]> git.donarmstrong.com Git - lilypond.git/blobdiff - lily/score-performer.cc
Doc-es: various updates.
[lilypond.git] / lily / score-performer.cc
index b5983bb25e25f3ce26806c8ebb6ea37a56355947..b44bb489f5b9ccec4b8c46eb97775ac9278eae25 100644 (file)
@@ -1,33 +1,55 @@
 /*
-  score-performer.cc -- implement Score_performer
+  This file is part of LilyPond, the GNU music typesetter.
 
-  source file of the GNU LilyPond music typesetter
+  Copyright (C) 1996--2015 Jan Nieuwenhuizen <janneke@gnu.org>
 
-  (c) 1996--2005 Jan Nieuwenhuizen <janneke@gnu.org>
+  LilyPond 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 3 of the License, or
+  (at your option) any later version.
+
+  LilyPond 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 LilyPond.  If not, see <http://www.gnu.org/licenses/>.
 */
 
 #include "score-performer.hh"
 
 #include "audio-column.hh"
 #include "audio-item.hh"
+#include "context-def.hh"
+#include "dispatcher.hh"
+#include "global-context.hh"
 #include "performance.hh"
 #include "midi-stream.hh"
+#include "output-def.hh"
 #include "string-convert.hh"
 #include "warn.hh"
-#include "context-def.hh"
-#include "output-def.hh"
-#include "context.hh"
+#include "audio-staff.hh"
 
 ADD_TRANSLATOR_GROUP (Score_performer,
-                     /* doc */ "",
-                     /* create */ "",
-                     /* accept */ "",
-                     /* read */ "",
-                     /* write */ "");
+                      /* doc */
+                      "",
+
+                      /* create */
+                      "",
+
+                      /* read */
+                      "",
+
+                      /* write */
+                      ""
+                     );
 
 Score_performer::Score_performer ()
 {
   performance_ = 0;
+  skipping_ = false;
+  audio_column_ = 0;
 }
 
 Score_performer::~Score_performer ()
@@ -35,56 +57,110 @@ Score_performer::~Score_performer ()
 }
 
 void
-Score_performer::play_element (Audio_element *p)
+Score_performer::announce_element (Audio_element_info info)
 {
-  if (Audio_item *i = dynamic_cast<Audio_item *> (p))
-    audio_column_->add_audio_item (i);
-  performance_->add_element (p);
+  announce_infos_.push_back (info);
+  if (Audio_staff *s = dynamic_cast<Audio_staff *> (info.elem_))
+    {
+      performance_->audio_staffs_.push_back (s);
+    }
+
+  performance_->add_element (info.elem_);
 }
 
 void
-Score_performer::announce_element (Audio_element_info info)
+Score_performer::acknowledge_audio_elements ()
 {
-  announce_infos_.push (info);
+  for (vsize i = 0; i < announce_infos_.size (); i++)
+    {
+      if (Audio_item *ai = dynamic_cast<Audio_item *> (announce_infos_[i].elem_))
+        audio_column_->add_audio_item (ai);
+    }
+  Performer_group::acknowledge_audio_elements ();
 }
 
 void
-Score_performer::prepare (Moment m)
+Score_performer::connect_to_context (Context *c)
 {
-  audio_column_ = new Audio_column (m);
-  play_element (audio_column_);
-  precomputed_recurse_over_translators (context (), START_TRANSLATION_TIMESTEP, UP);
+  Performer_group::connect_to_context (c);
+
+  Dispatcher *d = c->get_global_context ()->event_source ();
+  d->add_listener (GET_LISTENER (Score_performer, one_time_step), ly_symbol2scm ("OneTimeStep"));
+  d->add_listener (GET_LISTENER (Score_performer, prepare), ly_symbol2scm ("Prepare"));
+  d->add_listener (GET_LISTENER (Score_performer, finish), ly_symbol2scm ("Finish"));
 }
 
 void
-Score_performer::finish ()
+Score_performer::disconnect_from_context ()
 {
-  recurse_over_translators (context (),
-                           &Translator::finalize,
-                           &Translator_group::finalize,
-                           UP);
+  Dispatcher *d = context ()->get_global_context ()->event_source ();
+  d->remove_listener (GET_LISTENER (Score_performer, one_time_step), ly_symbol2scm ("OneTimeStep"));
+  d->remove_listener (GET_LISTENER (Score_performer, prepare), ly_symbol2scm ("Prepare"));
+  d->remove_listener (GET_LISTENER (Score_performer, finish), ly_symbol2scm ("Finish"));
+
+  Performer_group::disconnect_from_context ();
 }
 
 void
-Score_performer::one_time_step ()
+Score_performer::prepare (SCM sev)
 {
-  precomputed_recurse_over_translators (context (), PROCESS_MUSIC, UP);
-  do_announces ();
-  precomputed_recurse_over_translators (context (), STOP_TRANSLATION_TIMESTEP, UP);
+  Stream_event *ev = unsmob<Stream_event> (sev);
+  SCM sm = ev->get_property ("moment");
+  Moment *m = unsmob<Moment> (sm);
+  audio_column_ = new Audio_column (*m);
+  announce_element (Audio_element_info (audio_column_, 0));
+  precomputed_recurse_over_translators (context (), START_TRANSLATION_TIMESTEP, UP);
 }
 
-int
-Score_performer::get_tempo () const
+void
+Score_performer::finish (SCM)
 {
-  return ::get_tempo (performance_->midi_, Moment (Rational (1, 4)));
+  SCM channel_mapping = context ()->get_property ("midiChannelMapping");
+  bool use_ports = scm_is_eq (channel_mapping, ly_symbol2scm ("voice"));
+  performance_->ports_ = use_ports;
+  recurse_over_translators
+    (context (),
+     Callback0_wrapper::make_smob<Translator, &Translator::finalize> (),
+     Callback0_wrapper::make_smob<Translator_group, &Translator_group::finalize> (),
+     UP);
 }
 
-SCM
-Score_performer::get_output ()
+void
+Score_performer::one_time_step (SCM)
 {
-  Music_output *o = performance_;
-  performance_ = 0;
-  return o->self_scm ();
+  // audio_column_ can be 0 when prepare has not been called.  The
+  // condition is triggered when Simple_music_iterator implicitly
+  // creates a Score context, like when writing
+  //
+  // \score { { | c4 c c c } \midi { } }
+  //
+  // The same situation happens with the Score_engraver group, but it
+  // would appear not to suffer any bad side effects.
+
+  if (!audio_column_)
+    audio_column_ = new Audio_column (context ()->now_mom ());
+  if (to_boolean (context ()->get_property ("skipTypesetting")))
+    {
+      if (!skipping_)
+        {
+          skip_start_mom_ = audio_column_->when ();
+          skipping_ = true;
+        }
+    }
+  else
+    {
+      if (skipping_)
+        {
+          offset_mom_ -= audio_column_->when () - skip_start_mom_;
+          skipping_ = false;
+        }
+
+      audio_column_->offset_when (offset_mom_);
+      precomputed_recurse_over_translators (context (), PROCESS_MUSIC, UP);
+      do_announces ();
+    }
+
+  precomputed_recurse_over_translators (context (), STOP_TRANSLATION_TIMESTEP, UP);
 }
 
 void
@@ -93,7 +169,6 @@ Score_performer::derived_mark () const
   if (performance_)
     scm_gc_mark (performance_->self_scm ());
 
-  Score_translator::derived_mark ();
   Performer_group::derived_mark ();
 }
 
@@ -102,6 +177,7 @@ Score_performer::initialize ()
 {
   performance_ = new Performance;
   performance_->unprotect ();
+  context ()->set_property ("output", performance_->self_scm ());
   performance_->midi_ = context ()->get_output_def ();
 
   Translator_group::initialize ();