/*
 * Praat wrappers for libMAD (MPEG Audio Decoder) Copyright 2007 Erez Volk
 *
 * This program 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; either version 2 of the License, or (at
 * your option) any later version.
 *
 * This program 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, write to the Free Software
 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 *
 *
 *
 * Exact seeking in an MP3 file turns out to be quite complex.
 *
 * For constant bit rate (CBR) files, frame size is fixed and we can calculate
 * in advance the location of every frame.
 *
 * For variable bit rate (VBR) files, no option gives us more than a rough
 * estimate of where every percent of the file is lcoated, with no way of
 * checking (after we make a seek) where we ended up.  The only reliable way we
 * have, then, is to scan the entire file in advance and keep a table with the
 * offsets of all (or some) frames.  In the worst case, we will have to scan
 * the file *twice*, since we can't know the exact number of frames.
 *
 * What's worse, there is no guaranteed standard way of checking whether an MP3
 * file is CBR or VBR.  We therefore use the following compromise, which is
 * guaranteed to work on all files:
 *
 * - If there is a Xing header, read it to get the number of frames.
 * - Otherwise *estimate* the number of frames.
 * - Keep a reasonably (?) sized table of "key" offsets.
 * - Scan all the headers and keep required offsets in the table.
 * - After the scan, we also know the precise number of frames and samples.
 *
 * TODO: Find exactly what the encoder delay is.
 *       (see http://mp3decoders.mp3-tech.org/decoders_lame.html)
 * TODO: Compensate for end padding.
 * TODO: Better conversion to short.
 *
 */

/*#define MP3_DEBUG*/

#include <string.h>
#include "mp3.h"
#include "melder.h"

extern "C" {
	#include "mad_config.h"
	#include "mad_decoder.h"
}

#define MP3F_BUFFER_SIZE (8 * 1024)
#define MP3F_MAX_LOCATIONS 1024

/*
 * MP3 encoders and decoders add a number of silent samples at the beginning.
 * I don't know of any reliable way to detect these delays in an MP3 file.
 * The numbers below are (hopefully reasonable) estimates based on LAME and
 * some testing; it is more important to preserve all the meaningful samples
 * than to eliminate all silent ones.
 */
#define MP3F_DECODER_DELAY 529
/*#define MP3F_ENCODER_DELAY 576*/
#define MP3F_ENCODER_DELAY 96

#ifdef MP3_DEBUG
#   define MP3_DPRINTF(x) printf x
#   define MP3_PERCENT(base, value) (((float)(value) - (float)(base)) * 100.0 / (float)(base))
#else /* !MP3_DEBUG */
#   define MP3_DPRINTF(x)
#endif /* MP3_DEBUG */

struct _MP3_FILE
{
	struct mad_decoder decoder;

	FILE *f;
	unsigned char buffer [MP3F_BUFFER_SIZE];

	int xing;

	unsigned channels;
	unsigned frequency;
	unsigned frames;
	unsigned samples_per_frame;
	MP3F_OFFSET samples;

	MP3F_OFFSET locations[MP3F_MAX_LOCATIONS]; 
	unsigned num_locations;
	unsigned frames_per_location;

	unsigned delay;

	MP3F_CALLBACK callback;
	void *context;

	MP3F_OFFSET next_read_position;
	MP3F_OFFSET read_amount;
	MP3F_OFFSET first_offset;
	unsigned skip_amount;
	int need_seek;
};

static enum mad_flow mp3f_mad_error (void *context, struct mad_stream *stream, struct mad_frame *frame);
static enum mad_flow mp3f_mad_input (void *context, struct mad_stream *stream);
static enum mad_flow mp3f_mad_first_header (void *context, struct mad_header const *header);
static enum mad_flow mp3f_mad_first_filter (void *context,
		struct mad_stream const *stream,
		struct mad_frame *frame);
static int mp3f_check_xing (MP3_FILE mp3f, struct mad_stream const *stream);
static enum mad_flow mp3f_mad_scan_header (void *context, struct mad_header const *header);
static enum mad_flow mp3f_mad_report_samples (void *context, struct mad_header const *header, struct mad_pcm *pcm);

int mp3_recognize (int nread, const char *data)
{
	const unsigned char *bytes = (const unsigned char *)data;

	if (nread < 3)
		return 0;

	/* MP3 files can start with an ID3 tag */
	if (bytes [0] == 'I' && bytes [1] == 'D' && bytes [2] == '3')
		return 1;

	/* Otherwise the beginning of the file must be an MP3 frame */
	if (bytes [0] != 0xFF)
		return 0;

	/* This is not a foolproof check, but it is similar to file(1) */
	return
		((bytes [1] & 0xFE) == 0xFA) || /* MPEG ADTS, layer III, v1 */
		((bytes [1] & 0xFE) == 0xFC) || /* MPEG ADTS, layer II, v1 */
		((bytes [1] & 0xFE) == 0xFE) || /* MPEG ADTS, layer I, v1 */
		((bytes [1] & 0xFE) == 0xF2) || /* MPEG ADTS, layer III, v2 */
		((bytes [1] & 0xFE) == 0xF4) || /* MPEG ADTS, layer II, v2 */
		((bytes [1] & 0xFE) == 0xF6) || /* MPEG ADTS, layer I, v2 */
		((bytes [1] & 0xFE) == 0xE2);   /* MPEG ADTS, layer III, v2.5 */
}

MP3_FILE mp3f_new (void)
{
	try {
		return Melder_calloc (struct _MP3_FILE, 1);
	} catch (MelderError) {
		Melder_throw ("Cannot create MP3 file object.");
	}
}

void mp3f_delete (MP3_FILE mp3f)
{
	Melder_free (mp3f);
}

void mp3f_set_file (MP3_FILE mp3f, FILE *f)
{
	mp3f -> f = f;
	if (! f)
		return;
	fseek (f, 0, SEEK_SET);
	mp3f -> next_read_position = 0;
	mp3f -> need_seek = 0;
	mp3f -> delay = MP3F_DECODER_DELAY + MP3F_ENCODER_DELAY;
	mp3f -> skip_amount = mp3f -> delay;
	mp3f -> first_offset = 0;
}

int mp3f_analyze (MP3_FILE mp3f)
{
	struct mad_decoder *decoder = & mp3f -> decoder;
	int status;
#ifdef MP3_DEBUG
	unsigned estimate, last;
#endif /* MP3_DEBUG */

	if (! mp3f || ! mp3f -> f)
		return 0;

	fseek (mp3f -> f, 0, SEEK_SET);

	mp3f -> xing = 0;
	mp3f -> channels = 0;
	mp3f -> frequency = 0;
	mp3f -> frames = 0;
	mp3f -> samples = 0;
	mp3f -> samples_per_frame = 0;
	mp3f -> num_locations = 0;

	/* Read first frames to get basic parameters and hopefully Xing */
	mad_decoder_init (decoder, 
			mp3f,
			mp3f_mad_input,
			mp3f_mad_first_header,
			mp3f_mad_first_filter,
			NULL /* Output: Don't actually decode for now */,
			mp3f_mad_error,
			NULL /* Message */);

	status = mad_decoder_run (decoder, MAD_DECODER_MODE_SYNC);
	if (status != 0)
		goto end;

	/*
	 * If we don't have a Xing header we need to estimate the frame count.
	 * This doesn't have to be accurate since we're going to count them
	 * later when we scan for header offsets.
	 */
	if (! mp3f -> xing) {
		MP3F_OFFSET file_size, frame_size;

		/* Take size of first frame */
		frame_size = mp3f -> locations [1] - mp3f -> locations[0];

		/* For file size, seek to end */
		fseek (mp3f -> f, 0, SEEK_END);
		file_size = ftell (mp3f -> f);

		/* This estimate will be pretty accurate for CBR */
		mp3f -> frames = file_size / frame_size;

		MP3_DPRINTF (("File size: %lu bytes\n", (unsigned long)file_size));
		MP3_DPRINTF (("First frame size: %lu bytes\n", (unsigned long)frame_size));
		MP3_DPRINTF (("Estimated frames: %lu\n", (unsigned long)mp3f -> frames));
	}

	/* Calculate how many frames can fit in a "location" */
	if (mp3f -> frames <= MP3F_MAX_LOCATIONS)
		mp3f -> frames_per_location = 1;
	else
		mp3f -> frames_per_location = (mp3f -> frames + MP3F_MAX_LOCATIONS - 1) / MP3F_MAX_LOCATIONS;

	MP3_DPRINTF (("MP3: Each location is %u frame(s) (%.3fs), each %u samples\n",
				mp3f -> frames_per_location,
				mp3f -> frames_per_location * mp3f -> samples_per_frame / (float)mp3f -> frequency,
				mp3f -> samples_per_frame));

	/* Read all frames to get offsets*/
#ifdef MP3_DEBUG
	estimate = mp3f -> frames;
#endif /* MP3_DEBUG */
	mp3f -> num_locations = 0;
	mp3f -> frames = 0;
	mp3f -> samples = 0;

	fseek (mp3f -> f, 0, SEEK_SET);
	mad_decoder_init (decoder, 
			mp3f,
			mp3f_mad_input,
			mp3f_mad_scan_header,
			NULL /* Filter */,
			NULL /* Output */,
			mp3f_mad_error,
			NULL /* Message */);

	status = mad_decoder_run (decoder, MAD_DECODER_MODE_SYNC);

	MP3_DPRINTF (("MP3 Frames: %u, estimated %u (%+.2f%%)\n",
		       	mp3f -> frames,
		       	estimate,
			MP3_PERCENT (mp3f -> frames, estimate)));
#ifdef MP3_DEBUG
	last = mp3f -> frames - (mp3f -> frames_per_location * (mp3f -> num_locations - 1));
	MP3_DPRINTF (("MP3F: Last location frames = %u (%+.2f%%) = %.3fs\n",
				last,
				MP3_PERCENT (mp3f -> frames_per_location, last),
				last * mp3f -> samples_per_frame / (float)mp3f -> frequency));
#endif /* MP3_DEBUG */

	mp3f_seek (mp3f, 0);

end:
	mad_decoder_finish (decoder);

	return (status == 0);
}

unsigned mp3f_channels (MP3_FILE mp3f)
{
	return mp3f -> channels;
}

unsigned mp3f_frequency (MP3_FILE mp3f)
{
	return mp3f -> frequency;
}

MP3F_OFFSET mp3f_samples (MP3_FILE mp3f)
{
	return mp3f -> samples - mp3f -> delay;
}

void mp3f_set_callback (MP3_FILE mp3f,
	       	MP3F_CALLBACK callback, void *context)
{
	mp3f -> callback = callback;
	mp3f -> context = context;
}

int mp3f_seek (MP3_FILE mp3f, MP3F_OFFSET sample)
{
	MP3F_OFFSET frame, location, base, offset;

	if (! mp3f || ! mp3f -> f)
		return 0;

	if (! mp3f -> frames_per_location)
		if (! mp3f_analyze (mp3f))
			return 0;

	/* Compensate for initial empty frames */
	sample += mp3f -> delay;

	/* Calculate where we need to seek */
	frame = sample / mp3f -> samples_per_frame;
	if ( frame ) /* libMAD can skip the first frame... */
		-- frame; 
	if ( frame ) /* ...and the first frame it decodes is useless */
		-- frame; 
	location = frame / mp3f -> frames_per_location;
	if (location >= mp3f -> num_locations)
		location = mp3f -> num_locations - 1;
	frame = location * mp3f -> frames_per_location;
	base = frame * mp3f -> samples_per_frame;

	offset = mp3f -> locations [location];
	if (fseek (mp3f -> f, offset, SEEK_SET) < 0)
		return 0;

	mp3f -> first_offset = offset;
	mp3f -> skip_amount = sample - base;
	mp3f -> need_seek = 0;

	MP3_DPRINTF (("SEEK to %lu (%lu + %u): Frame %lu, location %lu, offset %lu, base %lu, skip %u\n",
			(unsigned long)sample, 
			(unsigned long)sample - mp3f -> delay,
		       	mp3f -> delay,
			(unsigned long)frame, 
			(unsigned long)location, 
			(unsigned long)offset,
			(unsigned long)base,
		       	mp3f -> skip_amount));

	return 1;
}

int mp3f_read (MP3_FILE mp3f, MP3F_OFFSET num_samples)
{
	int status;
	struct mad_decoder *decoder = &(mp3f -> decoder);

	if (! mp3f || ! mp3f -> f || ! mp3f -> callback)
		return 0;

	/* Seek if the last read left us in mid-frame */
	if (mp3f -> need_seek)
		if (! mp3f_seek (mp3f, mp3f -> next_read_position))
			return 0;

	mad_decoder_init (decoder, 
			mp3f,
			mp3f_mad_input,
			NULL /* Header */,
			NULL /* Filter */,
			mp3f_mad_report_samples,
			mp3f_mad_error,
			NULL /* Message */);

	mp3f -> read_amount = num_samples;
	status = mad_decoder_run (decoder, MAD_DECODER_MODE_SYNC);
	mad_decoder_finish (decoder);

	mp3f -> next_read_position += num_samples;
	return (status == 0);
}

static enum mad_flow mp3f_mad_report_samples (void *context, struct mad_header const *header, struct mad_pcm *pcm)
{
	MP3_FILE mp3f = (MP3_FILE) context;
	const int *channels [] = { pcm -> samples [0], pcm -> samples [1] };
	unsigned length = pcm -> length;

	if (! mp3f || ! mp3f -> callback)
		return MAD_FLOW_BREAK;
	
	if (mp3f -> first_offset) {
		/* libMAD can decide to skip the first frame */
		if (header -> offset > mp3f -> first_offset) {
			MP3_DPRINTF (("Skip %u of %lu\n", length, mp3f -> skip_amount));
			mp3f -> skip_amount -= length;
		}
		mp3f -> first_offset = 0;
	}

	if (mp3f -> skip_amount >= length) {
		mp3f -> skip_amount -= length;
		return MAD_FLOW_IGNORE;
	}

	if (mp3f -> skip_amount > 0) {
		channels [0] += mp3f -> skip_amount;
		channels [1] += mp3f -> skip_amount;
		length -= mp3f -> skip_amount;
		mp3f -> skip_amount = 0;
	}

	if (length > mp3f -> read_amount) {
		length = mp3f -> read_amount;
		mp3f -> need_seek = 1;
	}

	if (length > 0)
		mp3f -> callback (channels, length, mp3f -> context);

	mp3f -> read_amount -= length;
	return (mp3f -> read_amount > 0) ? MAD_FLOW_CONTINUE : MAD_FLOW_STOP;
}

/* This conversion function was taken from minimad, and it could be better */
short mp3f_sample_to_short (MP3F_SAMPLE sample)
{
	/* round */
	sample += (1L << (MAD_F_FRACBITS - 16));

	/* clip */
	if (sample >= MAD_F_ONE)
		sample = MAD_F_ONE - 1;
	else if (sample < -MAD_F_ONE)
		sample = -MAD_F_ONE;

	/* quantize */
	return sample >> (MAD_F_FRACBITS + 1 - 16);
}

/* This function was adapted from libmad */
static enum mad_flow mp3f_mad_error(void *context, struct mad_stream *stream, struct mad_frame *frame)
{
	(void) context;
	(void) stream;
	(void) frame;
	return MAD_FLOW_CONTINUE;
}

/* This function was adapted from Audacity */
static enum mad_flow mp3f_mad_input(void *context, struct mad_stream *stream)
{
	MP3_FILE mp3f = (MP3_FILE) context;
	FILE *f = mp3f -> f;
	unsigned char *buffer = NULL;
	unsigned nthrown = 0, ncopied = 0, size = 0;
	size_t nread = 0;
	MP3F_OFFSET offset;

	if (feof (f))
		return MAD_FLOW_STOP;

	if (stream -> next_frame) {
		nthrown = stream -> next_frame - mp3f -> buffer; 
		ncopied = MP3F_BUFFER_SIZE - nthrown;
		memmove (mp3f -> buffer, stream -> next_frame, ncopied);
	}

	buffer = mp3f -> buffer + ncopied;
	size = MP3F_BUFFER_SIZE - ncopied;
	offset = ftell (f) - ncopied;

	if (size > 0)
		nread = fread (buffer, 1, size, f);

	mad_stream_buffer_offset (stream, mp3f -> buffer, nread + ncopied, offset);
	return MAD_FLOW_CONTINUE;
}

static enum mad_flow mp3f_mad_first_header(void *context, struct mad_header const *header)
{
	MP3_FILE mp3f = (MP3_FILE) context;

	mp3f -> channels = MAD_NCHANNELS (header);
	mp3f -> frequency = header -> samplerate;
	mp3f -> samples_per_frame = 32 * MAD_NSBSAMPLES (header);
	/* Just in case there is no Xing header: */
	mp3f -> locations [mp3f -> num_locations ++] = header -> offset;

	return MAD_FLOW_CONTINUE;
}

static enum mad_flow mp3f_mad_first_filter (void *context,
		struct mad_stream const *stream,
		struct mad_frame *frame)
{
	MP3_FILE mp3f = (MP3_FILE) context;

	(void) frame;

	mp3f -> xing = mp3f_check_xing (mp3f, stream);

	/* Xing? No need to look further */
	if (mp3f -> xing)
		return MAD_FLOW_STOP;

	/* Otherwise, read two frames for size estimate */
	return (mp3f -> num_locations < 2) ? MAD_FLOW_IGNORE : MAD_FLOW_STOP;
}

static enum mad_flow mp3f_mad_scan_header(void *context, struct mad_header const *header)
{
	MP3_FILE mp3f = (MP3_FILE) context;

	/* Some sanity checks */
	if (mp3f -> channels != MAD_NCHANNELS (header))
		return MAD_FLOW_BREAK;

	if (mp3f -> frequency != header -> samplerate)
		return MAD_FLOW_BREAK;

	if (mp3f -> samples_per_frame != 32 * MAD_NSBSAMPLES (header))
		return MAD_FLOW_BREAK;

	/* Check whether to log this offset in the table */
	if ((mp3f -> frames % mp3f -> frames_per_location) == 0 &&
			mp3f -> num_locations < MP3F_MAX_LOCATIONS)
		mp3f -> locations [mp3f -> num_locations ++] = header -> offset;

	/* Count this frame */
	++ mp3f -> frames;
	mp3f -> samples += mp3f -> samples_per_frame;

	return MAD_FLOW_IGNORE;
}

/*
 * Identify a Xing VBR header.
 * This was adapted from madplay.
 */
# define XING_VBR_MAGIC	(('X' << 24) | ('i' << 16) | ('n' << 8) | 'g')
# define XING_CBR_MAGIC	(('I' << 24) | ('n' << 16) | ('f' << 8) | 'o')

enum {
  XING_FLAGS_FRAMES = 0x00000001L,
  XING_FLAGS_BYTES  = 0x00000002L,
  XING_FLAGS_TOC    = 0x00000004L,
  XING_FLAGS_SCALE  = 0x00000008L
};

static int mp3f_check_xing (MP3_FILE mp3f, struct mad_stream const *stream)
{
	struct mad_bitptr ptr = stream -> anc_ptr;
	unsigned long magic, flags;

	/* When we get here we have the following data from the header:
	 *  channels
	 *  frequency
	 *  samples_per_frame
	 * We need to calculate the total number of frames and samples.
	 */

	magic = mad_bit_read (&ptr, 32);
	if (magic != XING_CBR_MAGIC && magic != XING_VBR_MAGIC)
		return 0;

	flags = mad_bit_read (&ptr, 32);
	if ((flags & XING_FLAGS_FRAMES) != XING_FLAGS_FRAMES)
		return 0;

	mp3f -> frames = mad_bit_read (&ptr, 32);
	mp3f -> samples = mp3f -> samples_per_frame * mp3f -> frames;

	return 1;
}