mirror of
https://gitlab.freedesktop.org/gstreamer/gstreamer.git
synced 2024-12-30 04:00:37 +00:00
82f5a3508c
Original commit message from CVS: * ext/cairo/gsttimeoverlay.c: (gst_cairo_time_overlay_update_font_height): * ext/gdk_pixbuf/pixbufscale.c: (gst_pixbufscale_transform_caps): * ext/jpeg/gstjpegdec.c: (gst_jpeg_dec_parse_image_data): * ext/jpeg/gstjpegenc.c: (gst_jpegenc_chain): * ext/jpeg/gstsmokedec.c: (gst_smokedec_chain): * ext/jpeg/gstsmokeenc.c: (gst_smokeenc_chain): * ext/libpng/gstpngdec.c: (user_endrow_callback): * gst/auparse/gstauparse.c: (gst_au_parse_parse_header): * gst/avi/gstavidemux.c: (gst_avi_demux_parse_superindex), (gst_avi_demux_parse_subindex), (gst_avi_demux_parse_stream), (gst_avi_demux_stream_data): * gst/cutter/gstcutter.c: (gst_cutter_chain): * gst/debug/efence.c: (gst_efence_buffer_alloc), (gst_fenced_buffer_copy): * gst/id3demux/id3v2frames.c: (id3demux_id3v2_parse_frame): * gst/matroska/matroska-demux.c: (gst_matroska_demux_add_stream): * gst/matroska/matroska-mux.c: (gst_matroska_mux_start): * gst/rtsp/gstrtspsrc.c: (gst_rtspsrc_send), (gst_rtspsrc_handle_message): * gst/wavparse/gstwavparse.c: (gst_wavparse_stream_headers): * sys/ximage/ximageutil.c: (ximageutil_xcontext_get): Printf format fixes.
1017 lines
29 KiB
C
1017 lines
29 KiB
C
/* -*- Mode: C; tab-width: 2; indent-tabs-mode: t; c-basic-offset: 2 -*- */
|
|
/* Copyright 2006 Tim-Philipp Müller <tim centricular net>
|
|
* Copyright 2005 Jan Schmidt <thaytan@mad.scientist.com>
|
|
* Copyright 2002,2003 Scott Wheeler <wheeler@kde.org> (portions from taglib)
|
|
*
|
|
* This library is free software; you can redistribute it and/or
|
|
* modify it under the terms of the GNU Library General Public
|
|
* License as published by the Free Software Foundation; either
|
|
* version 2 of the License, or (at your option) any later version.
|
|
*
|
|
* This library is distributed in the hope that it will be useful,
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
* Library General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU Library General Public
|
|
* License along with this library; if not, write to the
|
|
* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
|
|
* Boston, MA 02111-1307, USA.
|
|
*/
|
|
|
|
#ifdef HAVE_CONFIG_H
|
|
#include "config.h"
|
|
#endif
|
|
|
|
#include <string.h>
|
|
#include <stdlib.h>
|
|
#include <gst/tag/tag.h>
|
|
#include <gst/base/gsttypefindhelper.h>
|
|
|
|
#ifdef HAVE_ZLIB
|
|
#include <zlib.h>
|
|
#endif
|
|
|
|
#include "id3tags.h"
|
|
|
|
GST_DEBUG_CATEGORY_EXTERN (id3demux_debug);
|
|
#define GST_CAT_DEFAULT (id3demux_debug)
|
|
|
|
static gboolean parse_comment_frame (ID3TagsWorking * work);
|
|
static GArray *parse_text_identification_frame (ID3TagsWorking * work);
|
|
static gchar *parse_user_text_identification_frame (ID3TagsWorking * work,
|
|
const gchar ** tag_name);
|
|
static gchar *parse_unique_file_identifier (ID3TagsWorking * work,
|
|
const gchar ** tag_name);
|
|
static gboolean parse_relative_volume_adjustment_two (ID3TagsWorking * work);
|
|
static gboolean id3v2_tag_to_taglist (ID3TagsWorking * work,
|
|
const gchar * tag_name, const gchar * tag_str);
|
|
/* Parse a single string into an array of gchar* */
|
|
static void parse_split_strings (guint8 encoding, gchar * data, gint data_size,
|
|
GArray ** out_fields);
|
|
static void free_tag_strings (GArray * fields);
|
|
static gboolean
|
|
id3v2_genre_fields_to_taglist (ID3TagsWorking * work, const gchar * tag_name,
|
|
GArray * tag_fields);
|
|
static gboolean parse_picture_frame (ID3TagsWorking * work);
|
|
|
|
#define ID3V2_ENCODING_ISO8859 0x00
|
|
#define ID3V2_ENCODING_UTF16 0x01
|
|
#define ID3V2_ENCODING_UTF16BE 0x02
|
|
#define ID3V2_ENCODING_UTF8 0x03
|
|
|
|
extern guint read_synch_uint (guint8 * data, guint size);
|
|
|
|
gboolean
|
|
id3demux_id3v2_parse_frame (ID3TagsWorking * work)
|
|
{
|
|
const gchar *tag_name;
|
|
gboolean result = FALSE;
|
|
gint i;
|
|
guint8 *frame_data = work->hdr.frame_data;
|
|
guint frame_data_size = work->cur_frame_size;
|
|
gchar *tag_str = NULL;
|
|
GArray *tag_fields = NULL;
|
|
|
|
/* Check that the frame id is valid */
|
|
for (i = 0; i < 5 && work->frame_id[i] != '\0'; i++) {
|
|
if (!g_ascii_isalnum (work->frame_id[i])) {
|
|
GST_DEBUG ("Encountered invalid frame_id");
|
|
return FALSE;
|
|
}
|
|
}
|
|
|
|
/* Can't handle encrypted frames right now */
|
|
if (work->frame_flags & ID3V2_FRAME_FORMAT_ENCRYPTION) {
|
|
GST_WARNING ("Encrypted frames are not supported");
|
|
return FALSE;
|
|
}
|
|
|
|
if (work->frame_flags & ID3V2_FRAME_FORMAT_UNSYNCHRONISATION) {
|
|
GST_WARNING ("ID3v2 frame with unsupported unsynchronisation applied. "
|
|
"May fail badly");
|
|
}
|
|
|
|
tag_name = gst_tag_from_id3_tag (work->frame_id);
|
|
if (tag_name == NULL &&
|
|
strncmp (work->frame_id, "RVA2", 4) != 0 &&
|
|
strncmp (work->frame_id, "TXXX", 4) != 0 &&
|
|
strncmp (work->frame_id, "UFID", 4) != 0) {
|
|
return FALSE;
|
|
}
|
|
|
|
if (work->frame_flags & (ID3V2_FRAME_FORMAT_COMPRESSION |
|
|
ID3V2_FRAME_FORMAT_DATA_LENGTH_INDICATOR)) {
|
|
if (work->hdr.frame_data_size <= 4)
|
|
return FALSE;
|
|
work->parse_size = read_synch_uint (frame_data, 4);
|
|
frame_data += 4;
|
|
frame_data_size -= 4;
|
|
if (work->parse_size < frame_data_size) {
|
|
GST_WARNING ("ID3v2 frame %s has invalid size %d.", tag_name,
|
|
frame_data_size);
|
|
return FALSE;
|
|
}
|
|
}
|
|
|
|
work->parse_size = frame_data_size;
|
|
|
|
if (work->frame_flags & ID3V2_FRAME_FORMAT_COMPRESSION) {
|
|
#ifdef HAVE_ZLIB
|
|
uLongf destSize = work->parse_size;
|
|
Bytef *dest, *src;
|
|
|
|
work->parse_data = g_malloc (work->parse_size);
|
|
g_return_val_if_fail (work->parse_data != NULL, FALSE);
|
|
|
|
dest = (Bytef *) work->parse_data;
|
|
src = (Bytef *) frame_data;
|
|
|
|
if (uncompress (dest, &destSize, src, frame_data_size) != Z_OK) {
|
|
g_free (work->parse_data);
|
|
return FALSE;
|
|
}
|
|
if (destSize != work->parse_size) {
|
|
GST_WARNING
|
|
("Decompressing ID3v2 frame %s did not produce expected size %d bytes (got %lu)",
|
|
tag_name, work->parse_size, destSize);
|
|
return FALSE;
|
|
}
|
|
#else
|
|
GST_WARNING ("Compressed ID3v2 tag frame could not be decompressed"
|
|
" because gstid3demux was compiled without zlib support");
|
|
return FALSE;
|
|
#endif
|
|
} else {
|
|
work->parse_data = frame_data;
|
|
}
|
|
|
|
if (work->frame_id[0] == 'T') {
|
|
if (strcmp (work->frame_id, "TXXX") != 0) {
|
|
/* Text identification frame */
|
|
tag_fields = parse_text_identification_frame (work);
|
|
} else {
|
|
/* Handle user text frame */
|
|
tag_str = parse_user_text_identification_frame (work, &tag_name);
|
|
}
|
|
} else if (!strcmp (work->frame_id, "COMM")) {
|
|
/* Comment */
|
|
result = parse_comment_frame (work);
|
|
} else if (!strcmp (work->frame_id, "APIC")) {
|
|
/* Attached picture */
|
|
result = parse_picture_frame (work);
|
|
} else if (!strcmp (work->frame_id, "RVA2")) {
|
|
/* Relative volume */
|
|
result = parse_relative_volume_adjustment_two (work);
|
|
} else if (!strcmp (work->frame_id, "UFID")) {
|
|
/* Unique file identifier */
|
|
tag_str = parse_unique_file_identifier (work, &tag_name);
|
|
}
|
|
|
|
if (work->frame_flags & ID3V2_FRAME_FORMAT_COMPRESSION)
|
|
g_free (work->parse_data);
|
|
|
|
if (tag_str != NULL) {
|
|
/* g_print ("Tag %s value %s\n", tag_name, tag_str); */
|
|
result = id3v2_tag_to_taglist (work, tag_name, tag_str);
|
|
g_free (tag_str);
|
|
}
|
|
if (tag_fields != NULL) {
|
|
if (strcmp (work->frame_id, "TCON") == 0) {
|
|
/* Genre strings need special treatment */
|
|
result |= id3v2_genre_fields_to_taglist (work, tag_name, tag_fields);
|
|
} else {
|
|
gint t;
|
|
|
|
for (t = 0; t < tag_fields->len; t++) {
|
|
tag_str = g_array_index (tag_fields, gchar *, t);
|
|
if (tag_str != NULL && tag_str[0] != '\0')
|
|
result |= id3v2_tag_to_taglist (work, tag_name, tag_str);
|
|
}
|
|
}
|
|
free_tag_strings (tag_fields);
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
static gboolean
|
|
parse_comment_frame (ID3TagsWorking * work)
|
|
{
|
|
guint dummy;
|
|
guint8 encoding;
|
|
gchar language[4];
|
|
GArray *fields = NULL;
|
|
gchar *description, *text;
|
|
|
|
if (work->parse_size < 6)
|
|
return FALSE;
|
|
|
|
encoding = work->parse_data[0];
|
|
language[0] = work->parse_data[1];
|
|
language[1] = work->parse_data[2];
|
|
language[2] = work->parse_data[3];
|
|
language[3] = 0;
|
|
|
|
parse_split_strings (encoding, (gchar *) work->parse_data + 4,
|
|
work->parse_size - 4, &fields);
|
|
|
|
if (fields == NULL || fields->len < 2) {
|
|
GST_WARNING ("Failed to decode comment frame");
|
|
goto fail;
|
|
}
|
|
description = g_array_index (fields, gchar *, 0);
|
|
text = g_array_index (fields, gchar *, 1);
|
|
|
|
if (!g_utf8_validate (text, -1, NULL)) {
|
|
GST_WARNING ("Converted string is not valid utf-8");
|
|
goto fail;
|
|
}
|
|
|
|
/* skip our own dummy descriptions (from id3v2mux) */
|
|
if (strlen (description) > 0 && g_utf8_validate (description, -1, NULL) &&
|
|
sscanf (description, "c%u", &dummy) != 1) {
|
|
gchar *s;
|
|
|
|
if (language[0] != '\0') {
|
|
s = g_strdup_printf ("%s[%s]=%s", description, language, text);
|
|
} else {
|
|
s = g_strdup_printf ("%s=%s", description, text);
|
|
}
|
|
gst_tag_list_add (work->tags, GST_TAG_MERGE_APPEND,
|
|
GST_TAG_EXTENDED_COMMENT, s, NULL);
|
|
g_free (s);
|
|
} else if (text != NULL && *text != '\0') {
|
|
gst_tag_list_add (work->tags, GST_TAG_MERGE_APPEND,
|
|
GST_TAG_COMMENT, text, NULL);
|
|
} else {
|
|
goto fail;
|
|
}
|
|
|
|
free_tag_strings (fields);
|
|
return TRUE;
|
|
|
|
fail:
|
|
{
|
|
GST_WARNING ("failed to parse COMM frame");
|
|
free_tag_strings (fields);
|
|
return FALSE;
|
|
}
|
|
}
|
|
|
|
static GArray *
|
|
parse_text_identification_frame (ID3TagsWorking * work)
|
|
{
|
|
guchar encoding;
|
|
GArray *fields = NULL;
|
|
|
|
if (work->parse_size < 2)
|
|
return NULL;
|
|
|
|
encoding = work->parse_data[0];
|
|
parse_split_strings (encoding, (gchar *) work->parse_data + 1,
|
|
work->parse_size - 1, &fields);
|
|
if (fields) {
|
|
if (fields->len > 0) {
|
|
GST_LOG ("Read %d fields from Text ID frame of size %d with encoding %d"
|
|
". First is '%s'", fields->len, work->parse_size - 1, encoding,
|
|
g_array_index (fields, gchar *, 0));
|
|
} else {
|
|
GST_LOG ("Read 0 fields from Text ID frame of size %d with encoding %d",
|
|
work->parse_size - 1, encoding);
|
|
}
|
|
}
|
|
|
|
return fields;
|
|
}
|
|
|
|
static gchar *
|
|
parse_user_text_identification_frame (ID3TagsWorking * work,
|
|
const gchar ** tag_name)
|
|
{
|
|
gchar *ret;
|
|
guchar encoding;
|
|
GArray *fields = NULL;
|
|
|
|
*tag_name = NULL;
|
|
|
|
if (work->parse_size < 2)
|
|
return NULL;
|
|
|
|
encoding = work->parse_data[0];
|
|
|
|
parse_split_strings (encoding, (gchar *) work->parse_data + 1,
|
|
work->parse_size - 1, &fields);
|
|
|
|
if (fields == NULL)
|
|
return NULL;
|
|
|
|
if (fields->len != 2) {
|
|
GST_WARNING ("Expected 2 fields in TXXX frame, but got %d", fields->len);
|
|
free_tag_strings (fields);
|
|
return NULL;
|
|
}
|
|
|
|
*tag_name =
|
|
gst_tag_from_id3_user_tag ("TXXX", g_array_index (fields, gchar *, 0));
|
|
|
|
GST_LOG ("TXXX frame of size %d. Mapped descriptor '%s' to GStreamer tag %s",
|
|
work->parse_size - 1, g_array_index (fields, gchar *, 0),
|
|
GST_STR_NULL (*tag_name));
|
|
|
|
if (*tag_name) {
|
|
ret = g_strdup (g_array_index (fields, gchar *, 1));
|
|
/* GST_LOG ("%s = %s", *tag_name, GST_STR_NULL (ret)); */
|
|
} else {
|
|
ret = NULL;
|
|
}
|
|
|
|
free_tag_strings (fields);
|
|
return ret;
|
|
}
|
|
|
|
static gboolean
|
|
parse_id_string (ID3TagsWorking * work, gchar ** p_str, gint * p_len,
|
|
gint * p_datalen)
|
|
{
|
|
gint len, datalen;
|
|
|
|
if (work->parse_size < 2)
|
|
return FALSE;
|
|
|
|
for (len = 0; len < work->parse_size - 1; ++len) {
|
|
if (work->parse_data[len] == '\0')
|
|
break;
|
|
}
|
|
|
|
datalen = work->parse_size - (len + 1);
|
|
if (len == 0 || datalen <= 0)
|
|
return FALSE;
|
|
|
|
*p_str = g_strndup ((gchar *) work->parse_data, len);
|
|
*p_len = len;
|
|
*p_datalen = datalen;
|
|
|
|
return TRUE;
|
|
}
|
|
|
|
static gchar *
|
|
parse_unique_file_identifier (ID3TagsWorking * work, const gchar ** tag_name)
|
|
{
|
|
gint len, datalen;
|
|
gchar *owner_id, *data, *ret = NULL;
|
|
|
|
GST_LOG ("parsing UFID frame of size %d", work->parse_size);
|
|
|
|
if (!parse_id_string (work, &owner_id, &len, &datalen))
|
|
return NULL;
|
|
|
|
data = (gchar *) work->parse_data + len + 1;
|
|
GST_LOG ("UFID owner ID: %s (+ %d bytes of data)", owner_id, datalen);
|
|
|
|
if (strcmp (owner_id, "http://musicbrainz.org") == 0 &&
|
|
g_utf8_validate (data, datalen, NULL)) {
|
|
*tag_name = GST_TAG_MUSICBRAINZ_TRACKID;
|
|
ret = g_strndup (data, datalen);
|
|
} else {
|
|
GST_INFO ("Unknown UFID owner ID: %s", owner_id);
|
|
}
|
|
g_free (owner_id);
|
|
|
|
return ret;
|
|
}
|
|
|
|
/* parse data and return length of the next string in the given encoding,
|
|
* including the NUL terminator */
|
|
static gint
|
|
scan_encoded_string (guint8 encoding, gchar * data, gint data_size)
|
|
{
|
|
gint i;
|
|
|
|
switch (encoding) {
|
|
case ID3V2_ENCODING_ISO8859:
|
|
case ID3V2_ENCODING_UTF8:
|
|
for (i = 0; i < data_size; ++i) {
|
|
if (data[i] == '\0')
|
|
return i + 1;
|
|
}
|
|
break;
|
|
case ID3V2_ENCODING_UTF16:
|
|
case ID3V2_ENCODING_UTF16BE:
|
|
/* we don't care about BOMs here and treat them as part of the string */
|
|
/* Find '\0\0' terminator */
|
|
for (i = 0; i < data_size - 1; i += 2) {
|
|
if (data[i] == '\0' && data[i + 1] == '\0')
|
|
return i + 2;
|
|
}
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
|
|
return 0;
|
|
}
|
|
|
|
static gboolean
|
|
parse_picture_frame (ID3TagsWorking * work)
|
|
{
|
|
GstBuffer *image = NULL;
|
|
GstCaps *image_caps = NULL;
|
|
gboolean is_pic_uri = FALSE;
|
|
guint8 txt_encoding, pic_type;
|
|
gchar *mime_str = NULL;
|
|
gint len, datalen;
|
|
|
|
GST_LOG ("APIC frame");
|
|
|
|
if (work->parse_size < 1 + 1 + 1 + 1 + 1)
|
|
goto not_enough_data;
|
|
|
|
txt_encoding = work->parse_data[0];
|
|
++work->parse_data;
|
|
--work->parse_size;
|
|
|
|
if (!parse_id_string (work, &mime_str, &len, &datalen)) {
|
|
return FALSE;
|
|
}
|
|
|
|
is_pic_uri = (mime_str != NULL && strcmp (mime_str, "-->") == 0);
|
|
|
|
if (mime_str && *mime_str && strchr (mime_str, '/') == NULL && !is_pic_uri) {
|
|
gchar *tmp;
|
|
|
|
tmp = g_strdup_printf ("image/%s", mime_str);
|
|
g_free (mime_str);
|
|
mime_str = tmp;
|
|
}
|
|
|
|
if (work->parse_size < (len + 1) + 1 + 1 + 1)
|
|
goto not_enough_data;
|
|
|
|
work->parse_data += (len + 1);
|
|
work->parse_size -= (len + 1);
|
|
|
|
pic_type = work->parse_data[0];
|
|
++work->parse_data;
|
|
--work->parse_size;
|
|
|
|
GST_LOG ("APIC frame mime type : %s", GST_STR_NULL (mime_str));
|
|
GST_LOG ("APIC frame picture type : 0x%02x", (guint) pic_type);
|
|
|
|
if (work->parse_size < 1 + 1)
|
|
goto not_enough_data;
|
|
|
|
len = scan_encoded_string (txt_encoding, (gchar *) work->parse_data,
|
|
work->parse_size);
|
|
|
|
if (len < 1)
|
|
goto error;
|
|
|
|
/* just skip the description string ... */
|
|
GST_LOG ("Skipping description string (%d bytes in original coding)", len);
|
|
|
|
if (work->parse_size < len + 1)
|
|
goto not_enough_data;
|
|
|
|
work->parse_data += len;
|
|
work->parse_size -= len;
|
|
|
|
GST_DEBUG ("image data is %u bytes", work->parse_size);
|
|
|
|
if (work->parse_size <= 0)
|
|
goto not_enough_data;
|
|
|
|
if (is_pic_uri) {
|
|
gchar *uri;
|
|
|
|
uri = g_strndup ((gchar *) work->parse_data, work->parse_size);
|
|
GST_DEBUG ("image URI: %s", uri);
|
|
|
|
image = gst_buffer_new ();
|
|
GST_BUFFER_MALLOCDATA (image) = (guint8 *) uri; /* take ownership */
|
|
GST_BUFFER_DATA (image) = (guint8 *) uri;
|
|
GST_BUFFER_SIZE (image) = work->parse_size;
|
|
|
|
image_caps = gst_caps_new_simple ("text/uri-list", NULL);
|
|
} else {
|
|
image = gst_buffer_new_and_alloc (work->parse_size);
|
|
memcpy (GST_BUFFER_DATA (image), work->parse_data, work->parse_size);
|
|
|
|
/* if possible use GStreamer media type rather than declared type */
|
|
image_caps = gst_type_find_helper_for_buffer (NULL, image, NULL);
|
|
if (image_caps) {
|
|
GST_DEBUG ("Found GStreamer media type: %" GST_PTR_FORMAT, image_caps);
|
|
} else if (mime_str && *mime_str) {
|
|
GST_DEBUG ("No GStreamer media type found, using declared type");
|
|
image_caps = gst_caps_new_simple (mime_str, NULL);
|
|
} else {
|
|
GST_DEBUG ("Empty declared mime type, ignoring image frame");
|
|
image = NULL;
|
|
image_caps = NULL;
|
|
goto error;
|
|
}
|
|
}
|
|
|
|
if (image && image_caps) {
|
|
/* FIXME: remove #ifdef once we depend on -base >= 0.10.9 */
|
|
#ifdef GST_TYPE_TAG_IMAGE_TYPE
|
|
if (pic_type > 0x14)
|
|
pic_type = GST_TAG_IMAGE_TYPE_UNDEFINED;
|
|
gst_structure_set (gst_caps_get_structure (image_caps, 0),
|
|
"image-type", GST_TYPE_TAG_IMAGE_TYPE,
|
|
(GstTagImageType) pic_type, NULL);
|
|
#endif
|
|
|
|
gst_buffer_set_caps (image, image_caps);
|
|
gst_caps_unref (image_caps);
|
|
if (pic_type == 0x01 || pic_type == 0x02) {
|
|
/* file icon of some sort */
|
|
gst_tag_list_add (work->tags, GST_TAG_MERGE_APPEND,
|
|
GST_TAG_PREVIEW_IMAGE, image, NULL);
|
|
} else {
|
|
gst_tag_list_add (work->tags, GST_TAG_MERGE_APPEND,
|
|
GST_TAG_IMAGE, image, NULL);
|
|
}
|
|
gst_buffer_unref (image);
|
|
}
|
|
|
|
g_free (mime_str);
|
|
return TRUE;
|
|
|
|
not_enough_data:
|
|
{
|
|
GST_DEBUG ("not enough data, skipping APIC frame");
|
|
/* fall through to error */
|
|
}
|
|
error:
|
|
{
|
|
GST_DEBUG ("problem parsing APIC frame, skipping");
|
|
g_free (mime_str);
|
|
return FALSE;
|
|
}
|
|
}
|
|
|
|
#define ID3V2_RVA2_CHANNEL_MASTER 1
|
|
|
|
static gboolean
|
|
parse_relative_volume_adjustment_two (ID3TagsWorking * work)
|
|
{
|
|
const gchar *gain_tag_name = NULL;
|
|
const gchar *peak_tag_name = NULL;
|
|
gdouble gain_dB, peak_val;
|
|
guint64 peak;
|
|
guint8 *data, chan, peak_bits;
|
|
gchar *id;
|
|
gint len, datalen, i;
|
|
|
|
if (!parse_id_string (work, &id, &len, &datalen))
|
|
return FALSE;
|
|
|
|
if (datalen < (1 + 2 + 1)) {
|
|
GST_WARNING ("broken RVA2 frame, data size only %d bytes", datalen);
|
|
g_free (id);
|
|
return FALSE;
|
|
}
|
|
|
|
data = work->parse_data + len + 1;
|
|
chan = GST_READ_UINT8 (data);
|
|
gain_dB = (gdouble) ((gint16) GST_READ_UINT16_BE (data + 1)) / 512.0;
|
|
/* The meaning of the peak value is not defined in the ID3v2 spec. However,
|
|
* the first/only implementation of this seems to have been in XMMS, and
|
|
* other libs (like mutagen) seem to follow that implementation as well:
|
|
* see http://bugs.xmms.org/attachment.cgi?id=113&action=view */
|
|
peak_bits = GST_READ_UINT8 (data + 1 + 2);
|
|
if (peak_bits > 64) {
|
|
GST_WARNING ("silly peak precision of %d bits, ignoring", (gint) peak_bits);
|
|
peak_bits = 0;
|
|
}
|
|
data += 1 + 2 + 1;
|
|
datalen -= 1 + 2 + 1;
|
|
if (peak_bits == 16) {
|
|
peak = GST_READ_UINT16_BE (data);
|
|
} else {
|
|
peak = 0;
|
|
for (i = 0; i < (GST_ROUND_UP_8 (peak_bits) / 8) && datalen > 0; ++i) {
|
|
peak = peak << 8;
|
|
peak |= GST_READ_UINT8 (data);
|
|
++data;
|
|
--datalen;
|
|
}
|
|
}
|
|
|
|
peak = peak << (64 - GST_ROUND_UP_8 (peak_bits));
|
|
peak_val =
|
|
gst_guint64_to_gdouble (peak) / gst_util_guint64_to_gdouble (G_MAXINT64);
|
|
GST_LOG ("RVA2 frame: id=%s, chan=%u, adj=%.2fdB, peak_bits=%u, peak=%.2f",
|
|
id, chan, gain_dB, (guint) peak_bits, peak_val);
|
|
|
|
if (chan == ID3V2_RVA2_CHANNEL_MASTER && strcmp (id, "track") == 0) {
|
|
gain_tag_name = GST_TAG_TRACK_GAIN;
|
|
peak_tag_name = GST_TAG_TRACK_PEAK;
|
|
} else if (chan == ID3V2_RVA2_CHANNEL_MASTER && strcmp (id, "album") == 0) {
|
|
gain_tag_name = GST_TAG_ALBUM_GAIN;
|
|
peak_tag_name = GST_TAG_ALBUM_PEAK;
|
|
} else {
|
|
GST_INFO ("Unhandled RVA2 frame id '%s' for channel %d", id, chan);
|
|
}
|
|
|
|
if (gain_tag_name) {
|
|
gst_tag_list_add (work->tags, GST_TAG_MERGE_APPEND,
|
|
gain_tag_name, gain_dB, NULL);
|
|
}
|
|
if (peak_tag_name && peak_bits > 0) {
|
|
gst_tag_list_add (work->tags, GST_TAG_MERGE_APPEND,
|
|
peak_tag_name, peak_val, NULL);
|
|
}
|
|
|
|
g_free (id);
|
|
|
|
return (gain_tag_name != NULL || peak_tag_name != NULL);
|
|
}
|
|
|
|
static gboolean
|
|
id3v2_tag_to_taglist (ID3TagsWorking * work, const gchar * tag_name,
|
|
const gchar * tag_str)
|
|
{
|
|
GType tag_type = gst_tag_get_type (tag_name);
|
|
GstTagList *tag_list = work->tags;
|
|
|
|
if (tag_str == NULL)
|
|
return FALSE;
|
|
|
|
switch (tag_type) {
|
|
case G_TYPE_UINT:
|
|
{
|
|
gint current, total;
|
|
|
|
if (sscanf (tag_str, "%d/%d", ¤t, &total) == 2) {
|
|
if (total <= 0) {
|
|
GST_WARNING ("Ignoring invalid value for total %d in tag %s",
|
|
total, tag_name);
|
|
} else {
|
|
if (strcmp (tag_name, GST_TAG_TRACK_NUMBER) == 0) {
|
|
gst_tag_list_add (tag_list, GST_TAG_MERGE_APPEND,
|
|
GST_TAG_TRACK_COUNT, total, NULL);
|
|
} else if (strcmp (tag_name, GST_TAG_ALBUM_VOLUME_NUMBER) == 0) {
|
|
gst_tag_list_add (tag_list, GST_TAG_MERGE_APPEND,
|
|
GST_TAG_ALBUM_VOLUME_COUNT, total, NULL);
|
|
}
|
|
}
|
|
} else if (sscanf (tag_str, "%d", ¤t) != 1) {
|
|
/* Not an integer in the string */
|
|
GST_WARNING ("Tag string for tag %s does not contain an integer - "
|
|
"ignoring", tag_name);
|
|
break;
|
|
}
|
|
|
|
if (current <= 0) {
|
|
GST_WARNING ("Ignoring invalid value %d in tag %s", current, tag_name);
|
|
} else {
|
|
gst_tag_list_add (tag_list, GST_TAG_MERGE_APPEND, tag_name, current,
|
|
NULL);
|
|
}
|
|
break;
|
|
}
|
|
case G_TYPE_UINT64:
|
|
{
|
|
guint64 tmp;
|
|
|
|
g_assert (strcmp (tag_name, GST_TAG_DURATION) == 0);
|
|
tmp = strtoul (tag_str, NULL, 10);
|
|
if (tmp == 0) {
|
|
break;
|
|
}
|
|
gst_tag_list_add (tag_list, GST_TAG_MERGE_APPEND,
|
|
GST_TAG_DURATION, tmp * 1000 * 1000, NULL);
|
|
break;
|
|
}
|
|
case G_TYPE_STRING:{
|
|
if (!strcmp (tag_name, GST_TAG_GENRE)) {
|
|
if (work->prev_genre && !strcmp (tag_str, work->prev_genre))
|
|
break; /* Same as the last genre */
|
|
g_free (work->prev_genre);
|
|
work->prev_genre = g_strdup (tag_str);
|
|
}
|
|
gst_tag_list_add (tag_list, GST_TAG_MERGE_APPEND,
|
|
tag_name, tag_str, NULL);
|
|
break;
|
|
}
|
|
|
|
default:{
|
|
gchar *tmp = NULL;
|
|
GValue src = { 0, };
|
|
GValue dest = { 0, };
|
|
|
|
/* Ensure that any date string is complete */
|
|
if (tag_type == GST_TYPE_DATE) {
|
|
guint year = 1901, month = 1, day = 1;
|
|
|
|
/* Dates can be yyyy-MM-dd, yyyy-MM or yyyy, but we need
|
|
* the first type */
|
|
if (sscanf (tag_str, "%04u-%02u-%02u", &year, &month, &day) == 0)
|
|
break;
|
|
|
|
tmp = g_strdup_printf ("%04u-%02u-%02u", year, month, day);
|
|
tag_str = tmp;
|
|
}
|
|
|
|
/* handles anything else */
|
|
g_value_init (&src, G_TYPE_STRING);
|
|
g_value_set_string (&src, (const gchar *) tag_str);
|
|
g_value_init (&dest, tag_type);
|
|
|
|
if (g_value_transform (&src, &dest)) {
|
|
gst_tag_list_add_values (tag_list, GST_TAG_MERGE_APPEND,
|
|
tag_name, &dest, NULL);
|
|
} else if (tag_type == G_TYPE_DOUBLE) {
|
|
/* replaygain tags in TXXX frames ... */
|
|
g_value_set_double (&dest, g_strtod (tag_str, NULL));
|
|
gst_tag_list_add_values (tag_list, GST_TAG_MERGE_KEEP,
|
|
tag_name, &dest, NULL);
|
|
GST_LOG ("Converted string '%s' to double %f", tag_str,
|
|
g_value_get_double (&dest));
|
|
} else {
|
|
GST_WARNING ("Failed to transform tag from string to type '%s'",
|
|
g_type_name (tag_type));
|
|
}
|
|
|
|
g_value_unset (&src);
|
|
g_value_unset (&dest);
|
|
g_free (tmp);
|
|
break;
|
|
}
|
|
}
|
|
|
|
return TRUE;
|
|
}
|
|
|
|
/* Check that an array of characters contains only digits */
|
|
static gboolean
|
|
id3v2_are_digits (const gchar * chars, gint size)
|
|
{
|
|
gint i;
|
|
|
|
for (i = 0; i < size; i++) {
|
|
if (!g_ascii_isdigit (chars[i]))
|
|
return FALSE;
|
|
}
|
|
return TRUE;
|
|
}
|
|
|
|
static gboolean
|
|
id3v2_genre_string_to_taglist (ID3TagsWorking * work, const gchar * tag_name,
|
|
const gchar * tag_str, gint len)
|
|
{
|
|
g_return_val_if_fail (tag_str != NULL, FALSE);
|
|
|
|
/* If it's a number, it might be a defined genre */
|
|
if (id3v2_are_digits (tag_str, len)) {
|
|
tag_str = gst_tag_id3_genre_get (strtol (tag_str, NULL, 10));
|
|
return id3v2_tag_to_taglist (work, tag_name, tag_str);
|
|
}
|
|
/* Otherwise it might be "RX" or "CR" */
|
|
if (len == 2) {
|
|
if (g_ascii_strncasecmp ("rx", tag_str, len) == 0)
|
|
return id3v2_tag_to_taglist (work, tag_name, "Remix");
|
|
|
|
if (g_ascii_strncasecmp ("cr", tag_str, len) == 0)
|
|
return id3v2_tag_to_taglist (work, tag_name, "Cover");
|
|
}
|
|
|
|
/* Otherwise it's a string */
|
|
return id3v2_tag_to_taglist (work, tag_name, tag_str);
|
|
}
|
|
|
|
static gboolean
|
|
id3v2_genre_fields_to_taglist (ID3TagsWorking * work, const gchar * tag_name,
|
|
GArray * tag_fields)
|
|
{
|
|
gchar *tag_str = NULL;
|
|
gboolean result = FALSE;
|
|
gint i;
|
|
|
|
for (i = 0; i < tag_fields->len; i++) {
|
|
gint len;
|
|
|
|
tag_str = g_array_index (tag_fields, gchar *, 0);
|
|
if (tag_str == NULL)
|
|
continue;
|
|
|
|
len = strlen (tag_str);
|
|
/* Only supposed to see '(n)' type numeric genre strings in ID3 <= 2.3.0
|
|
* but apparently we see them in 2.4.0 sometimes too */
|
|
if (TRUE || work->hdr.version <= 0x300) { /* <= 2.3.0 */
|
|
/* Check for genre numbers wrapped in parentheses, possibly
|
|
* followed by a string */
|
|
while (len >= 2) {
|
|
gint pos;
|
|
gboolean found = FALSE;
|
|
|
|
/* Double parenthesis ends the numeric genres, but we need
|
|
* to swallow the first one so we actually output '(' */
|
|
if (tag_str[0] == '(' && tag_str[1] == '(') {
|
|
tag_str++;
|
|
len--;
|
|
break;
|
|
}
|
|
|
|
/* If the first char is not a parenthesis, then stop
|
|
* looking for parenthesised genre strings */
|
|
if (tag_str[0] != '(')
|
|
break;
|
|
|
|
for (pos = 1; pos < len; pos++) {
|
|
if (tag_str[pos] == ')') {
|
|
gchar *tmp_str;
|
|
|
|
tmp_str = g_strndup (tag_str + 1, pos - 1);
|
|
result |=
|
|
id3v2_genre_string_to_taglist (work, tag_name, tmp_str,
|
|
pos - 1);
|
|
g_free (tmp_str);
|
|
tag_str += pos + 1;
|
|
len -= pos + 1;
|
|
found = TRUE;
|
|
break;
|
|
}
|
|
|
|
/* If we encounter a non-digit while searching for a closing
|
|
* parenthesis, we should not try and interpret this as a
|
|
* numeric genre string */
|
|
if (!g_ascii_isdigit (tag_str[pos]))
|
|
break;
|
|
}
|
|
if (!found)
|
|
break; /* There was no closing parenthesis */
|
|
}
|
|
}
|
|
|
|
if (len > 0 && tag_str != NULL)
|
|
result |= id3v2_genre_string_to_taglist (work, tag_name, tag_str, len);
|
|
}
|
|
return result;
|
|
}
|
|
|
|
static const gchar utf16enc[] = "UTF-16";
|
|
static const gchar utf16leenc[] = "UTF-16LE";
|
|
static const gchar utf16beenc[] = "UTF-16BE";
|
|
|
|
static gboolean
|
|
find_utf16_bom (gchar * data, const gchar ** p_in_encoding)
|
|
{
|
|
guint16 marker = (GST_READ_UINT8 (data) << 8) | GST_READ_UINT8 (data + 1);
|
|
|
|
switch (marker) {
|
|
case 0xFFFE:
|
|
*p_in_encoding = utf16leenc;
|
|
return TRUE;
|
|
case 0xFEFF:
|
|
*p_in_encoding = utf16beenc;
|
|
return TRUE;
|
|
default:
|
|
break;
|
|
}
|
|
return FALSE;
|
|
}
|
|
|
|
static void
|
|
parse_insert_string_field (guint8 encoding, gchar * data, gint data_size,
|
|
GArray * fields)
|
|
{
|
|
gchar *field = NULL;
|
|
|
|
switch (encoding) {
|
|
case ID3V2_ENCODING_UTF16:
|
|
case ID3V2_ENCODING_UTF16BE:
|
|
{
|
|
const gchar *in_encode;
|
|
|
|
if (encoding == ID3V2_ENCODING_UTF16)
|
|
in_encode = utf16enc;
|
|
else
|
|
in_encode = utf16beenc;
|
|
|
|
/* Sometimes we see strings with multiple BOM markers at the start.
|
|
* In that case, we assume the innermost one is correct. If that fails
|
|
* to produce valid UTF-8, we try the other endianness anyway */
|
|
while (data_size > 2 && find_utf16_bom (data, &in_encode)) {
|
|
data += 2; /* skip BOM */
|
|
data_size -= 2;
|
|
}
|
|
|
|
field = g_convert (data, data_size, "UTF-8", in_encode, NULL, NULL, NULL);
|
|
|
|
if (field == NULL || g_utf8_validate (field, -1, NULL) == FALSE) {
|
|
/* As a fallback, try interpreting UTF-16 in the other endianness */
|
|
if (in_encode == utf16beenc)
|
|
field = g_convert (data, data_size, "UTF-8", utf16leenc,
|
|
NULL, NULL, NULL);
|
|
}
|
|
}
|
|
|
|
break;
|
|
case ID3V2_ENCODING_ISO8859:
|
|
if (g_utf8_validate (data, data_size, NULL))
|
|
field = g_strndup (data, data_size);
|
|
else
|
|
field = g_convert (data, data_size, "UTF-8", "ISO-8859-1",
|
|
NULL, NULL, NULL);
|
|
break;
|
|
default:
|
|
field = g_strndup (data, data_size);
|
|
break;
|
|
}
|
|
|
|
if (field) {
|
|
if (g_utf8_validate (field, -1, NULL)) {
|
|
g_array_append_val (fields, field);
|
|
return;
|
|
}
|
|
|
|
GST_DEBUG ("%s was bad UTF-8 after conversion from encoding %d. Ignoring",
|
|
field, encoding);
|
|
g_free (field);
|
|
}
|
|
}
|
|
|
|
static void
|
|
parse_split_strings (guint8 encoding, gchar * data, gint data_size,
|
|
GArray ** out_fields)
|
|
{
|
|
GArray *fields = g_array_new (FALSE, TRUE, sizeof (gchar *));
|
|
gint text_pos;
|
|
gint prev = 0;
|
|
|
|
g_return_if_fail (out_fields != NULL);
|
|
|
|
switch (encoding) {
|
|
case ID3V2_ENCODING_ISO8859:
|
|
for (text_pos = 0; text_pos < data_size; text_pos++) {
|
|
if (data[text_pos] == 0) {
|
|
parse_insert_string_field (encoding, data + prev,
|
|
text_pos - prev + 1, fields);
|
|
prev = text_pos + 1;
|
|
}
|
|
}
|
|
if (data_size - prev > 0 && data[prev] != 0x00) {
|
|
parse_insert_string_field (encoding, data + prev,
|
|
data_size - prev, fields);
|
|
}
|
|
|
|
break;
|
|
case ID3V2_ENCODING_UTF8:
|
|
for (prev = 0, text_pos = 0; text_pos < data_size; text_pos++) {
|
|
if (data[text_pos] == '\0') {
|
|
parse_insert_string_field (encoding, data + prev,
|
|
text_pos - prev + 1, fields);
|
|
prev = text_pos + 1;
|
|
}
|
|
}
|
|
if (data_size - prev > 0 && data[prev] != 0x00) {
|
|
parse_insert_string_field (encoding, data + prev,
|
|
data_size - prev, fields);
|
|
}
|
|
break;
|
|
case ID3V2_ENCODING_UTF16:
|
|
case ID3V2_ENCODING_UTF16BE:
|
|
{
|
|
/* Find '\0\0' terminator */
|
|
for (text_pos = 0; text_pos < data_size - 1; text_pos += 2) {
|
|
if (data[text_pos] == '\0' && data[text_pos + 1] == '\0') {
|
|
/* found a delimiter */
|
|
parse_insert_string_field (encoding, data + prev,
|
|
text_pos - prev + 2, fields);
|
|
text_pos++; /* Advance to the 2nd NULL terminator */
|
|
prev = text_pos + 1;
|
|
break;
|
|
}
|
|
}
|
|
if (data_size - prev > 1 &&
|
|
(data[prev] != 0x00 || data[prev + 1] != 0x00)) {
|
|
/* There were 2 or more non-null chars left, convert those too */
|
|
parse_insert_string_field (encoding, data + prev,
|
|
data_size - prev, fields);
|
|
}
|
|
break;
|
|
}
|
|
}
|
|
if (fields->len > 0)
|
|
*out_fields = fields;
|
|
else
|
|
g_array_free (fields, TRUE);
|
|
}
|
|
|
|
static void
|
|
free_tag_strings (GArray * fields)
|
|
{
|
|
if (fields) {
|
|
gint i;
|
|
gchar *c;
|
|
|
|
for (i = 0; i < fields->len; i++) {
|
|
c = g_array_index (fields, gchar *, i);
|
|
g_free (c);
|
|
}
|
|
g_array_free (fields, TRUE);
|
|
}
|
|
}
|