+Initially, the idea behind the functional classes - MooSequence, MooTrack and MooNote - was that they would extend the classes provided by the Java MIDI API and provide additional functionality for simplifying data manipulation. Soon however, it was discovered that this was not possible because of the methods related to these classes in the API. It was then decided that these classes would imitate the data hierarchy of the MIDI file, as the Java classes do, but provide their own implementations. When the edited sequence was to be sent to the synthesizer for playback, the data would be assembled and inserted into the Java classes. At a later stage, this design was also scrapped, the reason being that creating and maintaining such a data structure would require more code than using the somewhat limited methods of the API. MooSequence and MooTrack was therefore removed.
+
+2.6.2 Adding and removing some of the GUI classes
+As this was the first time for all of us in designing an application before implementation, we were not able to predict exactly which classes would be required. During the course of the project, the following classes were added:
+x MooDialog A dialog generator class for all the application's dialogs.
+x MooInstrumentList A combo box with the 128 General MIDI instruments.
+x MooKeyboard A keyboard listener emulating a synthesizer.
+
+Two classes, MooStatus and MooNoteProp, were also removed. The status bar was reduced to a JLabel with a single method and was therefore merged inte the main GUI class. The note properties dialog was inserted into the new MooDialog class.
+------------------------------
+2.7 Problems
+Of all the problems we have come across in our work with Moosique, only one persisted and was until very recently unresolved. As always when working with Java, we were aware that the visualisation performance leaves more to be desired. But we did not expect the extremely low performance the program displayed. During playback, a thread updates the view port of the main scroll pane (MooView). However, the update delay on low-end hardware exceeded one second. This obviously made the GUI unresponsive, but also affected synthesizer playback performance. In order for the application to work properly, this delay must at the very least fall below the time it takes the synthesizer to play a sixteenth note. In a sequence of standard tempo, 120 beats per minute, this would mean 1/8 of a second.
+
+This problem was especially unfortunate since it was the only performance aspect that was mentioned in the requirements specification.
+
+[Ändra font och dra ner textstorlek, citat juh!]
+"Delays during execution should be minimized, but are allowed when loading files. Playback should be smooth and delays should not change the beat or tempo."
+
+However, as we recently discovered, the Sun developers were also aware of this issue and, in trying to find a remedy for it, implemented some additional methods for controlling the viewport's visualisation during scrolling. This allowed us to reduce the time lag to a fraction.
+
+Another issue, that could be resolved only through trial-and-error, was the application's interaction with the setSequence method of the Sequencer class. The method employs, as opposed to what is customary in Java, dereferencing of its argument. This means that the MIDI sequence passed to the method is somehow cloned before it is sent to the sequencer, and therefore when the data of the sequence is updated, the changes are not reflected in the sequence currently loaded into the sequencer. In the Programmer's Guide, this method is described as "[tying] together an existing Sequence with the Sequencer, which is somewhat analogous to loading a tape onto a tape recorder". However, the actual implementation of the method (which may be vendor specific) does not work that way, since the "tape recorder" reads the contents of the "tape", but then does not respond to any changes in the "tape" unless the tape is reinserted. This incomprehensible implementation, which is not even hinted at in the API or the Programmer's Guide, didn't cause us a lot of extra work (since we could not understand where the problem was), but postponed a vital step in our implementation to the very last stage of the project.
+
+During our project some time has also been wasted when wrestling with Java Swing components showing up at the wrong position whith the wrong size. No matter how much we studied the API we could not find the answer of this problem. In the end we found out that rebooting our FreeBSD machines was the only cure to this problem. We suspect there is some kind of bug, either in the FreeBSD OS, or the windomaker application we use. After all, we never ran into this problem those rare times we were working on a Windows-platform
+
+The major changes and that not enough work design of the graphical classes resulted in that the different graphical classes couldn't be implemented on their own. Instead changes often had to take place in many different classes when some feature was added. This made the implementation slow, especially in the beginning, since there where many different ideas on how things should be implemented. This got better through time though, when the basic foundation was implemented it got easier to add the rest of the features.
+------------------------------
+3. Conclusions
+During the course of this project, the importance of extensive system design has become clear to us. With no prior experience in writing such documents or with working with the API in question, designing every aspect of our application would have been nearly impossible without at least approaching the implementation phase. In the future, our proficiency in this field will hopefully improve drastically, and we will be able to save a lot of implementation time by writing more complete design documentation.
+
+------------------------------
+4. References