Logo Search packages:      
Sourcecode: ardour version File versions

audio_diskstream.cc

/*
    Copyright (C) 2000-2006 Paul Davis 

    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.
*/

#include <fstream>
#include <cstdio>
#include <unistd.h>
#include <cmath>
#include <cerrno>
#include <cassert>
#include <string>
#include <climits>
#include <fcntl.h>
#include <cstdlib>
#include <ctime>
#include <sys/stat.h>
#include <sys/mman.h>

#include <pbd/error.h>
#include <pbd/basename.h>
#include <glibmm/thread.h>
#include <pbd/xml++.h>
#include <pbd/memento_command.h>
#include <pbd/enumwriter.h>
#include <pbd/stacktrace.h>

#include <ardour/ardour.h>
#include <ardour/audioengine.h>
#include <ardour/analyser.h>
#include <ardour/audio_diskstream.h>
#include <ardour/utils.h>
#include <ardour/configuration.h>
#include <ardour/audiofilesource.h>
#include <ardour/send.h>
#include <ardour/region_factory.h>
#include <ardour/audioplaylist.h>
#include <ardour/playlist_factory.h>
#include <ardour/cycle_timer.h>
#include <ardour/audioregion.h>
#include <ardour/source_factory.h>

#include "i18n.h"
#include <locale.h>

using namespace std;
using namespace ARDOUR;
using namespace PBD;

size_t  AudioDiskstream::_working_buffers_size = 0;
Sample* AudioDiskstream::_mixdown_buffer       = 0;
gain_t* AudioDiskstream::_gain_buffer          = 0;

AudioDiskstream::AudioDiskstream (Session &sess, const string &name, Diskstream::Flag flag)
      : Diskstream(sess, name, flag)
      , deprecated_io_node(NULL)
      , channels (new ChannelList)
{
      /* prevent any write sources from being created */

      in_set_state = true;

      init(flag);
      use_new_playlist ();

      in_set_state = false;
}
      
AudioDiskstream::AudioDiskstream (Session& sess, const XMLNode& node)
      : Diskstream(sess, node)
      , deprecated_io_node(NULL)
      , channels (new ChannelList)
{
      in_set_state = true;
      init (Recordable);

      if (set_state (node)) {
            in_set_state = false;
            throw failed_constructor();
      }

      in_set_state = false;

      if (destructive()) {
            use_destructive_playlist ();
      }
}

void
AudioDiskstream::init (Diskstream::Flag f)
{
      Diskstream::init(f);

      /* there are no channels at this point, so these
         two calls just get speed_buffer_size and wrap_buffer
         size setup without duplicating their code.
      */

      set_block_size (_session.get_block_size());
      allocate_temporary_buffers ();

      add_channel (1);
      assert(_n_channels == 1);
}

AudioDiskstream::~AudioDiskstream ()
{
      notify_callbacks ();

      {
            RCUWriter<ChannelList> writer (channels);
            boost::shared_ptr<ChannelList> c = writer.get_copy();
            
            for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {
                  delete *chan;
            }

            c->clear();
      }

      channels.flush ();
}

void
AudioDiskstream::allocate_working_buffers()
{
      assert(disk_io_frames() > 0);

      _working_buffers_size = disk_io_frames();
      _mixdown_buffer       = new Sample[_working_buffers_size];
      _gain_buffer          = new gain_t[_working_buffers_size];
}

void
AudioDiskstream::free_working_buffers()
{
      delete [] _mixdown_buffer;
      delete [] _gain_buffer;
      _working_buffers_size = 0;
      _mixdown_buffer       = 0;
      _gain_buffer          = 0;
}

void
AudioDiskstream::non_realtime_input_change ()
{
      {
            Glib::Mutex::Lock lm (state_lock);

            if (input_change_pending == NoChange) {
                  return;
            }

            {
                  RCUWriter<ChannelList> writer (channels);
                  boost::shared_ptr<ChannelList> c = writer.get_copy();
                  
                  _n_channels = c->size();
                  
                  if (_io->n_inputs() > _n_channels) {
                        add_channel_to (c, _io->n_inputs() - _n_channels);
                  } else if (_io->n_inputs() < _n_channels) {
                        remove_channel_from (c, _n_channels - _io->n_inputs());
                  }
            }
            
            get_input_sources ();
            set_capture_offset ();
            
            if (first_input_change) {
                  set_align_style (_persistent_alignment_style);
                  first_input_change = false;
            } else {
                  set_align_style_from_io ();
            }
            
            input_change_pending = NoChange;

            /* implicit unlock */
      }
      
      /* reset capture files */

      reset_write_sources (false);

      /* now refill channel buffers */

      if (speed() != 1.0f || speed() != -1.0f) {
            seek ((nframes_t) (_session.transport_frame() * (double) speed()));
      } else {
            seek (_session.transport_frame());
      }
}

void
AudioDiskstream::get_input_sources ()
{
      boost::shared_ptr<ChannelList> c = channels.reader();

      uint32_t n;
      ChannelList::iterator chan;
      uint32_t ni = _io->n_inputs();

      for (n = 0, chan = c->begin(); chan != c->end() && n < ni; ++chan, ++n) {
            
            const char **connections = _io->input(n)->get_connections ();
            
            if (connections == 0 || connections[0] == 0) {
                  
                  if ((*chan)->source) {
                        // _source->disable_metering ();
                  }
                  
                  (*chan)->source = 0;
                  
            } else {
                  (*chan)->source = _session.engine().get_port_by_name (connections[0]);
            }
            
            if (connections) {
                  free (connections);
            }
      }
}           

int
AudioDiskstream::find_and_use_playlist (const string& name)
{
      boost::shared_ptr<AudioPlaylist> playlist;
            
      if ((playlist = boost::dynamic_pointer_cast<AudioPlaylist> (_session.playlist_by_name (name))) == 0) {
            playlist = boost::dynamic_pointer_cast<AudioPlaylist> (PlaylistFactory::create (_session, name));
      }

      if (!playlist) {
            error << string_compose(_("AudioDiskstream: Playlist \"%1\" isn't an audio playlist"), name) << endmsg;
            return -1;
      }

      return use_playlist (playlist);
}

int
AudioDiskstream::use_playlist (boost::shared_ptr<Playlist> playlist)
{
      assert(boost::dynamic_pointer_cast<AudioPlaylist>(playlist));

      Diskstream::use_playlist(playlist);

      return 0;
}

int
AudioDiskstream::use_new_playlist ()
{
      string newname;
      boost::shared_ptr<AudioPlaylist> playlist;
      
      if (!in_set_state && destructive()) {
            return 0;
      }

      if (_playlist) {
            newname = Playlist::bump_name (_playlist->name(), _session);
      } else {
            newname = Playlist::bump_name (_name, _session);
      }

      if ((playlist = boost::dynamic_pointer_cast<AudioPlaylist> (PlaylistFactory::create (_session, newname, hidden()))) != 0) {
            
            playlist->set_orig_diskstream_id (id());
            return use_playlist (playlist);

      } else { 
            return -1;
      }
}

int
AudioDiskstream::use_copy_playlist ()
{
      assert(audio_playlist());

      if (destructive()) {
            return 0;
      }

      if (_playlist == 0) {
            error << string_compose(_("AudioDiskstream %1: there is no existing playlist to make a copy of!"), _name) << endmsg;
            return -1;
      }

      string newname;
      boost::shared_ptr<AudioPlaylist> playlist;

      newname = Playlist::bump_name (_playlist->name(), _session);
      
      if ((playlist  = boost::dynamic_pointer_cast<AudioPlaylist>(PlaylistFactory::create (audio_playlist(), newname))) != 0) {
            playlist->set_orig_diskstream_id (id());
            return use_playlist (playlist);
      } else { 
            return -1;
      }
}

void
AudioDiskstream::setup_destructive_playlist ()
{
      SourceList srcs;
      boost::shared_ptr<ChannelList> c = channels.reader();
      
      for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {
            srcs.push_back ((*chan)->write_source);
      }

      /* a single full-sized region */

      boost::shared_ptr<Region> region (RegionFactory::create (srcs, 0, max_frames - srcs.front()->natural_position(), _name));
      _playlist->add_region (region, srcs.front()->natural_position());       
}

void
AudioDiskstream::use_destructive_playlist ()
{
      /* this is called from the XML-based constructor or ::set_destructive. when called,
         we already have a playlist and a region, but we need to
         set up our sources for write. we use the sources associated 
         with the (presumed single, full-extent) region.
      */

      boost::shared_ptr<Region> rp = _playlist->find_next_region (_session.current_start_frame(), Start, 1);

      if (!rp) {
            reset_write_sources (false, true);
            return;
      }

      boost::shared_ptr<AudioRegion> region = boost::dynamic_pointer_cast<AudioRegion> (rp);

      if (region == 0) {
            throw failed_constructor();
      }

      /* be sure to stretch the region out to the maximum length */

      region->set_length (max_frames - region->position(), this);

      uint32_t n;
      ChannelList::iterator chan;
      boost::shared_ptr<ChannelList> c = channels.reader();

      for (n = 0, chan = c->begin(); chan != c->end(); ++chan, ++n) {
            (*chan)->write_source = boost::dynamic_pointer_cast<AudioFileSource>(region->source (n));
            assert((*chan)->write_source);
            (*chan)->write_source->set_allow_remove_if_empty (false);

            /* this might be false if we switched modes, so force it */

            (*chan)->write_source->set_destructive (true);
      }

      /* the source list will never be reset for a destructive track */
}

void
AudioDiskstream::check_record_status (nframes_t transport_frame, nframes_t nframes, bool can_record)
{
      int possibly_recording;
      int rolling;
      int change;
      const int transport_rolling = 0x4;
      const int track_rec_enabled = 0x2;
      const int global_rec_enabled = 0x1;

      /* merge together the 3 factors that affect record status, and compute
         what has changed.
      */

      rolling = _session.transport_speed() != 0.0f;
      possibly_recording = (rolling << 2) | (record_enabled() << 1) | can_record;
      change = possibly_recording ^ last_possibly_recording;

      if (possibly_recording == last_possibly_recording) {
            return;
      }

      /* change state */

      /* if per-track or global rec-enable turned on while the other was already on, we've started recording */

      if ((change & track_rec_enabled) && record_enabled() && (!(change & global_rec_enabled) && can_record) || 
          ((change & global_rec_enabled) && can_record && (!(change & track_rec_enabled) && record_enabled()))) {
            
            /* starting to record: compute first+last frames */

            first_recordable_frame = transport_frame + _capture_offset;
            last_recordable_frame = max_frames;
            capture_start_frame = transport_frame;

            if (!(last_possibly_recording & transport_rolling) && (possibly_recording & transport_rolling)) {

                  /* was stopped, now rolling (and recording) */

                  if (_alignment_style == ExistingMaterial) {
                        first_recordable_frame += _session.worst_output_latency();
                  } else {
                        first_recordable_frame += _roll_delay;
                  }

            } else {

                  /* was rolling, but record state changed */

                  if (_alignment_style == ExistingMaterial) {

                        if (!Config->get_punch_in()) {

                              /* manual punch in happens at the correct transport frame
                                 because the user hit a button. but to get alignment correct 
                                 we have to back up the position of the new region to the 
                                 appropriate spot given the roll delay.
                              */

                              capture_start_frame -= _roll_delay;

                              /* XXX paul notes (august 2005): i don't know why
                                 this is needed.
                              */

                              first_recordable_frame += _capture_offset;

                        } else {

                              /* autopunch toggles recording at the precise
                                 transport frame, and then the DS waits
                                 to start recording for a time that depends
                                 on the output latency.
                              */

                              first_recordable_frame += _session.worst_output_latency();
                        }

                  } else {

                        if (Config->get_punch_in()) {
                              first_recordable_frame += _roll_delay;
                        } else {
                              capture_start_frame -= _roll_delay;
                        }
                  }
                  
            }

            if (recordable() && destructive()) {
                  boost::shared_ptr<ChannelList> c = channels.reader();
                  for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {
                        
                        RingBufferNPT<CaptureTransition>::rw_vector transvec;
                        (*chan)->capture_transition_buf->get_write_vector(&transvec);
                        
                        if (transvec.len[0] > 0) {
                              transvec.buf[0]->type = CaptureStart;
                              transvec.buf[0]->capture_val = capture_start_frame;
                              (*chan)->capture_transition_buf->increment_write_ptr(1);
                        }
                        else {
                              // bad!
                              fatal << X_("programming error: capture_transition_buf is full on rec start!  inconceivable!") 
                                    << endmsg;
                        }
                  }
            }

      } else if (!record_enabled() || !can_record) {
            
            /* stop recording */

            last_recordable_frame = transport_frame + _capture_offset;
            
            if (_alignment_style == ExistingMaterial) {
                  last_recordable_frame += _session.worst_output_latency();
            } else {
                  last_recordable_frame += _roll_delay;
            }
      }

      last_possibly_recording = possibly_recording;
}

int
AudioDiskstream::process (nframes_t transport_frame, nframes_t nframes, nframes_t offset, bool can_record, bool rec_monitors_input)
{
      uint32_t n;
      boost::shared_ptr<ChannelList> c = channels.reader();
      ChannelList::iterator chan;
      int ret = -1;
      nframes_t rec_offset = 0;
      nframes_t rec_nframes = 0;
      bool nominally_recording;
      bool re = record_enabled ();
      bool collect_playback = false;

      /* if we've already processed the frames corresponding to this call,
         just return. this allows multiple routes that are taking input
         from this diskstream to call our ::process() method, but have
         this stuff only happen once. more commonly, it allows both
         the AudioTrack that is using this AudioDiskstream *and* the Session
         to call process() without problems.
      */

      if (_processed) {
            return 0;
      }

      commit_should_unlock = false;

      if (!_io->active()) {
            _processed = true;
            return 0;
      }

      check_record_status (transport_frame, nframes, can_record);

      nominally_recording = (can_record && re);

      if (nframes == 0) {
            _processed = true;
            return 0;
      }

      /* This lock is held until the end of AudioDiskstream::commit, so these two functions
         must always be called as a pair. The only exception is if this function
         returns a non-zero value, in which case, ::commit should not be called.
      */

      // If we can't take the state lock return.
      if (!state_lock.trylock()) {
            return 1;
      } 
      commit_should_unlock = true;
      adjust_capture_position = 0;

      for (chan = c->begin(); chan != c->end(); ++chan) {
            (*chan)->current_capture_buffer = 0;
            (*chan)->current_playback_buffer  = 0;
      }

      if (nominally_recording || (_session.get_record_enabled() && Config->get_punch_in())) {
            OverlapType ot;
            
            // Safeguard against situations where process() goes haywire when autopunching and last_recordable_frame < first_recordable_frame
            if (last_recordable_frame < first_recordable_frame) {
                        last_recordable_frame = max_frames;

            }
            
            ot = coverage (first_recordable_frame, last_recordable_frame, transport_frame, transport_frame + nframes);

            switch (ot) {
            case OverlapNone:
                  rec_nframes = 0;
                  break;
                  
            case OverlapInternal:
            /*     ----------    recrange
                         |---|       transrange
            */
                  rec_nframes = nframes;
                  rec_offset = 0;
                  break;
                  
            case OverlapStart:
                  /*    |--------|    recrange
                            -----|          transrange
                  */
                  rec_nframes = transport_frame + nframes - first_recordable_frame;
                  if (rec_nframes) {
                        rec_offset = first_recordable_frame - transport_frame;
                  }
                  break;
                  
            case OverlapEnd:
                  /*    |--------|    recrange
                                 |--------  transrange
                  */
                  rec_nframes = last_recordable_frame - transport_frame;
                  rec_offset = 0;
                  break;
                  
            case OverlapExternal:
                  /*    |--------|    recrange
                            --------------  transrange
                  */
                  rec_nframes = last_recordable_frame - first_recordable_frame;
                  rec_offset = first_recordable_frame - transport_frame;
                  break;
            }

            if (rec_nframes && !was_recording) {
                  capture_captured = 0;
                  was_recording = true;
            }
      }


      if (can_record && !_last_capture_regions.empty()) {
            _last_capture_regions.clear ();
      }

      if (nominally_recording || rec_nframes) {

            uint32_t limit = _io->n_inputs ();

            /* one or more ports could already have been removed from _io, but our
               channel setup hasn't yet been updated. prevent us from trying to
               use channels that correspond to missing ports. note that the
               process callback (from which this is called) is always atomic
               with respect to port removal/addition.
            */

            for (n = 0, chan = c->begin(); chan != c->end() && n < limit; ++chan, ++n) {
                  
                  ChannelInfo* chaninfo (*chan);

                  chaninfo->capture_buf->get_write_vector (&chaninfo->capture_vector);

                  if (rec_nframes <= chaninfo->capture_vector.len[0]) {
                        
                        chaninfo->current_capture_buffer = chaninfo->capture_vector.buf[0];

                        /* note: grab the entire port buffer, but only copy what we were supposed to for recording, and use
                           rec_offset
                        */

                        memcpy (chaninfo->current_capture_buffer, _io->input(n)->get_buffer (rec_nframes) + offset + rec_offset, sizeof (Sample) * rec_nframes);

                  } else {

                        nframes_t total = chaninfo->capture_vector.len[0] + chaninfo->capture_vector.len[1];

                        if (rec_nframes > total) {
                              DiskOverrun ();
                              goto out;
                        }

                        Sample* buf = _io->input (n)->get_buffer (nframes) + offset;
                        nframes_t first = chaninfo->capture_vector.len[0];

                        memcpy (chaninfo->capture_wrap_buffer, buf, sizeof (Sample) * first);
                        memcpy (chaninfo->capture_vector.buf[0], buf, sizeof (Sample) * first);
                        memcpy (chaninfo->capture_wrap_buffer+first, buf + first, sizeof (Sample) * (rec_nframes - first));
                        memcpy (chaninfo->capture_vector.buf[1], buf + first, sizeof (Sample) * (rec_nframes - first));
                        
                        chaninfo->current_capture_buffer = chaninfo->capture_wrap_buffer;
                  }
            }

      } else {

            if (was_recording) {
                  finish_capture (rec_monitors_input, c);
            }

      }
      
      if (rec_nframes) {
            
            /* data will be written to disk */

            if (rec_nframes == nframes && rec_offset == 0) {

                  for (chan = c->begin(); chan != c->end(); ++chan) {
                        (*chan)->current_playback_buffer = (*chan)->current_capture_buffer;
                  }

                  playback_distance = nframes;

            } else {


                  /* we can't use the capture buffer as the playback buffer, because
                     we recorded only a part of the current process' cycle data
                     for capture.
                  */

                  collect_playback = true;
            }

            adjust_capture_position = rec_nframes;

      } else if (nominally_recording) {

            /* can't do actual capture yet - waiting for latency effects to finish before we start*/

            for (chan = c->begin(); chan != c->end(); ++chan) {
                  (*chan)->current_playback_buffer = (*chan)->current_capture_buffer;
            }

            playback_distance = nframes;

      } else {

            collect_playback = true;
      }

      if (collect_playback) {

            /* we're doing playback */

            nframes_t necessary_samples;

            /* no varispeed playback if we're recording, because the output .... TBD */

            if (rec_nframes == 0 && _actual_speed != 1.0f) {
                  necessary_samples = (nframes_t) floor ((nframes * fabs (_actual_speed))) + 1;
            } else {
                  necessary_samples = nframes;
            }
            
            for (chan = c->begin(); chan != c->end(); ++chan) {
                  (*chan)->playback_buf->get_read_vector (&(*chan)->playback_vector);
            }

            n = 0;                  

            for (chan = c->begin(); chan != c->end(); ++chan, ++n) {
                  
                  ChannelInfo* chaninfo (*chan);

                  if (necessary_samples <= chaninfo->playback_vector.len[0]) {

                        chaninfo->current_playback_buffer = chaninfo->playback_vector.buf[0];

                  } else {
                        nframes_t total = chaninfo->playback_vector.len[0] + chaninfo->playback_vector.len[1];
                        
                        if (necessary_samples > total) {
                              cerr << "underrun for " << _name << endl;
                              DiskUnderrun ();
                              goto out;
                              
                        } else {
                              
                              memcpy ((char *) chaninfo->playback_wrap_buffer, chaninfo->playback_vector.buf[0],
                                    chaninfo->playback_vector.len[0] * sizeof (Sample));
                              memcpy (chaninfo->playback_wrap_buffer + chaninfo->playback_vector.len[0], chaninfo->playback_vector.buf[1], 
                                    (necessary_samples - chaninfo->playback_vector.len[0]) * sizeof (Sample));
                              
                              chaninfo->current_playback_buffer = chaninfo->playback_wrap_buffer;
                        }
                  }
            } 

            if (rec_nframes == 0 && _actual_speed != 1.0f && _actual_speed != -1.0f) {
                  
                  uint64_t phase = last_phase;
                  int64_t phi_delta;
                  nframes_t i = 0;

                  // Linearly interpolate into the alt buffer
                  // using 40.24 fixp maths (swh)

                  if (phi != target_phi) {
                        phi_delta = ((int64_t)(target_phi - phi)) / nframes;
                  } else {
                        phi_delta = 0;
                  }

                  for (chan = c->begin(); chan != c->end(); ++chan) {

                        float fr;
                        ChannelInfo* chaninfo (*chan);

                        i = 0;
                        phase = last_phase;

                        for (nframes_t outsample = 0; outsample < nframes; ++outsample) {
                              i = phase >> 24;
                              fr = (phase & 0xFFFFFF) / 16777216.0f;
                              chaninfo->speed_buffer[outsample] = 
                                    chaninfo->current_playback_buffer[i] * (1.0f - fr) +
                                    chaninfo->current_playback_buffer[i+1] * fr;
                              phase += phi + phi_delta;
                        }
                        
                        chaninfo->current_playback_buffer = chaninfo->speed_buffer;
                  }

                  playback_distance = i; // + 1;
                  last_phase = (phase & 0xFFFFFF);

            } else {
                  playback_distance = nframes;
            }

            phi = target_phi;
      }

      ret = 0;

  out:
      _processed = true;

      if (ret) {

            /* we're exiting with failure, so ::commit will not
               be called. unlock the state lock.
            */
            
            commit_should_unlock = false;
            state_lock.unlock();
      } 

      return ret;
}

bool
AudioDiskstream::commit (nframes_t nframes)
{
      bool need_butler = false;

      if (!_io->active()) {
            return false;
      }

      if (_actual_speed < 0.0) {
            playback_sample -= playback_distance;
      } else {
            playback_sample += playback_distance;
      }

      boost::shared_ptr<ChannelList> c = channels.reader();
      for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {

            (*chan)->playback_buf->increment_read_ptr (playback_distance);
            
            if (adjust_capture_position) {
                  (*chan)->capture_buf->increment_write_ptr (adjust_capture_position);
            }
      }
      
      if (adjust_capture_position != 0) {
            capture_captured += adjust_capture_position;
            adjust_capture_position = 0;
      }
      
      if (_slaved) {
            if (_io && _io->active()) {
                  need_butler = c->front()->playback_buf->write_space() >= c->front()->playback_buf->bufsize() / 2;
            } else {
                  need_butler = false;
            }
      } else {
            if (_io && _io->active()) {
                  need_butler = c->front()->playback_buf->write_space() >= disk_io_chunk_frames
                        || c->front()->capture_buf->read_space() >= disk_io_chunk_frames;
            } else {
                  need_butler = c->front()->capture_buf->read_space() >= disk_io_chunk_frames;
            }
      }

      if (commit_should_unlock) {
            state_lock.unlock();
      }

      _processed = false;

      return need_butler;
}

void
AudioDiskstream::set_pending_overwrite (bool yn)
{
      /* called from audio thread, so we can use the read ptr and playback sample as we wish */
      
      pending_overwrite = yn;

      overwrite_frame = playback_sample;
      overwrite_offset = channels.reader()->front()->playback_buf->get_read_ptr();
}

int
AudioDiskstream::overwrite_existing_buffers ()
{
      boost::shared_ptr<ChannelList> c = channels.reader();
      Sample* mixdown_buffer;
      float* gain_buffer;
      int ret = -1;
      bool reversed = (_visible_speed * _session.transport_speed()) < 0.0f;

      overwrite_queued = false;

      /* assume all are the same size */
      nframes_t size = c->front()->playback_buf->bufsize();
      
      mixdown_buffer = new Sample[size];
      gain_buffer = new float[size];
      
      /* reduce size so that we can fill the buffer correctly. */
      size--;
      
      uint32_t n=0;
      nframes_t start;

      for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan, ++n) {

            start = overwrite_frame;
            nframes_t cnt = size;
            
            /* to fill the buffer without resetting the playback sample, we need to
               do it one or two chunks (normally two).

               |----------------------------------------------------------------------|

                                   ^
                               overwrite_offset
                |<- second chunk->||<----------------- first chunk ------------------>|
               
            */
            
            nframes_t to_read = size - overwrite_offset;

            if (read ((*chan)->playback_buf->buffer() + overwrite_offset, mixdown_buffer, gain_buffer, start, to_read, *chan, n, reversed)) {
                  error << string_compose(_("AudioDiskstream %1: when refilling, cannot read %2 from playlist at frame %3"),
                               _id, size, playback_sample) << endmsg;
                  goto out;
            }
                  
            if (cnt > to_read) {

                  cnt -= to_read;
            
                  if (read ((*chan)->playback_buf->buffer(), mixdown_buffer, gain_buffer,
                          start, cnt, *chan, n, reversed)) {
                        error << string_compose(_("AudioDiskstream %1: when refilling, cannot read %2 from playlist at frame %3"),
                                     _id, size, playback_sample) << endmsg;
                        goto out;
                  }
            }
      }

      ret = 0;
 
  out:
      pending_overwrite = false;
      delete [] gain_buffer;
      delete [] mixdown_buffer;
      return ret;
}

int
AudioDiskstream::seek (nframes_t frame, bool complete_refill)
{
      uint32_t n;
      int ret;
      ChannelList::iterator chan;
      boost::shared_ptr<ChannelList> c = channels.reader();

      Glib::Mutex::Lock lm (state_lock);
      
      for (n = 0, chan = c->begin(); chan != c->end(); ++chan, ++n) {
            (*chan)->playback_buf->reset ();
            (*chan)->capture_buf->reset ();
      }
      
      /* can't rec-enable in destructive mode if transport is before start */
      
      if (destructive() && record_enabled() && frame < _session.current_start_frame()) {
            disengage_record_enable ();
      }
      
      playback_sample = frame;
      file_frame = frame;
      
      if (complete_refill) {
            while ((ret = do_refill_with_alloc ()) > 0) ;
      } else {
            ret = do_refill_with_alloc ();
      }

      return ret;
}

int
AudioDiskstream::can_internal_playback_seek (nframes_t distance)
{
      ChannelList::iterator chan;
      boost::shared_ptr<ChannelList> c = channels.reader();

      for (chan = c->begin(); chan != c->end(); ++chan) {
            if ((*chan)->playback_buf->read_space() < distance) {
                  return false;
            } 
      }
      return true;
}

int
AudioDiskstream::internal_playback_seek (nframes_t distance)
{
      ChannelList::iterator chan;
      boost::shared_ptr<ChannelList> c = channels.reader();

      for (chan = c->begin(); chan != c->end(); ++chan) {
            (*chan)->playback_buf->increment_read_ptr (distance);
      }

      first_recordable_frame += distance;
      playback_sample += distance;
      
      return 0;
}

int
AudioDiskstream::read (Sample* buf, Sample* mixdown_buffer, float* gain_buffer, nframes_t& start, nframes_t cnt, 
                   ChannelInfo* channel_info, int channel, bool reversed)
{
      nframes_t this_read = 0;
      bool reloop = false;
      nframes_t loop_end = 0;
      nframes_t loop_start = 0;
      nframes_t loop_length = 0;
      nframes_t offset = 0;
      nframes_t xfade_samples = 0;
      Sample    xfade_buf[128];
      Location *loc = 0;

      /* XXX we don't currently play loops in reverse. not sure why */

      if (!reversed) {

            /* Make the use of a Location atomic for this read operation.
               
               Note: Locations don't get deleted, so all we care about
               when I say "atomic" is that we are always pointing to
               the same one and using a start/length values obtained
               just once.
            */
            
            if ((loc = loop_location) != 0) {
                  loop_start = loc->start();
                  loop_end = loc->end();
                  loop_length = loop_end - loop_start;
            }
            
            /* if we are looping, ensure that the first frame we read is at the correct
               position within the loop.
            */
            
            if (loc && start >= loop_end) {
                  //cerr << "start adjusted from " << start;
                  start = loop_start + ((start - loop_start) % loop_length);
                  //cerr << "to " << start << endl;
            }

            //cerr << "start is " << start << "  loopstart: " << loop_start << "  loopend: " << loop_end << endl;
      }

      while (cnt) {

            if (reversed) {
                  start -= cnt;
            }
                  
            /* take any loop into account. we can't read past the end of the loop. */

            if (loc && (loop_end - start < cnt)) {
                  this_read = loop_end - start;
                  //cerr << "reloop true: thisread: " << this_read << "  cnt: " << cnt << endl;
                  reloop = true;
            } else {
                  reloop = false;
                  this_read = cnt;
            }

            if (this_read == 0) {
                  break;
            }

            this_read = min(cnt,this_read);

            if (audio_playlist()->read (buf+offset, mixdown_buffer, gain_buffer, start, this_read, channel) != this_read) {
                  error << string_compose(_("AudioDiskstream %1: cannot read %2 from playlist at frame %3"), _id, this_read, 
                               start) << endmsg;
                  return -1;
            }

            // xfade loop boundary if appropriate

            if (xfade_samples > 0) {
                  // just do a linear xfade for this short bit

                  xfade_samples = min(xfade_samples, this_read);

                  float delta = 1.0f / xfade_samples;
                  float scale = 0.0f;
                  Sample * tmpbuf = buf+offset;

                  for (size_t i=0; i < xfade_samples; ++i) {
                        *tmpbuf = (*tmpbuf * scale) + xfade_buf[i]*(1.0f - scale);
                        ++tmpbuf;
                        scale += delta; 
                  }

                  xfade_samples = 0;
            }

            _read_data_count = _playlist->read_data_count();
            
            if (reversed) {

                  swap_by_ptr (buf, buf + this_read - 1);
                  
            } else {
                  start += this_read;
            
                  /* if we read to the end of the loop, go back to the beginning */
                  
                  if (reloop) {
                        // read crossfade samples to apply to the next read

                        xfade_samples = min((nframes_t) 128, cnt-this_read);

                        if (audio_playlist()->read (xfade_buf, mixdown_buffer, gain_buffer, start, xfade_samples, channel) != xfade_samples) {
                              error << string_compose(_("AudioDiskstream %1: cannot read xfade samples %2 from playlist at frame %3"), 
                                                _id, xfade_samples,start) << endmsg;
                              memset(xfade_buf, 0, xfade_samples * sizeof(Sample)); // just in case
                        }

                        start = loop_start;
                  }
            } 

            cnt -= this_read;
            offset += this_read;
      }

      return 0;
}

int
AudioDiskstream::do_refill_with_alloc ()
{
      Sample* mix_buf  = new Sample[disk_io_chunk_frames];
      float*  gain_buf = new float[disk_io_chunk_frames];

      int ret = _do_refill(mix_buf, gain_buf);
      
      delete [] mix_buf;
      delete [] gain_buf;

      return ret;
}

int
AudioDiskstream::_do_refill (Sample* mixdown_buffer, float* gain_buffer)
{
      int32_t ret = 0;
      nframes_t to_read;
      RingBufferNPT<Sample>::rw_vector vector;
      bool reversed = (_visible_speed * _session.transport_speed()) < 0.0f;
      nframes_t total_space;
      nframes_t zero_fill;
      uint32_t chan_n;
      ChannelList::iterator i;
      boost::shared_ptr<ChannelList> c = channels.reader();
      nframes_t ts;

      if (c->empty()) {
            return 0;
      }

      assert(mixdown_buffer);
      assert(gain_buffer);

      vector.buf[0] = 0;
      vector.len[0] = 0;
      vector.buf[1] = 0;
      vector.len[1] = 0;

      c->front()->playback_buf->get_write_vector (&vector);
      
      if ((total_space = vector.len[0] + vector.len[1]) == 0) {
            return 0;
      }

      /* if there are 2+ chunks of disk i/o possible for
         this track, let the caller know so that it can arrange
         for us to be called again, ASAP.
      */
      
      if (total_space >= (_slaved?3:2) * disk_io_chunk_frames) {
            ret = 1;
      }
      
      /* if we're running close to normal speed and there isn't enough 
         space to do disk_io_chunk_frames of I/O, then don't bother.  
         
         at higher speeds, just do it because the sync between butler
         and audio thread may not be good enough.
      */
      
      if ((total_space < disk_io_chunk_frames) && fabs (_actual_speed) < 2.0f) {
            return 0;
      }
      
      /* when slaved, don't try to get too close to the read pointer. this
         leaves space for the buffer reversal to have something useful to
         work with.
      */
      
      if (_slaved && total_space < (c->front()->playback_buf->bufsize() / 2)) {
            return 0;
      }

      /* never do more than disk_io_chunk_frames worth of disk input per call (limit doesn't apply for memset) */

      total_space = min (disk_io_chunk_frames, total_space);

      if (reversed) {

            if (file_frame == 0) {

                  /* at start: nothing to do but fill with silence */

                  for (chan_n = 0, i = c->begin(); i != c->end(); ++i, ++chan_n) {
                              
                        ChannelInfo* chan (*i);
                        chan->playback_buf->get_write_vector (&vector);
                        memset (vector.buf[0], 0, sizeof(Sample) * vector.len[0]);
                        if (vector.len[1]) {
                              memset (vector.buf[1], 0, sizeof(Sample) * vector.len[1]);
                        }
                        chan->playback_buf->increment_write_ptr (vector.len[0] + vector.len[1]);
                  }
                  return 0;
            }

            if (file_frame < total_space) {

                  /* too close to the start: read what we can, 
                     and then zero fill the rest 
                  */

                  zero_fill = total_space - file_frame;
                  total_space = file_frame;
                  file_frame = 0;

            } else {
                  
                  zero_fill = 0;
            }

      } else {

            if (file_frame == max_frames) {

                  /* at end: nothing to do but fill with silence */
                  
                  for (chan_n = 0, i = c->begin(); i != c->end(); ++i, ++chan_n) {
                              
                        ChannelInfo* chan (*i);
                        chan->playback_buf->get_write_vector (&vector);
                        memset (vector.buf[0], 0, sizeof(Sample) * vector.len[0]);
                        if (vector.len[1]) {
                              memset (vector.buf[1], 0, sizeof(Sample) * vector.len[1]);
                        }
                        chan->playback_buf->increment_write_ptr (vector.len[0] + vector.len[1]);
                  }
                  return 0;
            }
            
            if (file_frame > max_frames - total_space) {

                  /* to close to the end: read what we can, and zero fill the rest */

                  zero_fill = total_space - (max_frames - file_frame);
                  total_space = max_frames - file_frame;

            } else {
                  zero_fill = 0;
            }
      }
      
      nframes_t file_frame_tmp = 0;

      for (chan_n = 0, i = c->begin(); i != c->end(); ++i, ++chan_n) {

            ChannelInfo* chan (*i);
            Sample* buf1;
            Sample* buf2;
            nframes_t len1, len2;

            chan->playback_buf->get_write_vector (&vector);

            if (vector.len[0] > disk_io_chunk_frames) {
                  
                  /* we're not going to fill the first chunk, so certainly do not bother with the
                     other part. it won't be connected with the part we do fill, as in:
                     
                     .... => writable space
                     ++++ => readable space
                     ^^^^ => 1 x disk_io_chunk_frames that would be filled
                     
                     |......|+++++++++++++|...............................|
                     buf1                buf0
                                          ^^^^^^^^^^^^^^^
                     
                     
                     So, just pretend that the buf1 part isn't there.                           
                     
                  */
            
                  vector.buf[1] = 0;
                  vector.len[1] = 0;
            
            } 

            ts = total_space;
            file_frame_tmp = file_frame;

            buf1 = vector.buf[0];
            len1 = vector.len[0];
            buf2 = vector.buf[1];
            len2 = vector.len[1];

            to_read = min (ts, len1);
            to_read = min (to_read, disk_io_chunk_frames);

            if (to_read) {

                  if (read (buf1, mixdown_buffer, gain_buffer, file_frame_tmp, to_read, chan, chan_n, reversed)) {
                        ret = -1;
                        goto out;
                  }

                  chan->playback_buf->increment_write_ptr (to_read);
                  ts -= to_read;
            }

            to_read = min (ts, len2);

            if (to_read) {

                  /* we read all of vector.len[0], but it wasn't an entire disk_io_chunk_frames of data,
                     so read some or all of vector.len[1] as well.
                  */

                  if (read (buf2, mixdown_buffer, gain_buffer, file_frame_tmp, to_read, chan, chan_n, reversed)) {
                        ret = -1;
                        goto out;
                  }
            
                  chan->playback_buf->increment_write_ptr (to_read);
            }

            if (zero_fill) {
                  /* do something */
            }

      }
      
      file_frame = file_frame_tmp;

  out:

      return ret;
}     

/** Flush pending data to disk.
 *
 * Important note: this function will write *AT MOST* disk_io_chunk_frames
 * of data to disk. it will never write more than that.  If it writes that
 * much and there is more than that waiting to be written, it will return 1,
 * otherwise 0 on success or -1 on failure.
 * 
 * If there is less than disk_io_chunk_frames to be written, no data will be
 * written at all unless @a force_flush is true.
 */
int
AudioDiskstream::do_flush (Session::RunContext context, bool force_flush)
{
      uint32_t to_write;
      int32_t ret = 0;
      RingBufferNPT<Sample>::rw_vector vector;
      RingBufferNPT<CaptureTransition>::rw_vector transvec;
      nframes_t total;

      _write_data_count = 0;

      transvec.buf[0] = 0;
      transvec.buf[1] = 0;
      vector.buf[0] = 0;
      vector.buf[1] = 0;

      boost::shared_ptr<ChannelList> c = channels.reader();
      for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {
      
            (*chan)->capture_buf->get_read_vector (&vector);

            total = vector.len[0] + vector.len[1];

            if (total == 0 || (total < disk_io_chunk_frames && !force_flush && was_recording)) {
                  goto out;
            }

            /* if there are 2+ chunks of disk i/o possible for
               this track, let the caller know so that it can arrange
               for us to be called again, ASAP.
               
               if we are forcing a flush, then if there is* any* extra
               work, let the caller know.

               if we are no longer recording and there is any extra work,
               let the caller know too.
            */

            if (total >= 2 * disk_io_chunk_frames || ((force_flush || !was_recording) && total > disk_io_chunk_frames)) {
                  ret = 1;
            } 

            to_write = min (disk_io_chunk_frames, (nframes_t) vector.len[0]);
            
            // check the transition buffer when recording destructive
            // important that we get this after the capture buf

            if (destructive()) {
                  (*chan)->capture_transition_buf->get_read_vector(&transvec);
                  size_t transcount = transvec.len[0] + transvec.len[1];
                  bool have_start = false;
                  size_t ti;

                  for (ti=0; ti < transcount; ++ti) {
                        CaptureTransition & captrans = (ti < transvec.len[0]) ? transvec.buf[0][ti] : transvec.buf[1][ti-transvec.len[0]];
                        
                        if (captrans.type == CaptureStart) {
                              // by definition, the first data we got above represents the given capture pos

                              (*chan)->write_source->mark_capture_start (captrans.capture_val);
                              (*chan)->curr_capture_cnt = 0;

                              have_start = true;
                        }
                        else if (captrans.type == CaptureEnd) {

                              // capture end, the capture_val represents total frames in capture

                              if (captrans.capture_val <= (*chan)->curr_capture_cnt + to_write) {

                                    // shorten to make the write a perfect fit
                                    uint32_t nto_write = (captrans.capture_val - (*chan)->curr_capture_cnt); 

                                    if (nto_write < to_write) {
                                          ret = 1; // should we?
                                    }
                                    to_write = nto_write;

                                    (*chan)->write_source->mark_capture_end ();
                                    
                                    // increment past this transition, but go no further
                                    ++ti;
                                    break;
                              }
                              else {
                                    // actually ends just beyond this chunk, so force more work
                                    ret = 1;
                                    break;
                              }
                        }
                  }

                  if (ti > 0) {
                        (*chan)->capture_transition_buf->increment_read_ptr(ti);
                  }
            }

            if ((!(*chan)->write_source) || (*chan)->write_source->write (vector.buf[0], to_write) != to_write) {
                  error << string_compose(_("AudioDiskstream %1: cannot write to disk"), _id) << endmsg;
                  return -1;
            }

            (*chan)->capture_buf->increment_read_ptr (to_write);
            (*chan)->curr_capture_cnt += to_write;
            
            if ((to_write == vector.len[0]) && (total > to_write) && (to_write < disk_io_chunk_frames) && !destructive()) {
            
                  /* we wrote all of vector.len[0] but it wasn't an entire
                     disk_io_chunk_frames of data, so arrange for some part 
                     of vector.len[1] to be flushed to disk as well.
                  */
                  
                  to_write = min ((nframes_t)(disk_io_chunk_frames - to_write), (nframes_t) vector.len[1]);

                  if ((*chan)->write_source->write (vector.buf[1], to_write) != to_write) {
                        error << string_compose(_("AudioDiskstream %1: cannot write to disk"), _id) << endmsg;
                        return -1;
                  }

                  _write_data_count += (*chan)->write_source->write_data_count();
      
                  (*chan)->capture_buf->increment_read_ptr (to_write);
                  (*chan)->curr_capture_cnt += to_write;
            }
      }

  out:
      return ret;
}

void
AudioDiskstream::transport_stopped (struct tm& when, time_t twhen, bool abort_capture)
{
      uint32_t buffer_position;
      bool more_work = true;
      int err = 0;
      boost::shared_ptr<AudioRegion> region;
      nframes_t total_capture;
      SourceList srcs;
      SourceList::iterator src;
      ChannelList::iterator chan;
      vector<CaptureInfo*>::iterator ci;
      boost::shared_ptr<ChannelList> c = channels.reader();
      uint32_t n = 0; 
      bool mark_write_completed = false;

      finish_capture (true, c);

      /* butler is already stopped, but there may be work to do 
         to flush remaining data to disk.
      */

      while (more_work && !err) {
            switch (do_flush (Session::TransportContext, true)) {
            case 0:
                  more_work = false;
                  break;
            case 1:
                  break;
            case -1:
                  error << string_compose(_("AudioDiskstream \"%1\": cannot flush captured data to disk!"), _name) << endmsg;
                  err++;
            }
      }

      /* XXX is there anything we can do if err != 0 ? */
      Glib::Mutex::Lock lm (capture_info_lock);
      
      if (capture_info.empty()) {
            return;
      }

      if (abort_capture) {
            
            if (destructive()) {
                  goto outout;
            }

            for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {

                  if ((*chan)->write_source) {
                        
                        (*chan)->write_source->mark_for_remove ();
                        (*chan)->write_source->drop_references ();
                        (*chan)->write_source.reset ();
                  }
                  
                  /* new source set up in "out" below */
            }

            goto out;
      } 

      for (total_capture = 0, ci = capture_info.begin(); ci != capture_info.end(); ++ci) {
            total_capture += (*ci)->frames;
      }

      /* figure out the name for this take */

      for (n = 0, chan = c->begin(); chan != c->end(); ++chan, ++n) {

            boost::shared_ptr<AudioFileSource> s = (*chan)->write_source;
            
            if (s) {
                  srcs.push_back (s);
                  s->update_header (capture_info.front()->start, when, twhen);
                  s->set_captured_for (_name);
                  s->mark_immutable ();
                  if (Config->get_auto_analyse_audio()) {
                        Analyser::queue_source_for_analysis (s, true);
                  }
            }
      }

      /* destructive tracks have a single, never changing region */

      if (destructive()) {

            /* send a signal that any UI can pick up to do the right thing. there is 
               a small problem here in that a UI may need the peak data to be ready
               for the data that was recorded and this isn't interlocked with that
               process. this problem is deferred to the UI.
             */
            
            _playlist->Modified();

      } else {

            string whole_file_region_name;
            whole_file_region_name = region_name_from_path (c->front()->write_source->name(), true);

            /* Register a new region with the Session that
               describes the entire source. Do this first
               so that any sub-regions will obviously be
               children of this one (later!)
            */
            
            try {
                  boost::shared_ptr<Region> rx (RegionFactory::create (srcs, c->front()->write_source->last_capture_start_frame(), total_capture, 
                                                           whole_file_region_name,
                                                           0, AudioRegion::Flag (AudioRegion::DefaultFlags|AudioRegion::Automatic|AudioRegion::WholeFile)));

                  region = boost::dynamic_pointer_cast<AudioRegion> (rx);
                  region->special_set_position (capture_info.front()->start);
            }
            
            
            catch (failed_constructor& err) {
                  error << string_compose(_("%1: could not create region for complete audio file"), _name) << endmsg;
                  /* XXX what now? */
            }
            
            _last_capture_regions.push_back (region);

            // cerr << _name << ": there are " << capture_info.size() << " capture_info records\n";
            
                XMLNode &before = _playlist->get_state();
            _playlist->freeze ();
            
            for (buffer_position = c->front()->write_source->last_capture_start_frame(), ci = capture_info.begin(); ci != capture_info.end(); ++ci) {
                  
                  string region_name;

                  _session.region_name (region_name, whole_file_region_name, false);
                  
                  // cerr << _name << ": based on ci of " << (*ci)->start << " for " << (*ci)->frames << " add region " << region_name << endl;
                  
                  try {
                        boost::shared_ptr<Region> rx (RegionFactory::create (srcs, buffer_position, (*ci)->frames, region_name));
                        region = boost::dynamic_pointer_cast<AudioRegion> (rx);
                  }
                  
                  catch (failed_constructor& err) {
                        error << _("AudioDiskstream: could not create region for captured audio!") << endmsg;
                        continue; /* XXX is this OK? */
                  }
                  
                  region->GoingAway.connect (bind (mem_fun (*this, &Diskstream::remove_region_from_last_capture), boost::weak_ptr<Region>(region)));
                  
                  _last_capture_regions.push_back (region);
                  
                  i_am_the_modifier++;
                  _playlist->add_region (region, (*ci)->start);
                  i_am_the_modifier--;
                  
                  buffer_position += (*ci)->frames;
            }

            _playlist->thaw ();
                XMLNode &after = _playlist->get_state();
            _session.add_command (new MementoCommand<Playlist>(*_playlist, &before, &after));
      }

      mark_write_completed = true;

  out:
      reset_write_sources (mark_write_completed);

  outout:

      for (ci = capture_info.begin(); ci != capture_info.end(); ++ci) {
            delete *ci;
      }

      capture_info.clear ();
      capture_start_frame = 0;
}

void
AudioDiskstream::transport_looped (nframes_t transport_frame)
{
      if (was_recording) {
            // all we need to do is finish this capture, with modified capture length
            boost::shared_ptr<ChannelList> c = channels.reader();

            // adjust the capture length knowing that the data will be recorded to disk
            // only necessary after the first loop where we're recording
            if (capture_info.size() == 0) {
                  capture_captured += _capture_offset;

                  if (_alignment_style == ExistingMaterial) {
                        capture_captured += _session.worst_output_latency();
                  } else {
                        capture_captured += _roll_delay;
                  }
            }

            finish_capture (true, c);

            // the next region will start recording via the normal mechanism
            // we'll set the start position to the current transport pos
            // no latency adjustment or capture offset needs to be made, as that already happened the first time
            capture_start_frame = transport_frame;
            first_recordable_frame = transport_frame; // mild lie
            last_recordable_frame = max_frames;
            was_recording = true;

            if (recordable() && destructive()) {
                  for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {
                        
                        RingBufferNPT<CaptureTransition>::rw_vector transvec;
                        (*chan)->capture_transition_buf->get_write_vector(&transvec);
                        
                        if (transvec.len[0] > 0) {
                              transvec.buf[0]->type = CaptureStart;
                              transvec.buf[0]->capture_val = capture_start_frame;
                              (*chan)->capture_transition_buf->increment_write_ptr(1);
                        }
                        else {
                              // bad!
                              fatal << X_("programming error: capture_transition_buf is full on rec loop!  inconceivable!") 
                                    << endmsg;
                        }
                  }
            }

      }
}

void
AudioDiskstream::finish_capture (bool rec_monitors_input, boost::shared_ptr<ChannelList> c)
{
      was_recording = false;
      
      if (capture_captured == 0) {
            return;
      }

      if (recordable() && destructive()) {
            for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {
                  
                  RingBufferNPT<CaptureTransition>::rw_vector transvec;
                  (*chan)->capture_transition_buf->get_write_vector(&transvec);
                  
                  if (transvec.len[0] > 0) {
                        transvec.buf[0]->type = CaptureEnd;
                        transvec.buf[0]->capture_val = capture_captured;
                        (*chan)->capture_transition_buf->increment_write_ptr(1);
                  }
                  else {
                        // bad!
                        fatal << string_compose (_("programmer error: %1"), X_("capture_transition_buf is full when stopping record!  inconceivable!")) << endmsg;
                  }
            }
      }
      
      
      CaptureInfo* ci = new CaptureInfo;
      
      ci->start =  capture_start_frame;
      ci->frames = capture_captured;
      
      /* XXX theoretical race condition here. Need atomic exchange ? 
         However, the circumstances when this is called right 
         now (either on record-disable or transport_stopped)
         mean that no actual race exists. I think ...
         We now have a capture_info_lock, but it is only to be used
         to synchronize in the transport_stop and the capture info
         accessors, so that invalidation will not occur (both non-realtime).
      */

      // cerr << "Finish capture, add new CI, " << ci->start << '+' << ci->frames << endl;

      capture_info.push_back (ci);
      capture_captured = 0;
}

void
AudioDiskstream::set_record_enabled (bool yn)
{
      if (!recordable() || !_session.record_enabling_legal() || _io->n_inputs() == 0) {
            return;
      }

      /* can't rec-enable in destructive mode if transport is before start */

      if (destructive() && yn && _session.transport_frame() < _session.current_start_frame()) {
            return;
      }

      if (yn && channels.reader()->front()->source == 0) {

            /* pick up connections not initiated *from* the IO object
               we're associated with.
            */

            get_input_sources ();
      }

      /* yes, i know that this not proof against race conditions, but its
         good enough. i think.
      */

      if (record_enabled() != yn) {
            if (yn) {
                  engage_record_enable ();
            } else {
                  disengage_record_enable ();
            }
      }
}

void
AudioDiskstream::engage_record_enable ()
{
      bool rolling = _session.transport_speed() != 0.0f;
      boost::shared_ptr<ChannelList> c = channels.reader();

      g_atomic_int_set (&_record_enabled, 1);
      capturing_sources.clear ();

      if (Config->get_monitoring_model() == HardwareMonitoring) {

            for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {
                  if ((*chan)->source) {
                        (*chan)->source->ensure_monitor_input (!(Config->get_auto_input() && rolling));
                  }
                  capturing_sources.push_back ((*chan)->write_source);
            }
            
      } else {
            for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {
                  capturing_sources.push_back ((*chan)->write_source);
            }
      }

      RecordEnableChanged (); /* EMIT SIGNAL */
}

void
AudioDiskstream::disengage_record_enable ()
{
      g_atomic_int_set (&_record_enabled, 0);
      boost::shared_ptr<ChannelList> c = channels.reader();
      if (Config->get_monitoring_model() == HardwareMonitoring) {
            for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {
                  if ((*chan)->source) {
                        (*chan)->source->ensure_monitor_input (false);
                  }
            }
      }
      capturing_sources.clear ();
      RecordEnableChanged (); /* EMIT SIGNAL */
}

XMLNode&
AudioDiskstream::get_state ()
{
      XMLNode* node = new XMLNode ("AudioDiskstream");
      char buf[64] = "";
      LocaleGuard lg (X_("POSIX"));
      boost::shared_ptr<ChannelList> c = channels.reader();

      node->add_property ("flags", enum_2_string (_flags));

      snprintf (buf, sizeof(buf), "%zd", c->size());
      node->add_property ("channels", buf);

      node->add_property ("playlist", _playlist->name());
      
      snprintf (buf, sizeof(buf), "%.12g", _visible_speed);
      node->add_property ("speed", buf);

      node->add_property("name", _name);
      id().print (buf, sizeof (buf));
      node->add_property("id", buf);

      if (!capturing_sources.empty() && _session.get_record_enabled()) {

            XMLNode* cs_child = new XMLNode (X_("CapturingSources"));
            XMLNode* cs_grandchild;

            for (vector<boost::shared_ptr<AudioFileSource> >::iterator i = capturing_sources.begin(); i != capturing_sources.end(); ++i) {
                  cs_grandchild = new XMLNode (X_("file"));
                  cs_grandchild->add_property (X_("path"), (*i)->path());
                  cs_child->add_child_nocopy (*cs_grandchild);
            }

            /* store the location where capture will start */

            Location* pi;

            if (Config->get_punch_in() && ((pi = _session.locations()->auto_punch_location()) != 0)) {
                  snprintf (buf, sizeof (buf), "%" PRIu32, pi->start());
            } else {
                  snprintf (buf, sizeof (buf), "%" PRIu32, _session.transport_frame());
            }

            cs_child->add_property (X_("at"), buf);
            node->add_child_nocopy (*cs_child);
      }

      if (_extra_xml) {
            node->add_child_copy (*_extra_xml);
      }

      return* node;
}

int
AudioDiskstream::set_state (const XMLNode& node)
{
      const XMLProperty* prop;
      XMLNodeList nlist = node.children();
      XMLNodeIterator niter;
      uint32_t nchans = 1;
      XMLNode* capture_pending_node = 0;
      LocaleGuard lg (X_("POSIX"));

      in_set_state = true;

      for (niter = nlist.begin(); niter != nlist.end(); ++niter) {
            if ((*niter)->name() == IO::state_node_name) {
                  deprecated_io_node = new XMLNode (**niter);
            }

            if ((*niter)->name() == X_("CapturingSources")) {
                  capture_pending_node = *niter;
            }
      }

      /* prevent write sources from being created */
      
      in_set_state = true;
      
      if ((prop = node.property ("name")) != 0) {
            _name = prop->value();
      } 

      if (deprecated_io_node) {
            if ((prop = deprecated_io_node->property ("id")) != 0) {
                  _id = prop->value ();
            }
      } else {
            if ((prop = node.property ("id")) != 0) {
                  _id = prop->value ();
            }
      }

      if ((prop = node.property ("flags")) != 0) {
            _flags = Flag (string_2_enum (prop->value(), _flags));
      }

      if ((prop = node.property ("channels")) != 0) {
            nchans = atoi (prop->value().c_str());
      }
      
      // create necessary extra channels
      // we are always constructed with one and we always need one

      _n_channels = channels.reader()->size();
      
      if (nchans > _n_channels) {

            add_channel (nchans - _n_channels);
            IO::MoreOutputs(_n_channels);

      } else if (nchans < _n_channels) {

            remove_channel (_n_channels - nchans);
      }

      if ((prop = node.property ("playlist")) == 0) {
            return -1;
      }

      {
            bool had_playlist = (_playlist != 0);
      
            if (find_and_use_playlist (prop->value())) {
                  return -1;
            }

            if (!had_playlist) {
                  _playlist->set_orig_diskstream_id (_id);
            }
            
            if (!destructive() && capture_pending_node) {
                  /* destructive streams have one and only one source per channel,
                     and so they never end up in pending capture in any useful
                     sense.
                  */
                  use_pending_capture_data (*capture_pending_node);
            }

      }

      if ((prop = node.property ("speed")) != 0) {
            double sp = atof (prop->value().c_str());

            if (realtime_set_speed (sp, false)) {
                  non_realtime_set_speed ();
            }
      }

      in_set_state = false;

      /* make sure this is clear before we do anything else */

      capturing_sources.clear ();

      /* write sources are handled when we handle the input set 
         up of the IO that owns this DS (::non_realtime_input_change())
      */
            
      return 0;
}

int
AudioDiskstream::use_new_write_source (uint32_t n)
{
      boost::shared_ptr<ChannelList> c = channels.reader();

      if (!recordable()) {
            return 1;
      }

      if (n >= c->size()) {
            error << string_compose (_("AudioDiskstream: channel %1 out of range"), n) << endmsg;
            return -1;
      }

      ChannelInfo* chan = (*c)[n];
      
      if (chan->write_source) {
            chan->write_source->done_with_peakfile_writes ();
            chan->write_source->set_allow_remove_if_empty (true);
            chan->write_source.reset ();
      }

      try {
            if ((chan->write_source = _session.create_audio_source_for_session (*this, n, destructive())) == 0) {
                  throw failed_constructor();
            }
      } 

      catch (failed_constructor &err) {
            error << string_compose (_("%1:%2 new capture file not initialized correctly"), _name, n) << endmsg;
            chan->write_source.reset ();
            return -1;
      }

      /* do not remove destructive files even if they are empty */

      chan->write_source->set_allow_remove_if_empty (!destructive());

      return 0;
}

void
AudioDiskstream::reset_write_sources (bool mark_write_complete, bool force)
{
      ChannelList::iterator chan;
      boost::shared_ptr<ChannelList> c = channels.reader();
      uint32_t n;

      if (!recordable()) {
            return;
      }
      
      capturing_sources.clear ();

      for (chan = c->begin(), n = 0; chan != c->end(); ++chan, ++n) {
            if (!destructive()) {

                  if ((*chan)->write_source && mark_write_complete) {
                        (*chan)->write_source->mark_streaming_write_completed ();
                  }
                  use_new_write_source (n);

                  if (record_enabled()) {
                        capturing_sources.push_back ((*chan)->write_source);
                  }

            } else {
                  if ((*chan)->write_source == 0) {
                        use_new_write_source (n);
                  }
            }
      }

      if (destructive()) {

            /* we now have all our write sources set up, so create the
               playlist's single region.
            */

            if (_playlist->empty()) {
                  setup_destructive_playlist ();
            }
      }
}

int
AudioDiskstream::rename_write_sources ()
{
      ChannelList::iterator chan;
      boost::shared_ptr<ChannelList> c = channels.reader();
      uint32_t n;

      for (chan = c->begin(), n = 0; chan != c->end(); ++chan, ++n) {
            if ((*chan)->write_source != 0) {
                  (*chan)->write_source->set_name (_name, destructive());
                  /* XXX what to do if one of them fails ? */
            }
      }

      return 0;
}

void
AudioDiskstream::set_block_size (nframes_t nframes)
{
      if (_session.get_block_size() > speed_buffer_size) {
            speed_buffer_size = _session.get_block_size();
            boost::shared_ptr<ChannelList> c = channels.reader();

            for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {
                  if ((*chan)->speed_buffer) delete [] (*chan)->speed_buffer;
                  (*chan)->speed_buffer = new Sample[speed_buffer_size];
            }
      }
      allocate_temporary_buffers ();
}

void
AudioDiskstream::allocate_temporary_buffers ()
{
      /* make sure the wrap buffer is at least large enough to deal
         with the speeds up to 1.2, to allow for micro-variation
         when slaving to MTC, SMPTE etc.
      */

      double sp = max (fabsf (_actual_speed), 1.2f);
      nframes_t required_wrap_size = (nframes_t) floor (_session.get_block_size() * sp) + 1;

      if (required_wrap_size > wrap_buffer_size) {

            boost::shared_ptr<ChannelList> c = channels.reader();

            for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {
                  if ((*chan)->playback_wrap_buffer) delete [] (*chan)->playback_wrap_buffer;
                  (*chan)->playback_wrap_buffer = new Sample[required_wrap_size];   
                  if ((*chan)->capture_wrap_buffer) delete [] (*chan)->capture_wrap_buffer;
                  (*chan)->capture_wrap_buffer = new Sample[required_wrap_size];    
            }

            wrap_buffer_size = required_wrap_size;
      }
}

void
AudioDiskstream::monitor_input (bool yn)
{
      boost::shared_ptr<ChannelList> c = channels.reader();

      for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {
            
            if ((*chan)->source) {
                  (*chan)->source->ensure_monitor_input (yn);
            }
      }
}

void
AudioDiskstream::set_align_style_from_io ()
{
      bool have_physical = false;

      if (_io == 0) {
            return;
      }

      get_input_sources ();
      
      boost::shared_ptr<ChannelList> c = channels.reader();

      for (ChannelList::iterator chan = c->begin(); chan != c->end(); ++chan) {
            if ((*chan)->source && (*chan)->source->flags() & JackPortIsPhysical) {
                  have_physical = true;
                  break;
            }
      }

      if (have_physical) {
            set_align_style (ExistingMaterial);
      } else {
            set_align_style (CaptureTime);
      }
}

int
AudioDiskstream::add_channel_to (boost::shared_ptr<ChannelList> c, uint32_t how_many)
{
      while (how_many--) {
            c->push_back (new ChannelInfo(_session.diskstream_buffer_size(), speed_buffer_size, wrap_buffer_size));
      }

      _n_channels = c->size();

      return 0;
}

int
AudioDiskstream::add_channel (uint32_t how_many)
{
      RCUWriter<ChannelList> writer (channels);
      boost::shared_ptr<ChannelList> c = writer.get_copy();

      return add_channel_to (c, how_many);
}

int
AudioDiskstream::remove_channel_from (boost::shared_ptr<ChannelList> c, uint32_t how_many)
{
      while (how_many-- && !c->empty()) {
            delete c->back();
            c->pop_back();
      }

      _n_channels = c->size();

      return 0;
}

int
AudioDiskstream::remove_channel (uint32_t how_many)
{
      RCUWriter<ChannelList> writer (channels);
      boost::shared_ptr<ChannelList> c = writer.get_copy();
      
      return remove_channel_from (c, how_many);
}

float
AudioDiskstream::playback_buffer_load () const
{
      boost::shared_ptr<ChannelList> c = channels.reader();

      return (float) ((double) c->front()->playback_buf->read_space()/
                  (double) c->front()->playback_buf->bufsize());
}

float
AudioDiskstream::capture_buffer_load () const
{
      boost::shared_ptr<ChannelList> c = channels.reader();

      return (float) ((double) c->front()->capture_buf->write_space()/
                  (double) c->front()->capture_buf->bufsize());
}

int
AudioDiskstream::use_pending_capture_data (XMLNode& node)
{
      const XMLProperty* prop;
      XMLNodeList nlist = node.children();
      XMLNodeIterator niter;
      boost::shared_ptr<AudioFileSource> fs;
      boost::shared_ptr<AudioFileSource> first_fs;
      SourceList pending_sources;
      nframes_t position;

      if ((prop = node.property (X_("at"))) == 0) {
            return -1;
      }

      if (sscanf (prop->value().c_str(), "%" PRIu32, &position) != 1) {
            return -1;
      }

      for (niter = nlist.begin(); niter != nlist.end(); ++niter) {
            if ((*niter)->name() == X_("file")) {

                  if ((prop = (*niter)->property (X_("path"))) == 0) {
                        continue;
                  }

                  // This protects sessions from errant CapturingSources in stored sessions
                  struct stat sbuf;
                  if (stat (prop->value().c_str(), &sbuf)) {
                        continue;
                  }

                  try {
                        fs = boost::dynamic_pointer_cast<AudioFileSource> (SourceFactory::createWritable (_session, prop->value(), false, _session.frame_rate()));
                  }

                  catch (failed_constructor& err) {
                        error << string_compose (_("%1: cannot restore pending capture source file %2"),
                                      _name, prop->value())
                              << endmsg;
                        return -1;
                  }

                  pending_sources.push_back (fs);
                  
                  if (first_fs == 0) {
                        first_fs = fs;
                  }

                  fs->set_captured_for (_name);
            }
      }

      if (pending_sources.size() == 0) {
            /* nothing can be done */
            return 1;
      }

      if (pending_sources.size() != _n_channels) {
            error << string_compose (_("%1: incorrect number of pending sources listed - ignoring them all"), _name)
                  << endmsg;
            return -1;
      }

      boost::shared_ptr<AudioRegion> region;
      
      try {
            region = boost::dynamic_pointer_cast<AudioRegion> (RegionFactory::create (pending_sources, 0, first_fs->length(),
                                                                    region_name_from_path (first_fs->name(), true), 
                                                                    0, AudioRegion::Flag (AudioRegion::DefaultFlags|AudioRegion::Automatic|AudioRegion::WholeFile)));
            region->special_set_position (0);
      }

      catch (failed_constructor& err) {
            error << string_compose (_("%1: cannot create whole-file region from pending capture sources"),
                          _name)
                  << endmsg;
            
            return -1;
      }

      try {
            region = boost::dynamic_pointer_cast<AudioRegion> (RegionFactory::create (pending_sources, 0, first_fs->length(), region_name_from_path (first_fs->name(), true)));
      }

      catch (failed_constructor& err) {
            error << string_compose (_("%1: cannot create region from pending capture sources"),
                          _name)
                  << endmsg;
            
            return -1;
      }

      _playlist->add_region (region, position);

      return 0;
}

int
AudioDiskstream::set_destructive (bool yn)
{
      bool bounce_ignored;

      if (yn != destructive()) {
            
            if (yn) {
                  /* requestor should already have checked this and
                     bounced if necessary and desired 
                  */
                  if (!can_become_destructive (bounce_ignored)) {
                        return -1;
                  }
                  _flags = Flag (_flags | Destructive);
                  use_destructive_playlist ();
            } else {
                  _flags = Flag (_flags & ~Destructive);
                  reset_write_sources (true, true);
            }
      }

      return 0;
}

bool
AudioDiskstream::can_become_destructive (bool& requires_bounce) const
{
      if (!_playlist) {
            requires_bounce = false;
            return false;
      }

      /* is there only one region ? */

      if (_playlist->n_regions() != 1) {
            requires_bounce = true;
            return false;
      }

      boost::shared_ptr<Region> first = _playlist->find_next_region (_session.current_start_frame(), Start, 1);
      assert (first);

      /* do the source(s) for the region cover the session start position ? */
      
      if (first->position() != _session.current_start_frame()) {
            if (first->start() > _session.current_start_frame()) {
                  requires_bounce = true;
                  return false;
            }
      }

      /* is the source used by only 1 playlist ? */

      boost::shared_ptr<AudioRegion> afirst = boost::dynamic_pointer_cast<AudioRegion> (first);

      assert (afirst);

      if (afirst->source()->used() > 1) {
            requires_bounce = true; 
            return false;
      }

      requires_bounce = false;
      return true;
}

AudioDiskstream::ChannelInfo::ChannelInfo (nframes_t bufsize, nframes_t speed_size, nframes_t wrap_size)
{
      peak_power = 0.0f;
      source = 0;
      current_capture_buffer = 0;
      current_playback_buffer = 0;
      curr_capture_cnt = 0;

      speed_buffer = new Sample[speed_size];
      playback_wrap_buffer = new Sample[wrap_size];
      capture_wrap_buffer = new Sample[wrap_size];

      playback_buf = new RingBufferNPT<Sample> (bufsize);
      capture_buf = new RingBufferNPT<Sample> (bufsize);
      capture_transition_buf = new RingBufferNPT<CaptureTransition> (256);
      
      /* touch the ringbuffer buffers, which will cause
         them to be mapped into locked physical RAM if
         we're running with mlockall(). this doesn't do
         much if we're not.  
      */

      memset (playback_buf->buffer(), 0, sizeof (Sample) * playback_buf->bufsize());
      memset (capture_buf->buffer(), 0, sizeof (Sample) * capture_buf->bufsize());
      memset (capture_transition_buf->buffer(), 0, sizeof (CaptureTransition) * capture_transition_buf->bufsize());
}

AudioDiskstream::ChannelInfo::~ChannelInfo ()
{
      if (write_source) {
            write_source.reset ();
      }
            
      if (speed_buffer) {
            delete [] speed_buffer;
            speed_buffer = 0;
      }

      if (playback_wrap_buffer) {
            delete [] playback_wrap_buffer;
            playback_wrap_buffer = 0;
      }

      if (capture_wrap_buffer) {
            delete [] capture_wrap_buffer;
            capture_wrap_buffer = 0;
      }
      
      if (playback_buf) {
            delete playback_buf;
            playback_buf = 0;
      }

      if (capture_buf) {
            delete capture_buf;
            capture_buf = 0;
      }

      if (capture_transition_buf) {
            delete capture_transition_buf;
            capture_transition_buf = 0;
      }
}

Generated by  Doxygen 1.6.0   Back to index