Backport of 4dcf0b41c6b01f45e141https://github.com/taglib/taglib/pull/77
Tested with files larger than 2GB which have been created using
sox -n -r 44100 -C 320 large.mp3 synth 58916 sine 440 channels 2
sox -n -r 44100 -C 0 large.flac synth 25459 sine 440 channels 2
sox -n -r 44100 -C 10 large.ogg synth 229806 sine 440 channels 2
sox -n -r 44100 large.wav synth 6692 sine 440 channels 2
ffmpeg -f lavfi -i "sine=frequency=440:duration=244676" -y large.m4a
The only file which was readable with the tagreader example
before this commit was large.ogg. The problem is that long on
Windows is only 32-bit (also in LLP64 data model of 64-bit
compilation target) and all the file offsets using long are
too small for large files. Now long is replaced by offset_t
(defined to be long long on Windows and off_t on UNIX) for such
cases and some unsigned long are now size_t, which has the
correct size even on Windows.
Since the Doxygen output changed quite a bit since the styling was adapted for
TagLib, the API documentation generated with Doxygen 1.9.1 has several problems:
- In the namespaces and classes lists, the nodes cannot be expanded,
'N' and 'C' characters are prepended to the namespaces and classes instead
of displaying icons.
- Class members are only accessible for the letter 'a'.
- No "Deprecated" sections are displayed (only the "TAGLIB_DEPRECATED"
definition in the signature).
- Some things are strangely styled (e.g. an empty box on the left of the
"Introduction" title).
Now most of the custom styling is removed to get it running again and to avoid
nasty surprises with doxygen changes.
Some ID3v2.4.0 frames such as text information frames support multiple strings
separated by the termination code of the character encoding. If the encoding
is $01 UTF-16 with BOM, all strings shall have the same byte order. In the
multi strings written by TagLib, all string elements of such a multi string
have a BOM. However, I have often seen tags where a BOM exists only at the
beginning, i.e. at the start of the first string. In such a case, TagLib will
only return a list with the first string and a second empty string. This
commit will detect such cases and parse the strings without BOM according to
the BOM of the first string.
There are m4a files with regular (non-full) meta atoms. When such
a meta atom is not correctly parsed, the subsequent atoms are not
recognized and offsets will not be adjusted when atoms are added,
which will corrupt the MP4 file.
This change will look behind the meta atom to check if the next
atom follows directly, i.e. without the four bytes with version
and flags as they exist in full atoms. In such a case, these
four bytes will not be skipped.
Witnesses of this strange format specification are
https://leo-van-stee.github.io/https://github.com/axiomatic-systems/Bento4/blob/v1.6.0-639/Source/C%2B%2B/Core/Ap4ContainerAtom.cpp#L60
The main point of IOStream is to be able to use taglib with some content
that can't be fopen()'ed, for instance a network file, or a local file
on a system that doesn't support opening a file directly through fopen
& such.
This commit adds an overload for detectByResolvers that will stick to
using an IOStream all the way.
