view libmpdemux/audio_in.c @ 9278:caea8ed36b48

The reason why mplayer crashes (in some cases) when using x11 output and -wid (>0) parameter is this: Mplayer by default creates a colormap using DirectColor visual. If the window given to mplayer uses TrueColor visual there will be an error when mplayer sets the colormap for the window. This patch modifies mplayer to use TrueColor visual if the window given to mplayer uses TrueColor. Another solution is to make sure that the window given to mplayer is created using DirectColor visual if it is supported by the display. Jouni Tulkki <jitulkki@cc.hut.fi>
author arpi
date Tue, 04 Feb 2003 18:31:44 +0000
parents 772d6d27fd66
children 31f12f99118b
line wrap: on
line source

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>

#include "config.h"

#if defined(USE_TV) && defined(HAVE_TV_V4L)

#include "audio_in.h"
#include "mp_msg.h"
#include <string.h>
#include <errno.h>

// sanitizes ai structure before calling other functions
int audio_in_init(audio_in_t *ai, int type)
{
    ai->type = type;
    ai->setup = 0;

    ai->channels = -1;
    ai->samplerate = -1;
    ai->blocksize = -1;
    ai->bytes_per_sample = -1;
    ai->samplesize = -1;

    switch (ai->type) {
#ifdef HAVE_ALSA9	  
    case AUDIO_IN_ALSA:
	ai->alsa.handle = NULL;
	ai->alsa.log = NULL;
	ai->alsa.device = strdup("default");
	return 0;
#endif
#ifdef USE_OSS_AUDIO
    case AUDIO_IN_OSS:
	ai->oss.audio_fd = -1;
	ai->oss.device = strdup("/dev/dsp");
	return 0;
#endif
    default:
	return -1;
    }
}

int audio_in_setup(audio_in_t *ai)
{
    
    switch (ai->type) {
#ifdef HAVE_ALSA9	  
    case AUDIO_IN_ALSA:
	if (ai_alsa_init(ai) < 0) return -1;
	ai->setup = 1;
	return 0;
#endif
#ifdef USE_OSS_AUDIO
    case AUDIO_IN_OSS:
	if (ai_oss_init(ai) < 0) return -1;
	ai->setup = 1;
	return 0;
#endif
    default:
	return -1;
    }
}

int audio_in_set_samplerate(audio_in_t *ai, int rate)
{
    switch (ai->type) {
#ifdef HAVE_ALSA9	  
    case AUDIO_IN_ALSA:
	ai->req_samplerate = rate;
	if (!ai->setup) return 0;
	if (ai_alsa_setup(ai) < 0) return -1;
	return ai->samplerate;
#endif
#ifdef USE_OSS_AUDIO
    case AUDIO_IN_OSS:
	ai->req_samplerate = rate;
	if (!ai->setup) return 0;
	if (ai_oss_set_samplerate(ai) < 0) return -1;
	return ai->samplerate;
#endif
    default:
	return -1;
    }
}

int audio_in_set_channels(audio_in_t *ai, int channels)
{
    switch (ai->type) {
#ifdef HAVE_ALSA9	  
    case AUDIO_IN_ALSA:
	ai->req_channels = channels;
	if (!ai->setup) return 0;
	if (ai_alsa_setup(ai) < 0) return -1;
	return ai->channels;
#endif
#ifdef USE_OSS_AUDIO
    case AUDIO_IN_OSS:
	ai->req_channels = channels;
	if (!ai->setup) return 0;
	if (ai_oss_set_channels(ai) < 0) return -1;
	return ai->channels;
#endif
    default:
	return -1;
    }
}

int audio_in_set_device(audio_in_t *ai, char *device)
{
#ifdef HAVE_ALSA9	  
    int i;
#endif
    if (ai->setup) return -1;
    switch (ai->type) {
#ifdef HAVE_ALSA9	  
    case AUDIO_IN_ALSA:
	if (ai->alsa.device) free(ai->alsa.device);
	ai->alsa.device = strdup(device);
	/* mplayer cannot handle colons in arguments */
	for (i = 0; i < (int)strlen(ai->alsa.device); i++) {
	    if (ai->alsa.device[i] == '.') ai->alsa.device[i] = ':';
	}
	return 0;
#endif
#ifdef USE_OSS_AUDIO
    case AUDIO_IN_OSS:
	if (ai->oss.device) free(ai->oss.device);
	ai->oss.device = strdup(device);
	return 0;
#endif
    default:
	return -1;
    }
}

int audio_in_uninit(audio_in_t *ai)
{
    if (ai->setup) {
	switch (ai->type) {
#ifdef HAVE_ALSA9	  
	case AUDIO_IN_ALSA:
	    if (ai->alsa.log)
		snd_output_close(ai->alsa.log);
	    if (ai->alsa.handle) {
		snd_pcm_close(ai->alsa.handle);
	    }
	    ai->setup = 0;
	    return 0;
#endif
#ifdef USE_OSS_AUDIO
	case AUDIO_IN_OSS:
	    close(ai->oss.audio_fd);
	    ai->setup = 0;
	    return 0;
#endif
	}
    }
    return -1;
}

int audio_in_start_capture(audio_in_t *ai)
{
    switch (ai->type) {
#ifdef HAVE_ALSA9	  
    case AUDIO_IN_ALSA:
	return snd_pcm_start(ai->alsa.handle);
#endif
#ifdef USE_OSS_AUDIO
    case AUDIO_IN_OSS:
	return 0;
#endif
    default:
	return -1;
    }
}

int audio_in_read_chunk(audio_in_t *ai, unsigned char *buffer)
{
    int ret;
    
    switch (ai->type) {
#ifdef HAVE_ALSA9	  
    case AUDIO_IN_ALSA:
	ret = snd_pcm_readi(ai->alsa.handle, buffer, ai->alsa.chunk_size);
	if (ret != ai->alsa.chunk_size) {
	    if (ret < 0) {
		mp_msg(MSGT_TV, MSGL_ERR, "\nerror reading audio: %s\n", snd_strerror(ret));
		if (ret == -EPIPE) {
		    if (ai_alsa_xrun(ai) == 0) {
			mp_msg(MSGT_TV, MSGL_ERR, "Recovered from cross-run, some frames may be left out!\n");
		    } else {
			mp_msg(MSGT_TV, MSGL_ERR, "Fatal error, cannot recover!\n");
		    }
		}
	    } else {
		mp_msg(MSGT_TV, MSGL_ERR, "\nnot enough audio samples!\n");
	    }
	    return -1;
	}
	return ret;
#endif
#ifdef USE_OSS_AUDIO
    case AUDIO_IN_OSS:
	ret = read(ai->oss.audio_fd, buffer, ai->blocksize);
	if (ret != ai->blocksize) {
	    if (ret < 0) {
		mp_msg(MSGT_TV, MSGL_ERR, "\nerror reading audio: %s\n", strerror(errno));
	    } else {
		mp_msg(MSGT_TV, MSGL_ERR, "\nnot enough audio samples!\n");
	    }
	    return -1;
	}
	return ret;
#endif
    default:
	return -1;
    }
}

#endif