/* Nothing to do */
assert(html->selection_owner.none == true);
break;
case HTML_SELECTION_TEXTAREA:
textarea_clear_selection(html->selection_owner.textarea->
gadget->data.text.ta);
break;
case HTML_SELECTION_SELF:
assert(html->selection_owner.none == false);
selection_clear(&html->sel, true);
break;
case HTML_SELECTION_CONTENT:
content_clear_selection(html->selection_owner.content->object);
break;
default:
break;
}
/* There is no selection now. */
html->selection_type = HTML_SELECTION_NONE;
html->selection_owner.none = true;
}
/**
* Return an HTML content's selection context
*/
static char *html_get_selection(struct content *c)
{
html_content *html = (html_content *) c;
switch (html->selection_type) {
case HTML_SELECTION_TEXTAREA:
return textarea_get_selection(html->selection_owner.textarea->
gadget->data.text.ta);
case HTML_SELECTION_SELF:
assert(html->selection_owner.none == false);
return selection_get_copy(&html->sel);
case HTML_SELECTION_CONTENT:
return content_get_selection(
html->selection_owner.content->object);
case HTML_SELECTION_NONE:
/* Nothing to do */
assert(html->selection_owner.none == true);
break;
default:
break;
}
return NULL;
}
/**
* Get access to any content, link URLs and objects (images) currently
* at the given (x, y) coordinates.
*
* \param[in] c html content to look inside
* \param[in] x x-coordinate of point of interest
* \param[in] y y-coordinate of point of interest
* \param[out] data Positional features struct to be updated with any
* relevent content, or set to NULL if none.
* \return NSERROR_OK on success else appropriate error code.
*/
static nserror
html_get_contextual_content(struct content *c, int x, int y,
struct browser_window_features *data)
{
html_content *html = (html_content *) c;
struct box *box = html->layout;
struct box *next;
int box_x = 0, box_y = 0;
while ((next = box_at_point(box, x, y, &box_x, &box_y)) != NULL) {
box = next;
/* hidden boxes are ignored */
if ((box->style != NULL) &&
css_computed_visibility(box->style) == CSS_VISIBILITY_HIDDEN) {
continue;
}
if (box->iframe) {
browser_window_get_features(box->iframe,
x - box_x, y - box_y, data);
}
if (box->object)
content_get_contextual_content(box->object,
x - box_x, y - box_y, data);
if (box->object)
data->object = box->object;
if (box->href)
data->link = box->href;
if (box->usemap) {
const char *target = NULL;
nsurl *url = imagemap_get(html, box->usemap, box_x,
box_y, x, y, &target);
/* Box might have imagemap, but no actual link area
* at point */
if (url != NULL)
data->link = url;
}
if (box->gadget) {
switch (box->gadget->type) {
case GADGET_TEXTBOX:
case GADGET_TEXTAREA:
case GADGET_PASSWORD:
data->form_features = CTX_FORM_TEXT;
break;
case GADGET_FILE:
data->form_features = CTX_FORM_FILE;
break;
default:
data->form_features = CTX_FORM_NONE;
break;
}
}
}
return NSERROR_OK;
}
/**
* Scroll deepest thing within the content which can be scrolled at given point
*
* \param c html content to look inside
* \param x x-coordinate of point of interest
* \param y y-coordinate of point of interest
* \param scrx number of px try to scroll something in x direction
* \param scry number of px try to scroll something in y direction
* \return true iff scroll was consumed by something in the content
*/
static bool
html_scroll_at_point(struct content *c, int x, int y, int scrx, int scry)
{
html_content *html = (html_content *) c;
struct box *box = html->layout;
struct box *next;
int box_x = 0, box_y = 0;
bool handled_scroll = false;
/* TODO: invert order; visit deepest box first */
while ((next = box_at_point(box, x, y, &box_x, &box_y)) != NULL) {
box = next;
if (box->style && css_computed_visibility(box->style) ==
CSS_VISIBILITY_HIDDEN)
continue;
/* Pass into iframe */
if (box->iframe && browser_window_scroll_at_point(box->iframe,
x - box_x, y - box_y, scrx, scry) == true)
return true;
/* Pass into textarea widget */
if (box->gadget && (box->gadget->type == GADGET_TEXTAREA ||
box->gadget->type == GADGET_PASSWORD ||
box->gadget->type == GADGET_TEXTBOX) &&
textarea_scroll(box->gadget->data.text.ta,
scrx, scry) == true)
return true;
/* Pass into object */
if (box->object != NULL && content_scroll_at_point(
box->object, x - box_x, y - box_y,
scrx, scry) == true)
return true;
/* Handle box scrollbars */
if (box->scroll_y && scrollbar_scroll(box->scroll_y, scry))
handled_scroll = true;
if (box->scroll_x && scrollbar_scroll(box->scroll_x, scrx))
handled_scroll = true;
if (handled_scroll == true)
return true;
}
return false;
}
/** Helper for file gadgets to store their filename unencoded on the
* dom node associated with the gadget.
*
* \todo Get rid of this crap eventually
*/
static void html__dom_user_data_handler(dom_node_operation operation,
dom_string *key, void *_data, struct dom_node *src,
struct dom_node *dst)
{
char *oldfile;
char *data = (char *)_data;
if (!dom_string_isequal(corestring_dom___ns_key_file_name_node_data,
key) || data == NULL) {
return;
}
switch (operation) {
case DOM_NODE_CLONED:
if (dom_node_set_user_data(dst,
corestring_dom___ns_key_file_name_node_data,
strdup(data), html__dom_user_data_handler,
&oldfile) == DOM_NO_ERR) {
if (oldfile != NULL)
free(oldfile);
}
break;
case DOM_NODE_RENAMED:
case DOM_NODE_IMPORTED:
case DOM_NODE_ADOPTED:
break;
case DOM_NODE_DELETED:
free(data);
break;
default:
LOG(("User data operation not handled."));
assert(0);
}
}
static void html__set_file_gadget_filename(struct content *c,
struct form_control *gadget, const char *fn)
{
nserror ret;
char *utf8_fn, *oldfile = NULL;
html_content *html = (html_content *)c;
struct box *file_box = gadget->box;
ret = guit->utf8->local_to_utf8(fn, 0, &utf8_fn);
if (ret != NSERROR_OK) {
assert(ret != NSERROR_BAD_ENCODING);
LOG(("utf8 to local encoding conversion failed"));
/* Load was for us - just no memory */
return;
}
form_gadget_update_value(gadget, utf8_fn);
/* corestring_dom___ns_key_file_name_node_data */
if (dom_node_set_user_data((dom_node *)file_box->gadget->node,
corestring_dom___ns_key_file_name_node_data,
strdup(fn), html__dom_user_data_handler,
&oldfile) == DOM_NO_ERR) {
if (oldfile != NULL)
free(oldfile);
}
/* Redraw box. */
html__redraw_a_box(html, file_box);
}
void html_set_file_gadget_filename(struct hlcache_handle *hl,
struct form_control *gadget, const char *fn)
{
return html__set_file_gadget_filename(hlcache_handle_get_content(hl),
gadget, fn);
}
/**
* Drop a file onto a content at a particular point, or determine if a file
* may be dropped onto the content at given point.
*
* \param c html content to look inside
* \param x x-coordinate of point of interest
* \param y y-coordinate of point of interest
* \param file path to file to be dropped, or NULL to know if drop allowed
* \return true iff file drop has been handled, or if drop possible (NULL file)
*/
static bool html_drop_file_at_point(struct content *c, int x, int y, char *file)
{
html_content *html = (html_content *) c;
struct box *box = html->layout;
struct box *next;
struct box *file_box = NULL;
struct box *text_box = NULL;
int box_x = 0, box_y = 0;
/* Scan box tree for boxes that can handle drop */
while ((next = box_at_point(box, x, y, &box_x, &box_y)) != NULL) {
box = next;
if (box->style && css_computed_visibility(box->style) ==
CSS_VISIBILITY_HIDDEN)
continue;
if (box->iframe)
return browser_window_drop_file_at_point(box->iframe,
x - box_x, y - box_y, file);
if (box->object && content_drop_file_at_point(box->object,
x - box_x, y - box_y, file) == true)
return true;
if (box->gadget) {
switch (box->gadget->type) {
case GADGET_FILE:
file_box = box;
break;
case GADGET_TEXTBOX:
case GADGET_TEXTAREA:
case GADGET_PASSWORD:
text_box = box;
break;
default: /* appease compiler */
break;
}
}
}
if (!file_box && !text_box)
/* No box capable of handling drop */
return false;
if (file == NULL)
/* There is a box capable of handling drop here */
return true;
/* Handle the drop */
if (file_box) {
/* File dropped on file input */
html__set_file_gadget_filename(c, file_box->gadget, file);
} else {
/* File dropped on text input */
size_t file_len;
FILE *fp = NULL;
char *buffer;
char *utf8_buff;
nserror ret;
unsigned int size;
int bx, by;
/* Open file */
fp = fopen(file, "rb");
if (fp == NULL) {
/* Couldn't open file, but drop was for us */
return true;
}
/* Get filesize */
fseek(fp, 0, SEEK_END);
file_len = ftell(fp);
fseek(fp, 0, SEEK_SET);
/* Allocate buffer for file data */
buffer = malloc(file_len + 1);
if (buffer == NULL) {
/* No memory, but drop was for us */
fclose(fp);
return true;
}
/* Stick file into buffer */
if (file_len != fread(buffer, 1, file_len, fp)) {
/* Failed, but drop was for us */
free(buffer);
fclose(fp);
return true;
}
/* Done with file */
fclose(fp);
/* Ensure buffer's string termination */
buffer[file_len] = '\0';
/* TODO: Sniff for text? */
/* Convert to UTF-8 */
ret = guit->utf8->local_to_utf8(buffer, file_len, &utf8_buff);
if (ret != NSERROR_OK) {
/* bad encoding shouldn't happen */
assert(ret != NSERROR_BAD_ENCODING);
LOG(("local to utf8 encoding failed"));
free(buffer);
warn_user("NoMemory", NULL);
return true;
}
/* Done with buffer */
free(buffer);
/* Get new length */
size = strlen(utf8_buff);
/* Simulate a click over the input box, to place caret */
box_coords(text_box, &bx, &by);
textarea_mouse_action(text_box->gadget->data.text.ta,
BROWSER_MOUSE_PRESS_1, x - bx, y - by);
/* Paste the file as text */
textarea_drop_text(text_box->gadget->data.text.ta,
utf8_buff, size);
free(utf8_buff);
}
return true;
}
/**
* set debug status.
*
* \param c The content to debug
* \param op The debug operation type
*/
static nserror
html_debug(struct content *c, enum content_debug op)
{
html_redraw_debug = !html_redraw_debug;
return NSERROR_OK;
}
/**
* Dump debug info concerning the html_content
*
* \param c The content to debug
* \param f The file to dump to
* \param op The debug dump type
*/
static nserror
html_debug_dump(struct content *c, FILE *f, enum content_debug op)
{
html_content *htmlc = (html_content *)c;
dom_node *html;
dom_exception exc; /* returned by libdom functions */
nserror ret;
assert(htmlc != NULL);
if (op == CONTENT_DEBUG_RENDER) {
assert(htmlc->layout != NULL);
box_dump(f, htmlc->layout, 0, true);
ret = NSERROR_OK;
} else {
if (htmlc->document == NULL) {
LOG(("No document to dump"));
return NSERROR_DOM;
}
exc = dom_document_get_document_element(htmlc->document, (void *) &html);
if ((exc != DOM_NO_ERR) || (html == NULL)) {
LOG(("Unable to obtain root node"));
return NSERROR_DOM;
}
ret = libdom_dump_structure(html, f, 0);
LOG(("DOM structure dump returning %d", ret));
dom_node_unref(html);
}
return ret;
}
#if ALWAYS_DUMP_FRAMESET
/**
* Print a frameset tree to stderr.
*/
static void
html_dump_frameset(struct content_html_frames *frame, unsigned int depth)
{
unsigned int i;
int row, col, index;
const char *unit[] = {"px", "%", "*"};
const char *scrolling[] = {"auto", "yes", "no"};
assert(frame);
fprintf(stderr, "%p ", frame);
fprintf(stderr, "(%i %i) ", frame->rows, frame->cols);
fprintf(stderr, "w%g%s ", frame->width.value, unit[frame->width.unit]);
fprintf(stderr, "h%g%s ", frame->height.value,unit[frame->height.unit]);
fprintf(stderr, "(margin w%i h%i) ",
frame->margin_width, frame->margin_height);
if (frame->name)
fprintf(stderr, "'%s' ", frame->name);
if (frame->url)
fprintf(stderr, "<%s> ", frame->url);
if (frame->no_resize)
fprintf(stderr, "noresize ");
fprintf(stderr, "(scrolling %s) ", scrolling[frame->scrolling]);
if (frame->border)
fprintf(stderr, "border %x ",
(unsigned int) frame->border_colour);
fprintf(stderr, "\n");
if (frame->children) {
for (row = 0; row != frame->rows; row++) {
for (col = 0; col != frame->cols; col++) {
for (i = 0; i != depth; i++)
fprintf(stderr, " ");
fprintf(stderr, "(%i %i): ", row, col);
index = (row * frame->cols) + col;
html_dump_frameset(&frame->children[index],
depth + 1);
}
}
}
}
#endif
/**
* Retrieve HTML document tree
*
* \param h HTML content to retrieve document tree from
* \return Pointer to document tree
*/
dom_document *html_get_document(hlcache_handle *h)
{
html_content *c = (html_content *) hlcache_handle_get_content(h);
assert(c != NULL);
return c->document;
}
/**
* Retrieve box tree
*
* \param h HTML content to retrieve tree from
* \return Pointer to box tree
*
* \todo This API must die, as must all use of the box tree outside render/
*/
struct box *html_get_box_tree(hlcache_handle *h)
{
html_content *c = (html_content *) hlcache_handle_get_content(h);
assert(c != NULL);
return c->layout;
}
/**
* Retrieve the charset of an HTML document
*
* \param c Content to retrieve charset from
* \param op The content encoding operation to perform.
* \return Pointer to charset, or NULL
*/
static const char *html_encoding(const struct content *c, enum content_encoding_type op)
{
html_content *html = (html_content *) c;
static char enc_token[10] = "Encoding0";
assert(html != NULL);
if (op == CONTENT_ENCODING_SOURCE) {
enc_token[8] = '0' + html->encoding_source;
return messages_get(enc_token);
}
return html->encoding;
}
/**
* Retrieve framesets used in an HTML document
*
* \param h Content to inspect
* \return Pointer to framesets, or NULL if none
*/
struct content_html_frames *html_get_frameset(hlcache_handle *h)
{
html_content *c = (html_content *) hlcache_handle_get_content(h);
assert(c != NULL);
return c->frameset;
}
/**
* Retrieve iframes used in an HTML document
*
* \param h Content to inspect
* \return Pointer to iframes, or NULL if none
*/
struct content_html_iframe *html_get_iframe(hlcache_handle *h)
{
html_content *c = (html_content *) hlcache_handle_get_content(h);
assert(c != NULL);
return c->iframe;
}
/**
* Retrieve an HTML content's base URL
*
* \param h Content to retrieve base target from
* \return Pointer to URL
*/
nsurl *html_get_base_url(hlcache_handle *h)
{
html_content *c = (html_content *) hlcache_handle_get_content(h);
assert(c != NULL);
return c->base_url;
}
/**
* Retrieve an HTML content's base target
*
* \param h Content to retrieve base target from
* \return Pointer to target, or NULL if none
*/
const char *html_get_base_target(hlcache_handle *h)
{
html_content *c = (html_content *) hlcache_handle_get_content(h);
assert(c != NULL);
return c->base_target;
}
/**
* Retrieve layout coordinates of box with given id
*
* \param h HTML document to search
* \param frag_id String containing an element id
* \param x Updated to global x coord iff id found
* \param y Updated to global y coord iff id found
* \return true iff id found
*/
bool html_get_id_offset(hlcache_handle *h, lwc_string *frag_id, int *x, int *y)
{
struct box *pos;
struct box *layout;
if (content_get_type(h) != CONTENT_HTML)
return false;
layout = html_get_box_tree(h);
if ((pos = box_find_by_id(layout, frag_id)) != 0) {
box_coords(pos, x, y);
return true;
}
return false;
}
/**
* Compute the type of a content
*
* \return CONTENT_HTML
*/
static content_type html_content_type(void)
{
return CONTENT_HTML;
}
static void html_fini(void)
{
html_css_fini();
}
static const content_handler html_content_handler = {
.fini = html_fini,
.create = html_create,
.process_data = html_process_data,
.data_complete = html_convert,
.reformat = html_reformat,
.destroy = html_destroy,
.stop = html_stop,
.mouse_track = html_mouse_track,
.mouse_action = html_mouse_action,
.keypress = html_keypress,
.redraw = html_redraw,
.open = html_open,
.close = html_close,
.get_selection = html_get_selection,
.clear_selection = html_clear_selection,
.get_contextual_content = html_get_contextual_content,
.scroll_at_point = html_scroll_at_point,
.drop_file_at_point = html_drop_file_at_point,
.search = html_search,
.search_clear = html_search_clear,
.debug_dump = html_debug_dump,
.debug = html_debug,
.clone = html_clone,
.get_encoding = html_encoding,
.type = html_content_type,
.no_share = true,
};
nserror html_init(void)
{
uint32_t i;
nserror error;
error = html_css_init();
if (error != NSERROR_OK)
goto error;
for (i = 0; i < NOF_ELEMENTS(html_types); i++) {
error = content_factory_register_handler(html_types[i],
&html_content_handler);
if (error != NSERROR_OK)
goto error;
}
return NSERROR_OK;
error:
html_fini();
return error;
}
/**
* Get the browser window containing an HTML content
*
* \param c HTML content
* \return the browser window
*/
struct browser_window *html_get_browser_window(struct content *c)
{
html_content *html = (html_content *) c;
assert(c != NULL);
assert(c->handler == &html_content_handler);
return html->bw;
}