libsndfile : API Changes in Version 1.0.

Document Version 1.0 (2001/11/14)
Document Version 1.1 (2001/11/14)
Document Version 1.2 (2001/11/15)
Document Version 1.3 (2002/01/15) (Changes in green)
Document Version 1.4 (2002/04/16) (Changes in cyan)
Document Version 1.5 (2002/04/21) (Changes in yellow)
Document Version 1.6 (2002/04/30) (Changes in magenta)
Document Version 1.7 (2002/06/24) (Changes in red)

This document is now (Aug, 2002) out of date. It does however document the majority of the changes between version 0 and version 1. Some minor changes to the libsndfile API have been made and the changes are listed below. The full version 1 API can be found here.


File Open Functions

Version 0 of libsndfile supplied two functions for opening files; one for opening a file for read and one for write. Since it would also be desirable to allow opening a file for modification (ie read and write) it was decided to merge the two existing open functions and add the new read/write mode functionality to that. The new file open function which replaces sf_open_read and sf_open_write looks like this: 
	SNDFILE*    sf_open    (const char *path, int mode, SF_INFO *sfinfo) ;
where the mode parameter can be one of SFM_READ, SFM_WRITE and SFM_RDWR, for read, write and modification.

It is also desirable to allow completely independent read and write operations ie if the application programmer reads N frames followed by a write and then does a read of M frames, the application will retrieve the M frames immediately following the N read earlier.

Internally, libsndfile keeps track of the read and write locations using separate read and write pointers. In order to allow seeking to be specified on either the read pointer, the write pointer or both, the usage of the sf_seek function has been modified slightly.

The sf_seek function is defined as follows: 
    sf_count_t sf_seek (SNDFILE *sndfile, sf_count_t offset, int whence) ;
where the whence parameter is SEEK_SET, SEEK_CUR or SEEK_END. On files opened for read only or write only, this operation is unchanged. On files opened for modification, the standard seek operation will change both the read and the write location. To limit this the only read, the whence parameter should be logical ORed with SFM_READ, while to change only the write pointer, the whence parameter should be ORed with SFM_WRITE.




64 Bit File Offsets

Sonic Foundry's 64 bit riff/WAV is one of the new file formats which will be supported in version 1 of the library. This format allows for audio files with 64 bit file offsets. The AU file format, since it is simply a header followed by data should also support 64 bit file offsets. In addition, WAV and AIFF files use unsigned file offsets so they too can be larger than 2 gigabytes (although limited to 4 gigabytes) and it would be nice to support these as well.

After a bit of hacking and experimenting libsndfile now reads, writes and seeks files larger than 2 gigabytes :
  • Wherever the library is compiled as a 64 bit library on a 64 bit processor.
  • Linux (2.4 and later kernels).
  • Win32.
  • MacOSX (not sure about earlier MacOS).
This was achieved by dumping the ISO Standard C fopen/fread/fwrite/fseek/fclose functions and replacing them with wrapper functions around the POSIX open/read/ write/lseek/close functions. The big win here is that the POSIX definition of lseek() is : 
    off_t lseek(int fildes, off_t offset, int whence);
and on most BSD based systems (including MacOSX) off_t is a 64 bit value. On 32 bit Linux and Solaris, off_t is a 32 bit value but this can be overridden with some compiler flags resulting in a 64 bit off_t type.

To get around the problem of Win32 having a 32 bit off_t type and defining its 64 bit lseek function like this: 
    __int64 _lseeki64 (int handle, __int64 offset, int origin);
the sndfile header defines a type sf_count_t which will be set as needed.

Some systems like AmigaOS will still be limited to 2 gigabyte maximum file sizes.

As part of this shake out, the sf_seek () function call will now be defined as 
    sf_count_t sf_seek (SNDFILE *sndfile, sf_count_t offset, int whence) ;
Other changes in this area will be covered later in this document.


Command Interface (sf_command)

The sf_command () interface was previewed in version 0 of the library as the only way to turn normalisation on and off for the functions which read and wrote float data. This interface was modeled after the ioctl () system call in Unix.

In version 0 of the library, the prototype looked like this: 
    int  sf_command  (SNDFILE *sndfile, const char *cmd, void *data, size_t datasize) ;
For version 1 of the library, the command will not longer be encoded as a string but rather as an integer as follows: 
    int  sf_command  (SNDFILE *sndfile, int command, void *data, int datasize) ;
The valid values for command will have symbolic names beginning with SFC_ and will be listed in the header file <sndfile.h>.

Most of the commands (other than things like retrieving the library version) will work on a per file basis, allowing float normalisation to be turned on for some files and off for others.

When an integer values needs to be passed to the library via the sf_command interface, the data parameter should be a NULL pointer and the value passed in the datasize argument (see example below).

Here are some examples of its uses: 
        char buffer [1024] ;
        int	 value ;

        /* Retrieve a string containing the library version. No SNDFILE* 
        ** pointer needed 
        */
        sf_command  (NULL, SFC_GET_LIB_VERSION, buffer, sizeof (buffer)) ;

        /* Retrieve the log generated by the file header parser. */
        sf_command  (file, SFC_GET_LOG_INFO, buffer, sizeof (buffer)) ;

        /* Turn on normalisation of data read/written as float. */
        sf_command  (file, SFC_SET_NORM_FLOAT, NULL, SF_TRUE) ;

        /* Turn off normalisation of data read/written as float. */
        sf_command  (file, SFC_SET_NORM_FLOAT, NULL, SF_FALSE) ;

        /* Turn on normalisation of data read/written as double. */
        sf_command  (file, SFC_SET_NORM_DOUBLE, NULL, SF_TRUE) ;

		
        /* Retrieve current normalisation settings for doubles. */
        sf_command  (file, SFC_GET_NORM_DOUBLE, &value, sizeof (value)) ;
This function returns zero on success.

This interface will be put to other uses in the future.


File Open Functions

There are a number of minor changes here. Firstly the pcmbitwith field of the SF_INFO struct has been dropped, 
          typedef struct
          {   sf_count_t    frames ;
              int           samplerate ;
              int           channels ;
              int           format ;
              int           sections ;
              int           seekable ;
          } SF_INFO ;
and the types of these fields have been changed to signed types. The sample count is now an sf_count_t so that on machines where this is a 64 bit type, the sample count can contain a frame count of more than 2 giga samples.

The major format types will remain pretty much as they are. 
        SF_FORMAT_WAV           /* Microsoft WAV format (little endian). */
        SF_FORMAT_AIFF          /* Apple/SGI AIFF format (big endian). */
        SF_FORMAT_AU            /* Sun/NeXT AU format (big endian). */
        SF_FORMAT_RAW           /* RAW PCM data. */
        SF_FORMAT_PAF           /* Ensoniq PARIS file format. */
        SF_FORMAT_SVX           /* Amiga IFF / SVX8 / SV16 format. */
        SF_FORMAT_NIST          /* Sphere NIST format. */
        SF_FORMAT_SMPLTD        /* Sekd Samplitude. */
        SF_FORMAT_VOC           /* VOC files. */
        SF_FORMAT_SD2           /* Sound Designer 2 */
        SF_FORMAT_IRCAM         /* Berkeley/IRCAM/CARL */
        SF_FORMAT_W64           /* Sonic Foundry's 64 bit RIFF/WAV */
The really astute reader may notice that SF_FORMAT_AULE has disappeared. The reason for this will be dealt with below.

The following minor format types will be provided. Note that the bit width of PCM data is now encoded in the minor format. 
        SF_FORMAT_PCM_U8        /* Unsigned 8 bit data (WAV and RAW only) */
        SF_FORMAT_PCM_S8        /* Signed 8 bit data */
        SF_FORMAT_PCM_16        /* Signed 16 bit data */
        SF_FORMAT_PCM_24        /* Signed 24 bit data */
        SF_FORMAT_PCM_32        /* Signed 32 bit data */
        
        SF_FORMAT_FLOAT         /* 32 bit float data */
        SF_FORMAT_DOUBLE        /* 64 bit float data */
        
        SF_FORMAT_ULAW          /* U-Law encoded. */
        SF_FORMAT_ALAW          /* A-Law encoded. */
        SF_FORMAT_IMA_ADPCM     /* IMA ADPCM. */
        SF_FORMAT_MS_ADPCM      /* Microsoft ADPCM. */

        SF_FORMAT_GSM610        /* GSM 6.10 encoding. */

        SF_FORMAT_G721_32       /* 32kbs G721 ADPCM encoding. */
        SF_FORMAT_G723_24       /* 24kbs G723 ADPCM encoding. */

As with version 0 of the library, the major and minor format types are combined using the binary or operator. For instance, a 16 bit WAV file has a format obtained using (SF_FORMAT_WAV | SF_FORMAT_PCM_16). The constants SF_FORMAT_SUBMASK and SF_FORMAT_TYPEMASK behave as they do in version 0 of the library to mask out the major and minor types of the file format.

A number of file formats such as AU, PAF and IRCAM have both a big and a little endian version. For these files there are four new constants: 
        SF_ENDIAN_FILE          = 0x00000000,    /* file native */
        SF_ENDIAN_LITTLE        = 0x10000000,    /* forced little endian */
        SF_ENDIAN_BIG           = 0x20000000,    /* forced big endian */
        SF_ENDIAN_CPU           = 0x30000000,    /* forced CPU endian */

        SF_ENDIAN_MASK          = 0x30000000,    /* mask out the endian portion of format */
which are binary or-ed with the rest of the format. A format of (SF_FORMAT_AU | SF_FORMAT_FLOAT) would therefore create a big endian AU (AU is big endian by default) file, with 32 bit float data. By using the other three SF_ENDIAN_xxx constants, the default can be overridden. Of these three overrides, SF_FORMAT_CPU is the most interesting as data is stored in the native CPU endian-ness which is the endian-ness which will allow the fastest reads and writes of that data type regardless of the host processor type.


Read and Write Functions

To tie in with with 64 bit file offsets, sample and frame counts used with the read and write functions have been changed from type size_t to type sf_count_t. Wherever a function was defined in version 0 as : 
    size_t sf_read_type  (SNDFILE *sndfile, type *ptr, size_t items) ;
    size_t sf_readf_type (SNDFILE *sndfile, type *ptr, size_t frames) ;
this will now be defined as : 
    sf_count_t sf_read_type   (SNDFILE *sndfile, type *ptr, sf_count_t items) ;
    sf_count_t sf_readf_type  (SNDFILE *sndfile, type *ptr, sf_count_t frames) ;
These function will continue to return the number of items read or written. On error or end of file, a value of zero will be returned. A call to sf_error () will allow differentiation between error and end of file.


Read/Write double Functions

In version 0, the prototypes looked like this: 
    size_t  sf_read_double   (SNDFILE *sndfile, double *ptr, size_t items, int normalize) ;
    size_t  sf_readf_double  (SNDFILE *sndfile, double *ptr, size_t frames, int normalize) ;
    size_t  sf_write_double  (SNDFILE *sndfile, double *ptr, size_t items, int normalize) ;
    size_t  sf_writef_double (SNDFILE *sndfile, double *ptr, size_t frames, int normalize) ;
while in version 1 it will looks like this: 

    sf_count_t    sf_read_double   (SNDFILE *sndfile, double *ptr, sf_count_t items) ;
    sf_count_t    sf_readf_double  (SNDFILE *sndfile, double *ptr, sf_count_t frames) ;
    sf_count_t    sf_write_double  (SNDFILE *sndfile, double *ptr, sf_count_t items) ;
    sf_count_t    sf_writef_double (SNDFILE *sndfile, double *ptr, sf_count_t frames) ;
These have been changed to match the prototypes for sf_read/write_float () functions which do not have a normalize flag. Switching normalisation on and off can still be done using the sf_command () interface as is currently done for floats. Normalisation will be on by default.


PCM -> PCM Sample Scaling

There is currently an inconsistency in the way version 0 if the library handles conversion of PCM data from one sized container to another sized container.

Version 1 of the library will obey the following rule : whenever integer data is moved from one sized container to another sized container, the most significant bit in the source container will become the most significant bit in the destination container.


Faster float to int Conversions

Version 1 of the library uses the fast float to integer conversion techniques detailed here. This should work for Linux on all CPUs, Win3232 and MacOS (old and new).


Simple Format Enumeration Interface

During a discussion with Dominic Mazzoni it was decided that libsndfile needed an easy way for an application to present a choice of a small set of standard file formats to the user. This will be done using the sf_command () interface and a new typedef in <sndfile.h>: 
    typedef struct
    {   int              format;
        const char       *name;
        const char       *extension;
    } SF_FORMAT_INFO ;
There are two new commands which are used as follows: 
    SF_FORMAT_INFO  format_info ;
    int             format_count ;

    /* Retrieve the simple format count */
    sf_command (NULL, SFC_GET_SIMPLE_FORMAT_COUNT, &format_count, sizeof (format_count)) ;

    /* Retrieve format count 3 */
    format_info.format = 3 ;
    sf_command (NULL, SFC_GET_SIMPLE_FORMAT, &format_info, sizeof (format_info)) ;
The SF_FORMAT_INFO struct would then be filled with information like: 
    {   SF_FORMAT_AIFF | SF_FORMAT_PCM_S8, 
        "AIFF (Apple/SGI 8 bit)", 
        "aiff" 
        },
or 
    {   SF_FORMAT_WAV | SF_FORMAT_IMA_ADPCM, 
        "WAV (Microsoft 4 bit IMA ADPCM)", 
        "wav" 
        },
The SF_FORMAT_INFO structs retrieved from number 0 through the value obtained with SFC_GET_SIMPLE_FORMAT_COUNT, will be alphabetically sorted on the name field.


Format Enumeration Interface

After the above Simple Format Enumeration Interface was developed the need for a more extensive interface became apparent. Discussion with Dominic Mazzoni resulted in four new commands for the sf_command() interface; SFC_GET_FORMAT_MAJOR_COUNT, SFC_GET_FORMAT_MAJOR, SFC_GET_FORMAT_SUBTYPE_COUNT and SFC_GET_FORMAT_SUBTYPE.

These commands work very much like the Simple Format Enumeration Interface and use the same SF_FORMAT_INFO typedef. When combined with the sf_format_check () function, these new commands will allow an application program to build up a complete list of formats supported by the current version of libsndfile even if new formats have been added after the application was compiled.

The examples/ directory of the libsndfile-1.0.X source code distribution contains an example program list_formats.c which prints out a list of all the supported subtypes for all of the supported major formats.


Obsoleted Functions

The functions 
    size_t sf_get_header_info (SNDFILE *sndfile, char* buffer, size_t bufferlen, size_t offset) ;
    size_t sf_get_lib_version (char* buffer, size_t bufferlen) ;
    double sf_signal_max (SNDFILE *sndfile) ;
will be dropped and replaced with calls to sf_command which is documented here.

tm65e3b72e