+@item Finale has placed all of the rests at fixed heights on the staff.
+The user is free to adjust them as needed, but the software makes no
+attempt to consider the content of the other voice. As luck would have
+it, there are no true collisions between notes and rests in this example,
+but that has more to do with the positions of the notes than the rest.
+In other words, Bach deserves more credit for avoiding a complete
+collision than Finale does.
+
+@end itemize
+
+This example is not intended to suggest that Finale cannot be used to
+produce publication-quality output. On the contrary, in the hands of a
+skilled user it can and does, but it requires skill and time. One of the
+fundamental differences between LilyPond and commercial scorewriters is
+that LilyPond hopes to reduce the amount of human intervention to an
+absolute minimum, while other packages try to provide an attractive
+interface in which to make these types of edits.
+
+One particularly glaring omission we found from Finale is a missing flat
+in measure 33:
+
+@quotation
+@iftex
+@sourceimage{pdf/bwv861mm33-34-annotate,7.93cm,,}
+@end iftex
+@ifnottex
+@sourceimage{bwv861mm33-34-annotate,,,png}
+@end ifnottex
+@end quotation
+
+@noindent
+The flat symbol is required to cancel out the natural in the same
+measure, but Finale misses it because it occurred in a different voice.
+So in addition to running a beaming plug-in and checking the spacing on
+the noteheads and rests, the user must also check each measure for
+cross-voice accidentals to avoid interrupting a rehearsal over an
+engraving error.
+
+If you are interested in examining these examples in more detail, the
+full seven-measure excerpt can be found at the end of this essay along
+with four different published engravings. Close examination reveals that
+there is some acceptable variation among the hand-engravings, but that
+LilyPond compares reasonably well to that acceptable range. There are
+still some shortcomings in the LilyPond output, for example, it appears
+a bit too aggressive in shortening some of the stems, so there is room
+for further development and fine-tuning.
+
+Of course, typography relies on human judgment of appearance, so people
+cannot be replaced completely. However, much of the dull work can be
+automated. If LilyPond solves most of the common situations correctly,
+this will be a huge improvement over existing software. Over the course
+of years, the software can be refined to do more and more things
+automatically, so manual overrides are less and less necessary. Where
+manual adjustments are needed, LilyPond's structure has been designed
+with that flexibility in mind.
+
+@node Building software
+@section Building software
+
+This section describes some of the programming decisions that we made
+when designing LilyPond.
+
+@menu
+* Music representation::
+* What symbols to engrave?::
+* Flexible architecture::
+@end menu
+
+
+@node Music representation
+@unnumberedsubsec Music representation
+
+@cindex syntax
+@cindex recursive structures
+
+Ideally, the input format for any high-level formatting system is
+an abstract description of the content. In this case, that would
+be the music itself. This poses a formidable problem: how can we
+define what music really is? Instead of trying to find an answer,
+we have reversed the question. We write a program capable of
+producing sheet music, and adjust the format to be as lean as
+possible. When the format can no longer be trimmed down, by
+definition we are left with content itself. Our program serves as
+a formal definition of a music document.
+
+The syntax is also the user-interface for LilyPond, hence it is
+easy to type:
+
+@example
+@{
+ c'4 d'8
+@}
+@end example
+
+@noindent
+to create a quarter note on middle C (C1) and an eighth note on
+the D above middle C (D1).
+
+@lilypond[quote]
+{
+ c'4 d'8
+}
+@end lilypond
+
+On a microscopic scale, such syntax is easy to use. On a larger
+scale, syntax also needs structure. How else can you enter
+complex pieces like symphonies and operas? The structure is
+formed by the concept of music expressions: by combining small
+fragments of music into larger ones, more complex music can be
+expressed. For example
+
+@lilypond[quote,verbatim,fragment,relative=1]
+f4
+@end lilypond
+
+@noindent
+Simultaneous notes can be constructed by enclosing them with
+@code{<<} and @code{>>}:
+
+@example
+<<c4 d4 e4>>
+@end example
+
+@lilypond[quote,fragment,relative=1]
+\new Voice { <<c4 d4 e>> }
+@end lilypond
+
+@noindent
+This expression is put in sequence by enclosing it in curly braces
+@code{@{@tie{}@dots{}@tie{}@}}:
+
+@example
+@{ f4 <<c4 d4 e4>> @}
+@end example
+
+@lilypond[quote,relative=1,fragment]
+{ f4 <<c d e4>> }
+@end lilypond
+
+@noindent
+The above is also an expression, and so it may be combined again
+with another simultaneous expression (a half note) using
+@code{<<}, @code{\\}, and @code{>>}:
+
+@example
+<< g2 \\ @{ f4 <<c4 d4 e4>> @} >>
+@end example
+
+@lilypond[quote,fragment,relative=2]
+\new Voice { << g2 \\ { f4 <<c d e>> } >> }
+@end lilypond
+
+Such recursive structures can be specified neatly and formally in
+a context-free grammar. The parsing code is also generated from
+this grammar. In other words, the syntax of LilyPond is clearly
+and unambiguously defined.
+
+User-interfaces and syntax are what people see and deal with most.
+They are partly a matter of taste, and also the subject of much
+discussion. Although discussions on taste do have their merit,
+they are not very productive. In the larger picture of LilyPond,
+the importance of input syntax is small: inventing neat syntax is
+easy, while writing decent formatting code is much harder. This
+is also illustrated by the line-counts for the respective
+components: parsing and representation take up less than 10% of
+the source code.
+
+When designing the structures used in LilyPond, we made some different
+decisions than are apparent in other software. Consider the hierarchical
+nature of music notation:
+
+@lilypond[quote,fragment]
+<<
+ \new Staff \relative c'' {
+ \key g \major
+ \time 3/4
+ d4 g,8 a b c d4 g, g
+ }
+ \new Staff \relative c' {
+ \clef "bass"
+ \key g \major
+ <g b d>2 a4 b2.
+ }
+>>
+@end lilypond
+
+In this case, there are pitches grouped into chords that belong to
+measures, which belong to staves. This resembles a tidy structure of
+nested boxes:
+
+@quotation
+@iftex
+@sourceimage{pdf/nestedboxes,,4cm,}
+@end iftex
+@ifnottex
+@sourceimage{nestedboxes,,,png}
+@end ifnottex
+@end quotation
+
+Unfortunately, the structure is tidy because it is based on some
+excessively restrictive assumptions. This becomes apparent if we
+consider a more complicated musical example:
+
+@lilypond[quote]
+\layout {
+ \context {
+ \Score
+ \remove "Timing_translator"
+ \remove "Default_bar_line_engraver"
+ }
+ \context {
+ \Staff
+ \consists "Timing_translator"
+ \consists "Default_bar_line_engraver"
+ }
+}
+
+\new PianoStaff <<
+ \new Staff = "RH" <<
+ \new Voice = "I" \relative c''' {
+ \time 3/4
+ \voiceOne
+ \tuplet 7/6 { g8 g g g g g g }
+ \oneVoice
+ r4 <b,, fis' g bes> r4\fermata
+ }
+ \new Voice = "II" \relative c' {
+ \voiceTwo
+ c4
+ \tuplet 5/4 {
+ <c ees>8 f g
+ \change Staff = "LH" \oneVoice
+ \stemUp g,( c}
+ r4
+ \override Stem.cross-staff = ##t
+ \override Stem.length = #12
+ <fis, b>) r\fermata
+ }
+ >>
+ \new Staff = "LH" <<
+ \new Voice = "III" \relative c' {
+ \time 2/4
+ \clef "bass"
+ g4 \stopStaff s
+ \startStaff s2*2
+ }
+ >>
+>>
+@end lilypond
+
+In this example, staves start and stop at will, voices jump around
+between staves, and the staves have different time signatures. Many
+software packages would struggle with reproducing this example because
+they are built on the nested box structure. With LilyPond, on the other
+hand, we have tried to keep the input format and the structure as
+flexible as possible.