/* Hey Emacs, this is -*-C-*- ****************************************************************************** * linuxaudiodev.c -- Linux audio device for python. * * Author : Peter Bosch * Created On : Thu Mar 2 21:10:33 2000 * Last Modified By: Peter Bosch * Last Modified On: Fri Mar 24 11:27:00 2000 * Status : Unknown, Use with caution! * * Unless other notices are present in any part of this file * explicitly claiming copyrights for other people and/or * organizations, the contents of this file is fully copyright * (C) 2000 Peter Bosch, all rights reserved. ****************************************************************************** */ #include "Python.h" #include "structmember.h" #ifdef HAVE_UNISTD_H #include #endif #ifdef HAVE_FCNTL_H #include #endif #include #if defined(linux) #include typedef unsigned long uint32_t; #elif defined(__FreeBSD__) #include #ifndef SNDCTL_DSP_CHANNELS #define SNDCTL_DSP_CHANNELS SOUND_PCM_WRITE_CHANNELS #endif #endif typedef struct { PyObject_HEAD; int x_fd; /* The open file */ int x_icount; /* Input count */ int x_ocount; /* Output count */ uint32_t x_afmts; /* Supported audio formats */ } lad_t; static struct { int a_bps; uint32_t a_fmt; } audio_types[] = { { 8, AFMT_MU_LAW }, { 8, AFMT_U8 }, { 8, AFMT_S8 }, { 16, AFMT_U16_BE }, { 16, AFMT_U16_LE }, { 16, AFMT_S16_BE }, { 16, AFMT_S16_LE }, }; staticforward PyTypeObject Ladtype; static PyObject *LinuxAudioError; static lad_t * newladobject(PyObject *arg) { lad_t *xp; int fd, afmts, imode; char *mode; char *basedev; /* Check arg for r/w/rw */ if (!PyArg_ParseTuple(arg, "s:open", &mode)) return NULL; if (strcmp(mode, "r") == 0) imode = 0; else if (strcmp(mode, "w") == 0) imode = 1; else { PyErr_SetString(LinuxAudioError, "Mode should be one of 'r', or 'w'"); return NULL; } /* Open the correct device. The base device name comes from the * AUDIODEV environment variable first, then /dev/audio. The * control device tacks "ctl" onto the base device name. */ basedev = getenv("AUDIODEV"); if (!basedev) basedev = "/dev/dsp"; if ((fd = open(basedev, imode)) < 0) { PyErr_SetFromErrnoWithFilename(LinuxAudioError, basedev); return NULL; } if (imode && ioctl(fd, SNDCTL_DSP_NONBLOCK, NULL) < 0) { PyErr_SetFromErrnoWithFilename(LinuxAudioError, basedev); return NULL; } if (ioctl(fd, SNDCTL_DSP_GETFMTS, &afmts) < 0) { PyErr_SetFromErrnoWithFilename(LinuxAudioError, basedev); return NULL; } /* Create and initialize the object */ if ((xp = PyObject_New(lad_t, &Ladtype)) == NULL) { close(fd); return NULL; } xp->x_fd = fd; xp->x_icount = xp->x_ocount = 0; xp->x_afmts = afmts; return xp; } static void lad_dealloc(lad_t *xp) { /* if already closed, don't reclose it */ if (xp->x_fd != -1) close(xp->x_fd); PyObject_Del(xp); } static PyObject * lad_read(lad_t *self, PyObject *args) { int size, count; char *cp; PyObject *rv; if (!PyArg_ParseTuple(args, "i:read", &size)) return NULL; rv = PyString_FromStringAndSize(NULL, size); if (rv == NULL) return NULL; if (!(cp = PyString_AsString(rv))) { Py_DECREF(rv); return NULL; } if ((count = read(self->x_fd, cp, size)) < 0) { PyErr_SetFromErrno(LinuxAudioError); Py_DECREF(rv); return NULL; } self->x_icount += count; return rv; } static PyObject * lad_write(lad_t *self, PyObject *args) { char *cp; int rv, size; if (!PyArg_ParseTuple(args, "s#:write", &cp, &size)) return NULL; while (size > 0) { if ((rv = write(self->x_fd, cp, size)) < 0) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } self->x_ocount += rv; size -= rv; cp += rv; } Py_INCREF(Py_None); return Py_None; } static PyObject * lad_close(lad_t *self, PyObject *args) { if (!PyArg_ParseTuple(args, ":close")) return NULL; if (self->x_fd >= 0) { close(self->x_fd); self->x_fd = -1; } Py_INCREF(Py_None); return Py_None; } static PyObject * lad_fileno(lad_t *self, PyObject *args) { if (!PyArg_ParseTuple(args, ":fileno")) return NULL; return PyInt_FromLong(self->x_fd); } static PyObject * lad_setparameters(lad_t *self, PyObject *args) { int rate, ssize, nchannels, stereo, n, fmt; if (!PyArg_ParseTuple(args, "iiii:setparameters", &rate, &ssize, &nchannels, &fmt)) return NULL; if (rate < 0 || ssize < 0 || (nchannels != 1 && nchannels != 2)) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } if (ioctl(self->x_fd, SOUND_PCM_WRITE_RATE, &rate) < 0) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } if (ioctl(self->x_fd, SNDCTL_DSP_SAMPLESIZE, &ssize) < 0) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } stereo = (nchannels == 1)? 0: (nchannels == 2)? 1: -1; if (ioctl(self->x_fd, SNDCTL_DSP_STEREO, &stereo) < 0) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } for (n = 0; n != sizeof(audio_types) / sizeof(audio_types[0]); n++) if (fmt == audio_types[n].a_fmt) break; if (n == sizeof(audio_types) / sizeof(audio_types[0]) || audio_types[n].a_bps != ssize || (self->x_afmts & audio_types[n].a_fmt) == 0) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } if (ioctl(self->x_fd, SNDCTL_DSP_SETFMT, &audio_types[n].a_fmt) < 0) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } Py_INCREF(Py_None); return Py_None; } static int _ssize(lad_t *self, int *nchannels, int *ssize) { int fmt; fmt = 0; if (ioctl(self->x_fd, SNDCTL_DSP_SETFMT, &fmt) < 0) return -errno; switch (fmt) { case AFMT_MU_LAW: case AFMT_A_LAW: case AFMT_U8: case AFMT_S8: *ssize = sizeof(char); break; case AFMT_S16_LE: case AFMT_S16_BE: case AFMT_U16_LE: case AFMT_U16_BE: *ssize = sizeof(short); break; case AFMT_MPEG: case AFMT_IMA_ADPCM: default: return -EOPNOTSUPP; } *nchannels = 0; if (ioctl(self->x_fd, SNDCTL_DSP_CHANNELS, nchannels) < 0) return -errno; return 0; } /* bufsize returns the size of the hardware audio buffer in number of samples */ static PyObject * lad_bufsize(lad_t *self, PyObject *args) { audio_buf_info ai; int nchannels, ssize; if (!PyArg_ParseTuple(args, ":bufsize")) return NULL; if (_ssize(self, &nchannels, &ssize) < 0) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } if (ioctl(self->x_fd, SNDCTL_DSP_GETOSPACE, &ai) < 0) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } return PyInt_FromLong((ai.fragstotal * ai.fragsize) / (nchannels * ssize)); } /* obufcount returns the number of samples that are available in the hardware for playing */ static PyObject * lad_obufcount(lad_t *self, PyObject *args) { audio_buf_info ai; int nchannels, ssize; if (!PyArg_ParseTuple(args, ":obufcount")) return NULL; if (_ssize(self, &nchannels, &ssize) < 0) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } if (ioctl(self->x_fd, SNDCTL_DSP_GETOSPACE, &ai) < 0) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } return PyInt_FromLong((ai.fragstotal * ai.fragsize - ai.bytes) / (ssize * nchannels)); } /* obufcount returns the number of samples that can be played without blocking */ static PyObject * lad_obuffree(lad_t *self, PyObject *args) { audio_buf_info ai; int nchannels, ssize; if (!PyArg_ParseTuple(args, ":obuffree")) return NULL; if (_ssize(self, &nchannels, &ssize) < 0) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } if (ioctl(self->x_fd, SNDCTL_DSP_GETOSPACE, &ai) < 0) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } return PyInt_FromLong(ai.bytes / (ssize * nchannels)); } /* Flush the device */ static PyObject * lad_flush(lad_t *self, PyObject *args) { if (!PyArg_ParseTuple(args, ":flush")) return NULL; if (ioctl(self->x_fd, SNDCTL_DSP_SYNC, NULL) < 0) { PyErr_SetFromErrno(LinuxAudioError); return NULL; } Py_INCREF(Py_None); return Py_None; } static PyMethodDef lad_methods[] = { { "read", (PyCFunction)lad_read, METH_VARARGS }, { "write", (PyCFunction)lad_write, METH_VARARGS }, { "setparameters", (PyCFunction)lad_setparameters, METH_VARARGS }, { "bufsize", (PyCFunction)lad_bufsize, METH_VARARGS }, { "obufcount", (PyCFunction)lad_obufcount, METH_VARARGS }, { "obuffree", (PyCFunction)lad_obuffree, METH_VARARGS }, { "flush", (PyCFunction)lad_flush, METH_VARARGS }, { "close", (PyCFunction)lad_close, METH_VARARGS }, { "fileno", (PyCFunction)lad_fileno, METH_VARARGS }, { NULL, NULL} /* sentinel */ }; static PyObject * lad_getattr(lad_t *xp, char *name) { return Py_FindMethod(lad_methods, (PyObject *)xp, name); } static PyTypeObject Ladtype = { PyObject_HEAD_INIT(&PyType_Type) 0, /*ob_size*/ "linux_audio_device", /*tp_name*/ sizeof(lad_t), /*tp_size*/ 0, /*tp_itemsize*/ /* methods */ (destructor)lad_dealloc, /*tp_dealloc*/ 0, /*tp_print*/ (getattrfunc)lad_getattr, /*tp_getattr*/ 0, /*tp_setattr*/ 0, /*tp_compare*/ 0, /*tp_repr*/ }; static PyObject * ladopen(PyObject *self, PyObject *args) { return (PyObject *)newladobject(args); } static PyMethodDef linuxaudiodev_methods[] = { { "open", ladopen, METH_VARARGS }, { 0, 0 }, }; void initlinuxaudiodev(void) { PyObject *m, *d, *x; m = Py_InitModule("linuxaudiodev", linuxaudiodev_methods); d = PyModule_GetDict(m); LinuxAudioError = PyErr_NewException("linuxaudiodev.error", NULL, NULL); if (LinuxAudioError) PyDict_SetItemString(d, "error", LinuxAudioError); x = PyInt_FromLong((long) AFMT_MU_LAW); if (x == NULL || PyDict_SetItemString(d, "AFMT_MU_LAW", x) < 0) goto error; Py_DECREF(x); x = PyInt_FromLong((long) AFMT_U8); if (x == NULL || PyDict_SetItemString(d, "AFMT_U8", x) < 0) goto error; Py_DECREF(x); x = PyInt_FromLong((long) AFMT_S8); if (x == NULL || PyDict_SetItemString(d, "AFMT_S8", x) < 0) goto error; Py_DECREF(x); x = PyInt_FromLong((long) AFMT_U16_BE); if (x == NULL || PyDict_SetItemString(d, "AFMT_U16_BE", x) < 0) goto error; Py_DECREF(x); x = PyInt_FromLong((long) AFMT_U16_LE); if (x == NULL || PyDict_SetItemString(d, "AFMT_U16_LE", x) < 0) goto error; Py_DECREF(x); x = PyInt_FromLong((long) AFMT_S16_BE); if (x == NULL || PyDict_SetItemString(d, "AFMT_S16_BE", x) < 0) goto error; Py_DECREF(x); x = PyInt_FromLong((long) AFMT_S16_LE); if (x == NULL || PyDict_SetItemString(d, "AFMT_S16_LE", x) < 0) goto error; Py_DECREF(x); error: return; }