diff options
author | Max Kellermann <max@duempel.org> | 2009-08-04 00:15:43 +0200 |
---|---|---|
committer | Max Kellermann <max@duempel.org> | 2009-08-04 00:15:43 +0200 |
commit | 09571fcc553127822b3f3967c2593fe72ae6b21b (patch) | |
tree | 04d6a948d98257e4ae4055fd2a08bf70ab89138d | |
parent | c7563a5783c951b75c40123be3bbfdc08794864e (diff) | |
download | mpd-09571fcc553127822b3f3967c2593fe72ae6b21b.tar.gz mpd-09571fcc553127822b3f3967c2593fe72ae6b21b.tar.xz mpd-09571fcc553127822b3f3967c2593fe72ae6b21b.zip |
tag_id3: splitted function tag_id3_import_frame()
Splitted tag_id3_import_frame() into two specialized functions:
tag_id3_import_text() and tag_id3_import_comment(). Use
id3_frame_field() instead of directly accessing id3_frame.fields.
-rw-r--r-- | src/tag_id3.c | 210 |
1 files changed, 91 insertions, 119 deletions
diff --git a/src/tag_id3.c b/src/tag_id3.c index 84e980ec3..66d37be9e 100644 --- a/src/tag_id3.c +++ b/src/tag_id3.c @@ -119,9 +119,16 @@ import_id3_string(bool is_id3v1, const id3_ucs4_t *ucs4, enum tag_type type) return utf8_stripped; } +/** + * Import a "Text information frame" (ID3v2.4.0 section 4.2). It + * contains 2 fields: + * + * - encoding + * - string list + */ static void -tag_id3_import_frame(struct tag *dest, struct id3_tag *tag, const char *id, - enum tag_type type) +tag_id3_import_text(struct tag *dest, struct id3_tag *tag, const char *id, + enum tag_type type) { struct id3_frame const *frame; id3_ucs4_t const *ucs4; @@ -130,110 +137,75 @@ tag_id3_import_frame(struct tag *dest, struct id3_tag *tag, const char *id, unsigned int nstrings, i; frame = id3_tag_findframe(tag, id, 0); - /* Check frame */ - if (!frame) - { + if (frame == NULL || frame->nfields != 2) return; - } - /* Check fields in frame */ - if(frame->nfields == 0) - { - g_debug("Frame has no fields"); + + /* check the encoding field */ + + field = id3_frame_field(frame, 0); + if (field == NULL || field->type != ID3_FIELD_TYPE_TEXTENCODING) return; - } - /* Starting with T is a stringlist */ - if (id[0] == 'T') - { - /* This one contains 2 fields: - * 1st: Text encoding - * 2: Stringlist - * Shamefully this isn't the RL case. - * But I am going to enforce it anyway. - */ - if(frame->nfields != 2) - { - g_debug("Invalid number '%i' of fields for TXX frame", - frame->nfields); - return; - } - field = &frame->fields[0]; - /** - * First field is encoding field. - * This is ignored by mpd. - */ - if(field->type != ID3_FIELD_TYPE_TEXTENCODING) - { - g_debug("Expected encoding, found: %i", - field->type); - } - /* Process remaining fields, should be only one */ - field = &frame->fields[1]; - /* Encoding field */ - if(field->type == ID3_FIELD_TYPE_STRINGLIST) { - /* Get the number of strings available */ - nstrings = id3_field_getnstrings(field); - for (i = 0; i < nstrings; i++) { - ucs4 = id3_field_getstrings(field,i); - if(!ucs4) - continue; - utf8 = import_id3_string(tag_is_id3v1(tag), - ucs4, type); - if(!utf8) - continue; - - tag_add_item(dest, type, (char *)utf8); - g_free(utf8); - } - } - else { - g_warning("Field type not processed: %i", - (int)id3_field_gettextencoding(field)); - } - } - /* A comment frame */ - else if(!strcmp(ID3_FRAME_COMMENT, id)) - { - /* A comment frame is different... */ - /* 1st: encoding - * 2nd: Language - * 3rd: String - * 4th: FullString. - * The 'value' we want is in the 4th field - */ - if(frame->nfields == 4) - { - /* for now I only read the 4th field, with the fullstring */ - field = &frame->fields[3]; - if(field->type == ID3_FIELD_TYPE_STRINGFULL) - { - ucs4 = id3_field_getfullstring(field); - if(ucs4) - { - utf8 = import_id3_string(tag_is_id3v1(tag), - ucs4, type); - if(utf8) - { - tag_add_item(dest, type, (char *)utf8); - g_free(utf8); - } - } - } - else - { - g_debug("4th field in comment frame differs from expected, got '%i': ignoring", - field->type); - } - } - else - { - g_debug("Invalid 'comments' tag, got '%i' fields instead of 4", - frame->nfields); - } + /* process the value(s) */ + + field = id3_frame_field(frame, 1); + if (field == NULL || field->type != ID3_FIELD_TYPE_STRINGLIST) + return; + + /* Get the number of strings available */ + nstrings = id3_field_getnstrings(field); + for (i = 0; i < nstrings; i++) { + ucs4 = id3_field_getstrings(field, i); + if (ucs4 == NULL) + continue; + + utf8 = import_id3_string(tag_is_id3v1(tag), + ucs4, type); + if (utf8 == NULL) + continue; + + tag_add_item(dest, type, (char *)utf8); + g_free(utf8); } - /* Unsupported */ - else - g_debug("Unsupported tag type requrested"); +} + +/** + * Import a "Comment frame" (ID3v2.4.0 section 4.10). It + * contains 4 fields: + * + * - encoding + * - language + * - string + * - full string (we use this one) + */ +static void +tag_id3_import_comment(struct tag *dest, struct id3_tag *tag, const char *id, + enum tag_type type) +{ + struct id3_frame const *frame; + id3_ucs4_t const *ucs4; + id3_utf8_t *utf8; + union id3_field const *field; + + frame = id3_tag_findframe(tag, id, 0); + if (frame == NULL || frame->nfields != 4) + return; + + /* for now I only read the 4th field, with the fullstring */ + field = id3_frame_field(frame, 3); + if (field == NULL) + return; + + ucs4 = id3_field_getfullstring(field); + if (ucs4 == NULL) + return; + + utf8 = import_id3_string(tag_is_id3v1(tag), ucs4, type); + if (utf8 == NULL) + return; + + tag_add_item(dest, type, (char *)utf8); + g_free(utf8); } /** @@ -339,23 +311,23 @@ struct tag *tag_id3_import(struct id3_tag * tag) { struct tag *ret = tag_new(); - tag_id3_import_frame(ret, tag, ID3_FRAME_ARTIST, TAG_ITEM_ARTIST); - tag_id3_import_frame(ret, tag, ID3_FRAME_ALBUM_ARTIST, - TAG_ITEM_ALBUM_ARTIST); - tag_id3_import_frame(ret, tag, ID3_FRAME_ARTIST_SORT, - TAG_ARTIST_SORT); - tag_id3_import_frame(ret, tag, ID3_FRAME_ALBUM_ARTIST_SORT, - TAG_ALBUM_ARTIST_SORT); - tag_id3_import_frame(ret, tag, ID3_FRAME_TITLE, TAG_ITEM_TITLE); - tag_id3_import_frame(ret, tag, ID3_FRAME_ALBUM, TAG_ITEM_ALBUM); - tag_id3_import_frame(ret, tag, ID3_FRAME_TRACK, TAG_ITEM_TRACK); - tag_id3_import_frame(ret, tag, ID3_FRAME_YEAR, TAG_ITEM_DATE); - tag_id3_import_frame(ret, tag, ID3_FRAME_GENRE, TAG_ITEM_GENRE); - tag_id3_import_frame(ret, tag, ID3_FRAME_COMPOSER, TAG_ITEM_COMPOSER); - tag_id3_import_frame(ret, tag, "TPE3", TAG_ITEM_PERFORMER); - tag_id3_import_frame(ret, tag, "TPE4", TAG_ITEM_PERFORMER); - tag_id3_import_frame(ret, tag, ID3_FRAME_COMMENT, TAG_ITEM_COMMENT); - tag_id3_import_frame(ret, tag, ID3_FRAME_DISC, TAG_ITEM_DISC); + tag_id3_import_text(ret, tag, ID3_FRAME_ARTIST, TAG_ITEM_ARTIST); + tag_id3_import_text(ret, tag, ID3_FRAME_ALBUM_ARTIST, + TAG_ITEM_ALBUM_ARTIST); + tag_id3_import_text(ret, tag, ID3_FRAME_ARTIST_SORT, + TAG_ARTIST_SORT); + tag_id3_import_text(ret, tag, ID3_FRAME_ALBUM_ARTIST_SORT, + TAG_ALBUM_ARTIST_SORT); + tag_id3_import_text(ret, tag, ID3_FRAME_TITLE, TAG_ITEM_TITLE); + tag_id3_import_text(ret, tag, ID3_FRAME_ALBUM, TAG_ITEM_ALBUM); + tag_id3_import_text(ret, tag, ID3_FRAME_TRACK, TAG_ITEM_TRACK); + tag_id3_import_text(ret, tag, ID3_FRAME_YEAR, TAG_ITEM_DATE); + tag_id3_import_text(ret, tag, ID3_FRAME_GENRE, TAG_ITEM_GENRE); + tag_id3_import_text(ret, tag, ID3_FRAME_COMPOSER, TAG_ITEM_COMPOSER); + tag_id3_import_text(ret, tag, "TPE3", TAG_ITEM_PERFORMER); + tag_id3_import_text(ret, tag, "TPE4", TAG_ITEM_PERFORMER); + tag_id3_import_comment(ret, tag, ID3_FRAME_COMMENT, TAG_ITEM_COMMENT); + tag_id3_import_text(ret, tag, ID3_FRAME_DISC, TAG_ITEM_DISC); tag_id3_import_musicbrainz(ret, tag); tag_id3_import_ufid(ret, tag); |