[lilypond.git] / Documentation / contributor / programming-work.itexi

@c -*- coding: utf-8; mode: texinfo; -*-
@node Programming work
@chapter Programming work

@menu
* Overview of LilyPond architecture::
* LilyPond programming languages::
* Programming without compiling::
* Finding functions::
* Code style::
* Warnings Errors Progress and Debug Output::
* Debugging LilyPond::
* Tracing object relationships::
* Adding or modifying features::
* Iterator tutorial::
* Engraver tutorial::
* Callback tutorial::
* Understanding pure properties::
* LilyPond scoping::
* Scheme->C interface::
* LilyPond miscellany::
@end menu

@node Overview of LilyPond architecture
@section Overview of LilyPond architecture

LilyPond processes the input file into graphical and musical output in a
number of stages.  This process, along with the types of routines that
accomplish the various stages of the process, is described in this section.  A
more complete description of the LilyPond architecture and internal program
execution is found in Erik Sandberg's
@uref{http://lilypond.org/website/pdf/thesis-erik-sandberg.pdf, master's
thesis}.

The first stage of LilyPond processing is @emph{parsing}.  In the parsing
process, music expressions in LilyPond input format are converted to music
expressions in Scheme format.  In Scheme format, a music expression is a list
in tree form, with nodes that indicate the relationships between various music
events.  The LilyPond parser is written in Bison.

The second stage of LilyPond processing is @emph{iterating}.  Iterating
assigns each music event to a context, which is the environment in which the
music will be finally engraved.  The context is responsible for all further
processing of the music.  It is during the iteration stage that contexts are
created as necessary to ensure that every note has a Voice type context (e.g.
Voice, TabVoice, DrumVoice, CueVoice, MensuralVoice, VaticanaVoice,
GregorianTranscriptionVoice), that the Voice type contexts exist in
appropriate Staff type contexts, and that parallel Staff type contexts exist
in StaffGroup type contexts.  In addition, during the iteration stage each
music event is assigned a moment, or a time in the music when the event
begins.

Each type of music event has an associated iterator.  Iterators are defined in
@file{*-iterator.cc}.  During iteration, an
event's iterator is called to deliver that music event to the appropriate
context(s).

The final stage of LilyPond processing is @emph{translation}.  During
translation, music events are prepared for graphical or midi output.  The
translation step is accomplished by the polymorphic base class Translator
through its two derived classes: Engraver (for graphical output) and
Performer (for midi output).

Translators are defined in C++ files named @file{*-engraver.cc}
and @file{*-performer.cc}.
Much of the work of translating is handled by Scheme functions,
which is one of the keys to LilyPond's exceptional flexibility.

@sourceimage{architecture-diagram,,,png}


@node LilyPond programming languages
@section LilyPond programming languages

Programming in LilyPond is done in a variety of programming languages.  Each
language is used for a specific purpose or purposes.  This section describes
the languages used and provides links to reference manuals and tutorials for
the relevant language.

@subsection C++

The core functionality of LilyPond is implemented in C++.

C++ is so ubiquitous that it is difficult to identify either a reference
manual or a tutorial.  Programmers unfamiliar with C++ will need to spend some
time to learn the language before attempting to modify the C++ code.

The C++ code calls Scheme/GUILE through the GUILE interface, which is
documented in the
@uref{http://www.gnu.org/software/guile/manual/html_node/index.html, GUILE
  Reference Manual}.

@subsection Flex

The LilyPond lexer is implemented in Flex, an implementation of the Unix lex
lexical analyser generator.  Resources for Flex can be found
@uref{http://flex.sourceforge.net/, here}.

@subsection GNU Bison

The LilyPond parser is implemented in Bison, a GNU parser generator.  The
Bison homepage is found at @uref{http://www.gnu.org/software/bison/,
gnu.org}.  The manual (which includes both a reference and tutorial) is
@uref{http://www.gnu.org/software/bison/manual/index.html, available} in a
variety of formats.

@subsection GNU Make

GNU Make is used to control the compiling process and to build the
documentation and the website.  GNU Make documentation is available at
@uref{http://www.gnu.org/software/make/manual/, the GNU website}.

@subsection GUILE or Scheme

GUILE is the dialect of Scheme that is used as LilyPond's extension language.
Many extensions to LilyPond are written entirely in GUILE.  The
@uref{http://www.gnu.org/software/guile/manual/html_node/index.html,
GUILE Reference Manual} is available online.

@uref{http://mitpress.mit.edu/sicp/full-text/book/book.html, Structure and
Interpretation of Computer Programs}, a popular textbook used to teach
programming in Scheme is available in its entirety online.

An introduction to Guile/Scheme as used in LilyPond can be found in the
@rextend{Scheme tutorial}.

@subsection MetaFont

MetaFont is used to create the music fonts used by LilyPond.  A MetaFont
tutorial is available at @uref{http://metafont.tutorial.free.fr/, the
METAFONT tutorial page}.

@subsection PostScript

PostScript is used to generate graphical output.  A brief PostScript tutorial
is @uref{http://local.wasp.uwa.edu.au/~pbourke/dataformats/postscript/,
available online}.  The
@uref{http://www.adobe.com/products/postscript/pdfs/PLRM.pdf, PostScript Language
Reference} is available online in PDF format.

@subsection Python

Python is used for XML2ly and is used for building the documentation and the
website.

Python documentation is available at @uref{http://www.python.org/doc/,
python.org}.

@node Programming without compiling
@section Programming without compiling

Much of the development work in LilyPond takes place by changing @file{*.ly} or
@file{*.scm} files.  These changes can be made without compiling LilyPond.  Such
changes are described in this section.


@subsection Modifying distribution files

Much of LilyPond is written in Scheme or LilyPond input files.  These
files are interpreted when the program is run, rather than being compiled
when the program is built, and are present in all LilyPond distributions.
You will find @file{.ly} files in the @file{ly/} directory and the Scheme files in the
@file{scm/} directory.  Both Scheme files and @file{.ly} files can be modified and
saved with any text editor.  It's probably wise to make a backup copy of
your files before you modify them, although you can reinstall if the
files become corrupted.

Once you've modified the files, you can test the changes just by running
LilyPond on some input file.  It's a good idea to create a file that
demonstrates the feature you're trying to add.  This file will eventually
become a regression test and will be part of the LilyPond distribution.

@subsection Desired file formatting

Files that are part of the LilyPond distribution have Unix-style line
endings (LF), rather than DOS (CR+LF) or MacOS 9 and earlier (CR).  Make
sure you use the necessary tools to ensure that Unix-style line endings are
preserved in the patches you create.

Tab characters should not be included in files for distribution.  All
indentation should be done with spaces.  Most editors have settings to
allow the setting of tab stops and ensuring that no tab characters are
included in the file.

Scheme files and LilyPond files should be written according to standard
style guidelines.  Scheme file guidelines can be found at
@uref{http://community.schemewiki.org/?scheme-style}.  Following these
guidelines will make your code easier to read.  Both you and others that
work on your code will be glad you followed these guidelines.

For LilyPond files, you should follow the guidelines for LilyPond snippets
in the documentation.  You can find these guidelines at
@ref{Texinfo introduction and usage policy}.

@node Finding functions
@section Finding functions

When making changes or fixing bugs in LilyPond, one of the initial
challenges is finding out where in the code tree the functions to
be modified live.  With nearly 3000 files in the source tree,
trial-and-error searching is generally ineffective.  This section
describes a process for finding interesting code.

@subsection Using the ROADMAP

The file ROADMAP is located in the main directory of the lilypond source.
ROADMAP lists all of the directories in the LilyPond source tree, along
with a brief description of the kind of files found in each directory.
This can be a very helpful tool for deciding which directories to search
when looking for a function.


@subsection Using grep to search

Having identified a likely subdirectory to search, the grep utility can
be used to search for a function name.  The format of the grep command is

@example
grep -i functionName subdirectory/*
@end example

This command will search all the contents of the directory subdirectory/
and display every line in any of the files that contains
functionName.  The @option{-i} option makes @command{grep} ignore
case -- this can be very useful if you are not yet familiar with
our capitalization conventions.

The most likely directories to grep for function names are @file{scm/} for
scheme files, ly/ for lilypond input (@file{*.ly}) files, and @file{lily/} for C++
files.


@subsection Using git grep to search

If you have used git to obtain the source, you have access to a
powerful tool to search for functions.  The command:

@example
git grep functionName
@end example

will search through all of the files that are present in the git
repository looking for functionName.  It also presents the results
of the search using @code{less}, so the results are displayed one page
at a time.

@subsection Searching on the git repository at Savannah

You can also use the equivalent of git grep on the Savannah server.

@itemize

@item
Go to http://git.sv.gnu.org/gitweb/?p=lilypond.git

@item
In the pulldown box that says commit, select grep.

@item
Type functionName in the search box, and hit enter/return

@end itemize

This will initiate a search of the remote git repository.


@node Code style
@section Code style

This section describes style guidelines for LilyPond
source code.

@menu
* Languages::
* Filenames::
* Indentation::
* Naming conventions::
* Broken code::
* Code comments::
* Handling errors::
* Localization::
@end menu


@node Languages
@subsection Languages

C++ and Python are preferred.  Python code should use PEP 8.


@node Filenames
@subsection Filenames

Definitions of classes that are only accessed via pointers (*) or
references (&) shall not be included as include files.

@verbatim
   filenames

        ".hh"   Include files
             ".cc"      Implementation files
             ".icc"     Inline definition files
             ".tcc"     non inline Template defs

   in emacs:

             (setq auto-mode-alist
                   (append '(("\\.make$" . makefile-mode)
                        ("\\.cc$" . c++-mode)
                        ("\\.icc$" . c++-mode)
                        ("\\.tcc$" . c++-mode)
                        ("\\.hh$" . c++-mode)
                        ("\\.pod$" . text-mode)
                        )
                      auto-mode-alist))
@end verbatim

The class Class_name is coded in @q{class-name.*}


@node Indentation
@subsection Indentation

Standard GNU coding style is used.

@subsubheading Indenting files with @code{fixcc.py} (recommended)

LilyPond provides a python script that will adjust the indentation
and spacing on a @code{.cc} or @code{.hh} file to very near the
GNU standard:

@example
scripts/auxiliar/fixcc.py FILENAME
@end example

This can be run on all files at once, but this is not recommended
for normal contributors or developers.

@smallexample
scripts/auxiliar/fixcc.py \
  $(find flower lily -name '*cc' -o -name '*hh' | grep -v /out)
@end smallexample


@subsubheading Indenting with emacs

The following hooks will produce indentation which is similar to
our official indentation as produced with @code{fixcc.py}.

@example
(add-hook 'c++-mode-hook
     '(lambda ()
        (c-set-style "gnu")
        (setq indent-tabs-mode nil))
@end example

If you like using font-lock, you can also add this to your
@file{.emacs}:

@example
(setq font-lock-maximum-decoration t)
(setq c++-font-lock-keywords-3
      (append
       c++-font-lock-keywords-3
       '(("\\b\\(a-zA-Z_?+_\\)\\b" 1 font-lock-variable-name-face) ("\\b\\(A-Z?+a-z_?+\\)\\b" 1 font-lock-type-face))
       ))
@end example 


@subheading Indenting with vim

Although emacs indentation is the GNU standard, acceptable
indentation can usually be accomplished with vim.  Some hints for
vim are as follows:

A workable .vimrc:

@example
set cindent
set smartindent
set autoindent
set expandtab
set softtabstop=2
set shiftwidth=2
filetype plugin indent on
set incsearch
set ignorecase smartcase
set hlsearch
set confirm
set statusline=%F%m%r%h%w\ %@{&ff@}\ %Y\ [ASCII=\%03.3b]\ [HEX=\%02.2B]\ %04l,%04v\ %p%%\ [LEN=%L]
set laststatus=2
set number
" Remove trailing whitespace on write
autocmd BufWritePre * :%s/\s\+$//e
@end example

With this @file{.vimrc}, files can be reindented automatically by
highlighting the lines to be indented in visual mode (use V to
enter visual mode) and pressing @code{=}.

A @file{scheme.vim} file will help improve the indentation.  This
one was suggested by Patrick McCarty.  It should be saved in
@file{~/.vim/after/syntax/scheme.vim}.

@example
" Additional Guile-specific 'forms'
syn keyword schemeSyntax define-public define*-public
syn keyword schemeSyntax define* lambda* let-keywords*
syn keyword schemeSyntax defmacro defmacro* define-macro
syn keyword schemeSyntax defmacro-public defmacro*-public
syn keyword schemeSyntax use-modules define-module
syn keyword schemeSyntax define-method define-class

" Additional LilyPond-specific 'forms'
syn keyword schemeSyntax define-markup-command define-markup-list-command
syn keyword schemeSyntax define-safe-public define-music-function
syn keyword schemeSyntax def-grace-function

" All of the above should influence indenting too
set lw+=define-public,define*-public
set lw+=define*,lambda*,let-keywords*
set lw+=defmacro,defmacro*,define-macro
set lw+=defmacro-public,defmacro*-public
set lw+=use-modules,define-module
set lw+=define-method,define-class
set lw+=define-markup-command,define-markup-list-command
set lw+=define-safe-public,define-music-function
set lw+=def-grace-function

" These forms should not influence indenting
set lw-=if
set lw-=set!

" Try to highlight all ly: procedures
syn match schemeFunc "ly:[^) ]\+"
@end example


@node Naming conventions
@subsection Naming Conventions

Naming conventions have been established for LilyPond
source code.

@subheading Classes and Types

Classes begin with an uppercase letter, and words
in class names are separated with @code{_}:

@verbatim
This_is_a_class
@end verbatim

@subheading Members

Member variable names end with an underscore:

@verbatim
Type Class::member_
@end verbatim

@subheading Macros

Macro names should be written in uppercase completely,
with words separated by @code{_}:

@verbatim
THIS_IS_A_MACRO
@end verbatim

@subheading Variables

Variable names should be complete words, rather than abbreviations.
For example, it is preferred to use @code{thickness} rather than
@code{th} or @code{t}.

Multi-word variable names in C++ should have the words separated
by the underscore character (@q{_}):

@verbatim
cxx_multiword_variable
@end verbatim

Multi-word variable names in Scheme should have the words separated
by a hyphen (@q{-}):

@verbatim
scheme-multiword-variable
@end verbatim

@node Broken code
@subsection Broken code

Do not write broken code.  This includes hardwired dependencies,
hardwired constants, slow algorithms and obvious limitations.  If
you can not avoid it, mark the place clearly, and add a comment
explaining shortcomings of the code.

Ideally, the comment marking the shortcoming would include
TODO, so that it is marked for future fixing.

We reject broken-in-advance on principle.


@node Code comments
@subsection Code comments

Comments may not be needed if descriptive variable names are used
in the code and the logic is straightforward.  However, if the
logic is difficult to follow, and particularly if non-obvious
code has been included to resolve a bug, a comment describing
the logic and/or the need for the non-obvious code should be included.

There are instances where the current code could be commented better.
If significant time is required to understand the code as part of
preparing a patch, it would be wise to add comments reflecting your
understanding to make future work easier.


@node Handling errors
@subsection Handling errors

As a general rule, you should always try to continue computations,
even if there is some kind of error.  When the program stops, it
is often very hard for a user to pinpoint what part of the input
causes an error.  Finding the culprit is much easier if there is
some viewable output.

So functions and methods do not return errorcodes, they never
crash, but report a programming_error and try to carry on.

Error and warning messages need to be localized.


@node Localization
@subsection Localization

This document provides some guidelines to help programmers write
proper user
messages.  To help translations, user messages must follow
uniform conventions.  Follow these rules when coding for LilyPond.
Hopefully, this can be replaced by general GNU guidelines in the
future.  Even better would be to have an English (en_BR, en_AM)
guide helping programmers writing consistent messages for all GNU
programs.

Non-preferred messages are marked with `+'.  By convention,
ungrammatical examples are marked with `*'.  However, such ungrammatical
examples may still be preferred.

@itemize

@item
Every message to the user should be localized (and thus be marked
for localization).  This includes warning and error messages.

@item
Do not localize/gettextify:

@itemize
@item
`programming_error ()'s

@item
`programming_warning ()'s

@item
debug strings

@item
output strings (PostScript, TeX, etc.)

@end itemize

@item
Messages to be localized must be encapsulated in `_ (STRING)' or
`_f (FORMAT, ...)'. E.g.:

@example
warning (_ ("need music in a score"));
error (_f ("cannot open file: `%s'", file_name));
@end example

In some rare cases you may need to call `gettext ()' by hand.  This
happens when you pre-define (a list of) string constants for later
use.  In that case, you'll probably also need to mark these string
constants for translation, using `_i (STRING)'.  The `_i' macro is
a no-op, it only serves as a marker for `xgettext'.

@example
char const* messages[] = @{
  _i ("enable debugging output"),
  _i ("ignore lilypond version"),
  0
@};

void
foo (int i)
@{
  puts (gettext (messages i));
@}
@end example

See also @file{flower/getopt-long.cc} and @file{lily/main.cc}.

@item
Do not use leading or trailing whitespace in messages.  If you need
whitespace to be printed, prepend or append it to the translated
message

@example
message ("Calculating line breaks..." + " ");
@end example

@item
Error or warning messages displayed with a file name and line
number never start with a capital, eg,

@example
foo.ly: 12: not a duration: 3
@end example

Messages containing a final verb, or a gerund (`-ing'-form) always
start with a capital.  Other (simpler) messages start with a
lowercase letter

@example
Processing foo.ly...
`foo': not declared.
Not declaring: `foo'.
@end example

@item
Avoid abbreviations or short forms, use `cannot' and `do not'
rather than `can't' or `don't'
To avoid having a number of different messages for the same
situation, well will use quoting like this `"message: `%s'"' for all
strings.  Numbers are not quoted:

@example
_f ("cannot open file: `%s'", name_str)
_f ("cannot find character number: %d", i)
@end example

@item
Think about translation issues.  In a lot of cases, it is better to
translate a whole message.  English grammar must not be imposed on the
translator.  So, instead of

@example
stem at  + moment.str () +  does not fit in beam
@end example

have

@example
_f ("stem at %s does not fit in beam", moment.str ())
@end example

@item
Split up multi-sentence messages, whenever possible.  Instead of

@example
warning (_f ("out of tune!  Can't find: `%s'", "Key_engraver"));
warning (_f ("cannot find font `%s', loading default", font_name));
@end example

rather say:

@example
warning (_ ("out of tune:"));
warning (_f ("cannot find: `%s', "Key_engraver"));
warning (_f ("cannot find font: `%s', font_name));
warning (_f ("Loading default font"));
@end example

@item
If you must have multiple-sentence messages, use full punctuation.
Use two spaces after end of sentence punctuation.  No punctuation
(esp. period) is used at the end of simple messages.

@example
_f ("Non-matching braces in text `%s', adding braces", text)
_ ("Debug output disabled.  Compiled with NPRINT.")
_f ("Huh?  Not a Request: `%s'.  Ignoring.", request)
@end example

@item
Do not modularize too much; words frequently cannot be translated
without context.  It is probably safe to treat most occurrences of
words like stem, beam, crescendo as separately translatable words.

@item
When translating, it is preferable to put interesting information
at the end of the message, rather than embedded in the middle.
This especially applies to frequently used messages, even if this
would mean sacrificing a bit of eloquence.  This holds for original
messages too, of course.

@example
en: cannot open: `foo.ly'
+   nl: kan `foo.ly' niet openen (1)
kan niet openen: `foo.ly'*   (2)
niet te openen: `foo.ly'*    (3)
@end example


The first nl message, although grammatically and stylistically
correct, is not friendly for parsing by humans (even if they speak
dutch).  I guess we would prefer something like (2) or (3).

@item
Do not run make po/po-update with GNU gettext < 0.10.35

@end itemize


@node Warnings Errors Progress and Debug Output
@section Warnings, Errors, Progress and Debug Output

@unnumberedsubsec Available log levels

LilyPond has several loglevels, which specify how verbose the output on
the console should be:
@itemize
@item NONE: No output at all, even on failure
@item ERROR: Only error messages
@item WARN: Only error messages and warnings
@item BASIC_PROGRESS: Warnings, errors and basic progress (success, etc.)
@item PROGRESS: Warnings, errors and full progress messages
@item INFO: Warnings, errors, progress and more detailed information (default)
@item DEBUG: All messages, including full debug messages (very verbose!)
@end itemize

The loglevel can either be set with the environment variable
@code{LILYPOND_LOGLEVEL} or on the command line with the @option{--loglevel=...}
option.

@unnumberedsubsec Functions for debug and log output

LilyPond has two different types of error and log functions:
@itemize 

@item
If a warning or error is caused by an identified position in the input file,
e.g. by a grob or by a music expression, the functions of the @code{Input}
class provide logging functionality that prints the position of the message
in addition to the message.

@item
If a message can not be associated with a particular position in an input file,
e.g. the output file cannot be written, then the functions in the 
@code{flower/include/warn.hh} file will provide logging functionality that 
only prints out the message, but no location.

@end itemize

There are also Scheme functions to access all of these logging functions from
scheme.  In addition, the Grob class contains some convenience wrappers for
even easier access to these functions.

The message and debug functions in @code{warn.hh} also have an optional 
argument @code{newline}, which specifies whether the message should always
start on a new line or continue a previous message.
By default, @code{progress_indication} does NOT start on a new line, but rather
continue the previous output.  They also do not have a particular input
position associated, so there are no progress functions in the Input class.
All other functions by default start their output on a new line.

The error functions come in three different flavors: fatal error messages,
programming error messages and normal error messages.  Errors written
by the @code{error ()} function will cause LilyPond to exit immediately,
errors by @code{Input::error ()} will continue the compilation, but
return a non-zero return value of the LilyPond call (i.e. indicate an
unsuccessful program execution).  All other errors will be printed on the
console, but not exit LilyPond or indicate an unsuccessful return code.
Their only differences to a warnings are the displayed text and that
they will be shown with loglevel @code{ERROR}.

If the Scheme option @code{warning-as-error} is set, any warning will be
treated as if @code{Input::error} was called.


@unnumberedsubsec All logging functions at a glance

@multitable @columnfractions 0.16 0.42 0.42
@headitem
@tab C++, no location
@tab C++ from input location

@item ERROR
@tab @code{error ()}, @code{programming_error (msg)}, @code{non_fatal_error (msg)}
@tab @code{Input::error (msg)}, @code{Input::programming_error (msg)}

@item WARN
@tab @code{warning (msg)}
@tab @code{Input::warning (msg)}

@item BASIC
@tab @code{basic_progress (msg)}
@tab -

@item PROGRESS
@tab @code{progress_indication (msg)}
@tab -

@item INFO
@tab @code{message (msg)}
@tab @code{Input::message (msg)}

@item DEBUG
@tab @code{debug_output (msg)}
@tab @code{Input::debug_output (msg)}

@item @tab @tab

@headitem
@tab C++ from a Grob
@tab Scheme, music expression

@item ERROR
@tab @code{Grob::programming_error (msg)}
@tab -

@item WARN
@tab @code{Grob::warning (msg)}
@tab @code{(ly:music-warning music msg)}

@item BASIC
@tab -
@tab -

@item PROGRESS
@tab -
@tab -

@item INFO
@tab -
@tab @code{(ly:music-message music msg)}

@item DEBUG
@tab -
@tab -

@item @tab @tab

@headitem
@tab Scheme, no location
@tab Scheme, input location

@item ERROR
@tab -
@tab @code{(ly:error msg args)}, @code{(ly:programming-error msg args)}

@item WARN
@tab @code{(ly:warning msg args)}
@tab @code{(ly:input-warning input msg args)}

@item BASIC
@tab @code{(ly:basic-progress msg args)}
@tab -

@item PROGRESS
@tab @code{(ly:progress msg args)}
@tab -

@item INFO
@tab @code{(ly:message msg args)}
@tab @code{(ly:input-message input msg args)}

@item DEBUG
@tab @code{(ly:debug msg args)}
@tab -

@end multitable




@node Debugging LilyPond
@section Debugging LilyPond

The most commonly used tool for debugging LilyPond is the GNU
debugger gdb.  The gdb tool is used for investigating and debugging
core LilyPond code written in C++.  Another tool is available for
debugging Scheme code using the Guile debugger.  This section
describes how to use both gdb and the Guile Debugger.

@menu
* Debugging overview::
* Debugging C++ code::
* Debugging Scheme code::
@end menu

@node Debugging overview
@subsection Debugging overview

Using a debugger simplifies troubleshooting in at least two ways.

First, breakpoints can be set to pause execution at any desired point.
Then, when execution has paused, debugger commands can be issued to
explore the values of various variables or to execute functions.

Second, the debugger can display a stack trace, which shows the
sequence in which functions have been called and the arguments
passed to the called functions.

@node Debugging C++ code
@subsection Debugging C++ code

The GNU debugger, gdb, is the principal tool for debugging C++ code.

@subheading Compiling LilyPond for use with gdb

In order to use gdb with LilyPond, it is necessary to compile
LilyPond with debugging information.  This is the current default
mode of compilation.  Often debugging becomes more complicated
when the compiler has optimised variables and function calls away.
In that case it may be helpful to run the following command in the
main LilyPond source directory:

@example
./configure --disable-optimising
make
@end example

This will create a version of LilyPond with minimal optimization
which will allow the debugger to access all variables and step
through the source code in-order.  It may not accurately reproduce
bugs encountered with the optimized version, however.

You should not do @var{make install} if you want to use a debugger
with LilyPond.  The @var{make install} command will strip debugging
information from the LilyPond binary.

@subheading Typical gdb usage

Once you have compiled the LilyPond image with the necessary
debugging information it will have been written to a location in a
subfolder of your current working directory:

@example
out/bin/lilypond
@end example

This is important as you will need to let gdb know where to find the
image containing the symbol tables.  You can invoke gdb from the
command line using the following:

@example
gdb out/bin/lilypond
@end example
@noindent
This loads the LilyPond symbol tables into gdb.  Then, to run
LilyPond on @file{test.ly} under the debugger, enter the following:

@example
run test.ly
@end example

@noindent
at the gdb prompt.

As an alternative to running gdb at the command line you may try
a graphical interface to gdb such as ddd:

@example
ddd out/bin/lilypond
@end example

You can also use sets of standard gdb commands stored in a .gdbinit
file (see next section).

@subheading Typical .gdbinit files

The behavior of gdb can be readily customized through the use of a
@var{.gdbinit} file.  A @var{.gdbinit} file is a file named
@var{.gdbinit} (notice the @qq{.} at the beginning of the file name)
that is placed in a user's home directory.

The @var{.gdbinit} file below is from Han-Wen.  It sets breakpoints
for all errors and defines functions for displaying scheme objects
(ps), grobs (pgrob), and parsed music expressions (pmusic).

@example
file $LILYPOND_GIT/build/out/bin/lilypond
b programming_error
b Grob::programming_error

define ps
   print ly_display_scm($arg0)
end
define pgrob
  print ly_display_scm($arg0->self_scm_)
  print ly_display_scm($arg0->mutable_property_alist_)
  print ly_display_scm($arg0->immutable_property_alist_)
  print ly_display_scm($arg0->object_alist_)
end
define pmusic
  print ly_display_scm($arg0->self_scm_)
  print ly_display_scm($arg0->mutable_property_alist_)
  print ly_display_scm($arg0->immutable_property_alist_)
end
@end example

@node Debugging Scheme code
@subsection Debugging Scheme code

Scheme code can be developed using the Guile command line
interpreter @code{top-repl}.  You can either investigate
interactively using just Guile or you can use the debugging
tools available within Guile.

@subheading Using Guile interactively with LilyPond

In order to experiment with Scheme programming in the LilyPond
environment, it is necessary to have a Guile interpreter that
has all the LilyPond modules loaded.  This requires the following
steps.

First, define a Scheme symbol for the active module in the @file{.ly} file:

@example
#(module-define! (resolve-module '(guile-user))
                 'lilypond-module (current-module))
@end example

Now place a Scheme function in the @file{.ly} file that gives an
interactive Guile prompt:

@example
#(top-repl)
@end example

When the @file{.ly} file is compiled, this causes the compilation to be
interrupted and an interactive guile prompt to appear.  Once the
guile prompt appears, the LilyPond active module must be set as the
current guile module:

@example
guile> (set-current-module lilypond-module)
@end example

You can demonstrate these commands are operating properly by typing the name
of a LilyPond public scheme function to check it has been defined:

@example
guile> fret-diagram-verbose-markup
#<procedure fret-diagram-verbose-markup (layout props marking-list)>
@end example

If the LilyPond module has not been correctly loaded, an error
message will be generated:

@example
guile> fret-diagram-verbose-markup
ERROR: Unbound variable: fret-diagram-verbose-markup
ABORT: (unbound-variable)
@end example

Once the module is properly loaded, any valid LilyPond Scheme
expression can be entered at the interactive prompt.

After the investigation is complete, the interactive guile
interpreter can be exited:

@example
guile> (quit)
@end example

The compilation of the @file{.ly} file will then continue.

@subheading Using the Guile debugger

To set breakpoints and/or enable tracing in Scheme functions, put

@example
\include "guile-debugger.ly"
@end example

in your input file after any scheme procedures you have defined in
that file.  This will invoke the Guile command-line after having set
up the environment for the debug command-line.  When your input file
is processed, a guile prompt will be displayed.  You may now enter
commands to set up breakpoints and enable tracing by the Guile debugger.

@subheading Using breakpoints

At the guile prompt, you can set breakpoints with
the @code{set-break!} procedure:

@example
guile> (set-break! my-scheme-procedure)
@end example

Once you have set the desired breakpoints, you exit the guile repl frame
by typing:

@example
guile> (quit)
@end example

Then, when one of the scheme routines for which you have set
breakpoints is entered, guile will interrupt execution in a debug
frame.  At this point you will have access to Guile debugging
commands.  For a listing of these commands, type:

@example
debug> help
@end example

Alternatively you may code the breakpoints in your LilyPond source
file using a command such as:

@example
#(set-break! my-scheme-procedure)
@end example

immediately after the @code{\include} statement.  In this case the
breakpoint will be set straight after you enter the @code{(quit)}
command at the guile prompt.

Embedding breakpoint commands like this is particularly useful if
you want to look at how the Scheme procedures in the @file{.scm}
files supplied with LilyPond work.  To do this, edit the file in
the relevant directory to add this line near the top:

@example
(use-modules (scm guile-debugger))
@end example

Now you can set a breakpoint after the procedure you are interested
in has been declared.  For example, if you are working on routines
called by @var{print-book-with} in @file{lily-library.scm}:

@example
(define (print-book-with book process-procedure)
  (let* ((paper (ly:parser-lookup '$defaultpaper))
         (layout (ly:parser-lookup '$defaultlayout))
         (outfile-name (get-outfile-name book)))
    (process-procedure book paper layout outfile-name)))

(define-public (print-book-with-defaults book)
  (print-book-with book ly:book-process))

(define-public (print-book-with-defaults-as-systems book)
  (print-book-with book ly:book-process-to-systems))
@end example

At this point in the code you could add this to set a breakpoint at
print-book-with:

@example
(set-break! print-book-with)
@end example

@subheading Tracing procedure calls and evaluator steps

Two forms of trace are available:

@example
(set-trace-call! my-scheme-procedure)
@end example

and

@example
(set-trace-subtree! my-scheme-procedure)
@end example

@code{set-trace-call!} causes Scheme to log a line to the standard
output to show when the procedure is called and when it exits.

@code{set-trace-subtree!} traces every step the Scheme evaluator
performs in evaluating the procedure.

@node Tracing object relationships
@section Tracing object relationships

Understanding the LilyPond source often boils down to figuring out what
is happening to the Grobs.  Where (and why) are they being created,
modified and destroyed? Tracing Lily through a debugger in order to
identify these relationships can be time-consuming and tedious.

In order to simplify this process, a facility has been added to
display the grobs that are created and the properties that are set
and modified.  Although it can be complex to get set up, once set up
it easily provides detailed information about the life of grobs
in the form of a network graph.

Each of the steps necessary to use the graphviz utility
is described below.

@enumerate

@item Installing graphviz

In order to create the graph of the object relationships, it is
first necessary to install Graphviz.  Graphviz is available for a
number of different platforms:

@example
@uref{http://www.graphviz.org/Download..php}
@end example

@item Modifying config.make

In order for the Graphviz tool to work, config.make must be modified.
It is probably a good idea to first save a copy of config.make under
a different name.

In order to have the required functionality available, LilyPond
needs to be compiled with the option @option{-DDEBUG}.  You can
achieve this by configuring with

@example
./configure --enable-checking
@end example

@item Rebuilding LilyPond

The executable code of LilyPond must be rebuilt from scratch:

@example
make -C lily clean && make -C lily
@end example

@item Create a graphviz-compatible @file{.ly} file

In order to use the graphviz utility, the @file{.ly} file must include
@file{ly/graphviz-init.ly}, and should then specify the
grobs and symbols that should be tracked.  An example of this
is found in @file{input/regression/graphviz.ly}.

@item Run LilyPond with output sent to a log file

The Graphviz data is sent to stderr by LilyPond, so it is
necessary to redirect stderr to a logfile:

@example
lilypond graphviz.ly 2> graphviz.log
@end example

@item Edit the logfile

The logfile has standard LilyPond output, as well as the Graphviz
output data.  Delete everything from the beginning of the file
up to but not including the first occurrence of @code{digraph}.

Also, delete the final LilyPond message about success from the end
of the file.

@item Process the logfile with @code{dot}

The directed graph is created from the log file with the program
@code{dot}:

@example
dot -Tpdf graphviz.log > graphviz.pdf
@end example

@end enumerate

The pdf file can then be viewed with any pdf viewer.

When compiled with @option{-DDEBUG}, LilyPond may run slower
than normal.  The original configuration can be restored by rerunning
@code{./configure} with @option{--disable-checking}.  Then
rebuild LilyPond with

@example
make -C lily clean && make -C lily
@end example


@node Adding or modifying features
@section Adding or modifying features

When a new feature is to be added to LilyPond, it is necessary to
ensure that the feature is properly integrated to maintain
its long-term support.  This section describes the steps necessary
for feature addition and modification.


@menu
* Write the code::
* Write regression tests::
* Write convert-ly rule::
* Automatically update documentation::
* Manually update documentation::
* Edit changes.tely::
* Verify successful build::
* Verify regression tests::
* Post patch for comments::
* Push patch::
* Closing the issues::
@end menu

@node Write the code
@subsection Write the code

You should probably create a new git branch for writing the code, as that
will separate it from the master branch and allow you to continue
to work on small projects related to master.

Please be sure to follow the rules for programming style discussed
earlier in this chapter.


@node Write regression tests
@subsection Write regression tests

In order to demonstrate that the code works properly, you will
need to write one or more regression tests.  These tests are
typically @file{.ly} files that are found in @file{input/regression}.

Regression tests should be as brief as possible to demonstrate the
functionality of the code.

Regression tests should generally cover one issue per test.  Several
short, single-issue regression tests are preferred to a single, long,
multiple-issue regression test.

If the change in the output is small or easy to overlook, use bigger
staff size -- 40 or more (up to 100 in extreme cases).  Size 30 means
"pay extra attention to details in general".

Use existing regression tests as templates to demonstrate the type of
header information that should be included in a regression test.


@node Write convert-ly rule
@subsection Write convert-ly rule

If the modification changes the input syntax, a convert-ly rule
should be written to automatically update input files from older
versions.

convert-ly rules are found in python/convertrules.py

If possible, the convert-ly rule should allow automatic updating
of the file.  In some cases, this will not be possible, so the
rule will simply point out to the user that the feature needs
manual correction.

@subsubheading Updating version numbers

If a development release occurs between you writing your patch and
having it approved+pushed, you will need to update the version
numbers in your tree.  This can be done with:

@example
scripts/auxiliar/update-patch-version old.version.number new.version.number
@end example

It will change all files in git, so use with caution and examine
the resulting diff.


@node Automatically update documentation
@subsection Automatically update documentation

@command{convert-ly} should be used to update the documentation,
the snippets, and the regression tests.  This not only makes the
necessary syntax changes, it also tests the @command{convert-ly}
rules.

The automatic updating is performed by moving to the top-level
source directory, then running:

@example
scripts/auxiliar/update-with-convert-ly.sh
@end example

If you did an out-of-tree build, pass in the relative path:

@example
LILYPOND_BUILD_DIR=../build-lilypond/ scripts/auxiliar/update-with-convert-ly.sh
@end example


@node Manually update documentation
@subsection Manually update documentation

Where the convert-ly rule is not able to automatically update the inline
LilyPond code in the documentation (i.e. if a NOT_SMART rule is used), the
documentation must be manually updated.  The inline snippets that require
changing must be changed in the English version of the docs and all
translated versions.  If the inline code is not changed in the
translated documentation, the old snippets will show up in the
English version of the documentation.

Where the convert-ly rule is not able to automatically update snippets
in Documentation/snippets/, those snippets must be manually updated.
Those snippets should be copied to Documentation/snippets/new.  The
comments at the top of the snippet describing its automatic generation
should be removed.  All translated texidoc strings should be removed.
The comment @qq{% begin verbatim} should be removed.  The syntax of
the snippet should then be manually edited.

Where snippets in Documentation/snippets are made obsolete, the snippet
should be copied to Documentation/snippets/new.  The comments and
texidoc strings should be removed as described above.  Then the body
of the snippet should be changed to:

@example
\markup @{
  This snippet is deprecated as of version X.Y.Z and
  will be removed from the documentation.
@}
@end example

@noindent
where X.Y.Z is the version number for which the convert-ly rule was
written.

Update the snippet files by running:

@example
scripts/auxiliar/makelsr.py
@end example

Where the convert-ly rule is not able to automatically update regression
tests, the regression tests in input/regression should be manually
edited.

Although it is not required, it is helpful if the developer
can write relevant material for inclusion in the Notation
Reference.  If the developer does not feel qualified to write
the documentation, a documentation editor will be able to
write it from the regression tests.  In this case the developer
should raise a new issue with the Type=Documentation tag containing
a reference to the original issue number and/or the committish of
the pushed patch so that the need for new documention is not
overlooked.

Any text that is added to or removed from the documentation should
be changed only in the English version.


@node Edit changes.tely
@subsection Edit changes.tely

An entry should be added to Documentation/changes.tely to describe
the feature changes to be implemented.  This is especially important
for changes that change input file syntax.

Hints for changes.tely entries are given at the top of the file.

New entries in changes.tely go at the top of the file.

The changes.tely entry should be written to show how the new change
improves LilyPond, if possible.


@node Verify successful build
@subsection Verify successful build

When the changes have been made, successful completion must be
verified by doing

@example
make all
make doc
@end example

When these commands complete without error, the patch is
considered to function successfully.

Developers on Windows who are unable to build LilyPond should
get help from a GNU/Linux or OSX developer to do the make tests.


@node Verify regression tests
@subsection Verify regression tests

In order to avoid breaking LilyPond, it is important to verify that
the regression tests succeed, and that no unwanted changes are
introduced into the output.  This process is described in
@ref{Regtest comparison}.

@subheading Typical developer's edit/compile/test cycle

@itemize
@item
Initial test:

@example
make [-j@var{X}]
make [-j@var{X} CPU_COUNT=@var{X}] test-baseline
make [-j@var{X} CPU_COUNT=@var{X}] check
@end example

@item
Edit/compile/test cycle:

@example
@emph{## edit source files, then...}

make clean                       @emph{## only if needed (see below)}
make [-j@var{X}]                       @emph{## only if needed (see below)}
make [-j@var{X} CPU_COUNT=@var{X}] test-redo @emph{## redo files differing from baseline}
make [-j@var{X} CPU_COUNT=@var{X}] check
@end example

@item
Reset:

@example
make test-clean
@end example
@end itemize

If you modify any source files that have to be compiled (such as
@file{.cc} or @file{.hh} files in @file{flower/} or @file{lily/}),
then you must run @command{make} before @command{make test-redo},
so @command{make} can compile the modified files and relink all
the object files.  If you only modify files which are interpreted,
like those in the @file{scm/} and @file{ly/} directories, then
@command{make} is not needed before @command{make test-redo}.

Also, if you modify any font definitions in the @file{mf/}
directory then you must run @command{make clean} and
@command{make} before running @command{make test-redo}.  This will
recompile everything, whether modified or not, and takes a lot
longer.

Running @command{make@tie{}check} will leave an HTML page
@file{out/test-results/index.html}.  This page shows all the
important differences that your change introduced, whether in the
layout, MIDI, performance or error reporting.

You only need to use @command{make test-clean} to start from
scratch, prior to running @command{make@tie{}test-baseline}. To
check new modifications, all that is needed is to repeat
@command{make@tie{}test-redo} and @command{make@tie{}test-check}
(not forgetting @command{make} if needed).




@node Post patch for comments
@subsection Post patch for comments

See @ref{Uploading a patch for review}.


@node Push patch
@subsection Push patch

Once all the comments have been addressed, the patch can be pushed.

If the author has push privileges, the author will push the patch.
Otherwise, a developer with push privileges will push the patch.


@node Closing the issues
@subsection Closing the issues

Once the patch has been pushed, all the relevant issues should be
closed.

On Rietveld, the author should log in and close the issue either by
using the @q{Edit Issue} link, or by clicking the circled x icon
to the left of the issue name.

If the changes were in response to a feature request on the Google
issue tracker for LilyPond, the author should change the status to
Fixed and a tag @q{fixed_x_y_z} should be added, where the patch was
fixed in version x.y.z.  If
the author does not have privileges to change the status, an email
should be sent to bug-lilypond requesting the BugMeister to change
the status.


@node Iterator tutorial
@section Iterator tutorial

TODO -- this is a placeholder for a tutorial on iterators

Iterators are routines written in C++ that process music expressions
and sent the music events to the appropriate engravers and/or
performers.

See a short example discussing iterators and their duties in
@ref{Articulations on EventChord}.


@node Engraver tutorial
@section Engraver tutorial

Engravers are C++ classes that catch music events and
create the appropriate grobs for display on the page.  Though the
majority of engravers are responsible for the creation of a single grob,
in some cases (e.g. @code{New_fingering_engraver}), several different grobs
may be created.

Engravers listen for events and acknowledge grobs.  Events are passed to
the engraver in time-step order during the iteration phase.  Grobs are
made available to the engraver when they are created by other engravers
during the iteration phase.


@menu
* Useful methods for information processing::
* Translation process::
* Preventing garbage collection for SCM member variables::
* Listening to music events::
* Acknowledging grobs::
* Engraver declaration/documentation::
@end menu

@node Useful methods for information processing
@subsection Useful methods for information processing

An engraver inherits the following public methods from the Translator
base class, which can be used to process listened events and acknowledged
grobs:

@itemize
@item @code{virtual void initialize ()}
@item @code{void start_translation_timestep ()}
@item @code{void process_music ()}
@item @code{void process_acknowledged ()}
@item @code{void stop_translation_timestep ()}
@item @code{virtual void finalize ()}
@end itemize

These methods are listed in order of translation time, with
@code{initialize ()} and @code{finalize ()} bookending the whole
process.  @code{initialize ()} can be used for one-time initialization
of context properties before translation starts, whereas
@code{finalize ()} is often used to tie up loose ends at the end of
translation: for example, an unterminated spanner might be completed
automatically or reported with a warning message.


@node Translation process
@subsection Translation process

At each timestep in the music, translation proceeds by calling the
following methods in turn:

@code{start_translation_timestep ()} is called before any user
information enters the translators, i.e., no property operations
(\set, \override, etc.) or events have been processed yet.

@code{process_music ()} and @code{process_acknowledged ()} are called
after all events in the current time step have been heard, or all
grobs in the current time step have been acknowledged.  The latter
tends to be used exclusively with engravers which only acknowledge
grobs, whereas the former is the default method for main processing
within engravers.

@code{stop_translation_timestep ()} is called after all user
information has been processed prior to beginning the translation for
the next timestep.


@node Preventing garbage collection for SCM member variables
@subsection Preventing garbage collection for SCM member variables

In certain cases, an engraver might need to ensure private Scheme
variables (with type SCM) do not get swept away by Guile's garbage
collector: for example, a cache of the previous key signature which
must persist between timesteps.  The method
@code{virtual derived_mark () const} can be used in such cases:

@example
Engraver_name::derived_mark ()
@{
  scm_gc_mark (private_scm_member_)
@}
@end example


@node Listening to music events
@subsection Listening to music events

External interfaces to the engraver are implemented by protected
macros including one or more of the following:

@itemize
@item @code{DECLARE_TRANSLATOR_LISTENER (event_name)}
@item @code{IMPLEMENT_TRANSLATOR_LISTENER (Engraver_name, event_name)}
@end itemize

@noindent
where @var{event_name} is the type of event required to provide the
input the engraver needs and @var{Engraver_name} is the name of the
engraver.

Following declaration of a listener, the method is implemented as follows:

@example
IMPLEMENT_TRANSLATOR_LISTENER (Engraver_name, event_name)
void
Engraver_name::listen_event_name (Stream event *event)
@{
  ...body of listener method...
@}
@end example


@node Acknowledging grobs
@subsection Acknowledging grobs

Some engravers also need information from grobs as they are created
and as they terminate.  The mechanism and methods to obtain this
information are set up by the macros:

@itemize
@item @code{DECLARE_ACKNOWLEDGER (grob_interface)}
@item @code{DECLARE_END_ACKNOWLEDGER (grob_interface)}
@end itemize

where @var{grob_interface} is an interface supported by the
grob(s) which should be acknowledged.  For example, the following
code would declare acknowledgers for a @code{NoteHead} grob (via the
@code{note-head-interface}) and any grobs which support the
@code{side-position-interface}:

@example
@code{DECLARE_ACKNOWLEDGER (note_head)}
@code{DECLARE_ACKNOWLEDGER (side_position)}
@end example

The @code{DECLARE_END_ACKNOWLEDGER ()} macro sets up a spanner-specific
acknowledger which will be called whenever a spanner ends.

Following declaration of an acknowledger, the method is coded as follows:

@example
void
Engraver_name::acknowledge_interface_name (Grob_info info)
@{
  ...body of acknowledger method...
@}
@end example

Acknowledge functions are called in the order engravers are
@code{\consist}-ed (the only exception is if you set
@code{must-be-last} to @code{#t}).

If useful things are to be done to the acknowledged grobs, this
should be deferred until all the acknowledging has finished, i.e.,
store the acknowledged grobs and process the information in a
@code{process-acknowledged ()} or @code{stop-translation-timestep ()} 
function.


@node Engraver declaration/documentation
@subsection Engraver declaration/documentation

An engraver must have a public macro

@itemize
@item @code{TRANSLATOR_DECLARATIONS (Engraver_name)}
@end itemize

@noindent
where @code{Engraver_name} is the name of the engraver.  This
defines the common variables and methods used by every engraver.

At the end of the engraver file, one or both of the following
macros are generally called to document the engraver in the
Internals Reference:

@itemize
@item @code{ADD_ACKNOWLEDGER (Engraver_name, grob_interface)}
@item @code{ADD_TRANSLATOR (Engraver_name, Engraver_doc,
    Engraver_creates, Engraver_reads, Engraver_writes)}
@end itemize

@noindent
where @code{Engraver_name} is the name of the engraver, @code{grob_interface}
is the name of the interface that will be acknowledged,
@code{Engraver_doc} is a docstring for the engraver,
@code{Engraver_creates} is the set of grobs created by the engraver,
@code{Engraver_reads} is the set of properties read by the engraver,
and @code{Engraver_writes} is the set of properties written by
the engraver.

The @code{ADD_ACKNOWLEDGER} and @code{ADD_TRANSLATOR} macros use a
non-standard indentation system.  Each interface, grob, read property,
and write property is on its own line, and the closing parenthesis
and semicolon for the macro all occupy a separate line beneath the final
interface or write property.  See existing engraver files for more
information.


@node Callback tutorial
@section Callback tutorial

TODO -- This is a placeholder for a tutorial on callback functions.


@node Understanding pure properties
@section Understanding pure properties

@menu
* Purity in LilyPond::
* Writing a pure function::
* How purity is defined and stored::
* Where purity is used::
* Case studies::
* Debugging tips::
@end menu

Pure properties are some of the most difficult properties to understand
in LilyPond but, once understood, it is much easier to work with
horizontal spacing.  This document provides an overview of what it means
for something to be @q{pure} in LilyPond, what this purity guarantees,
and where pure properties are stored and used.  It finishes by
discussing a few case studies for the pure programmer to save you some
time and to prevent you some major headaches.


@node Purity in LilyPond
@subsection Purity in LilyPond
Pure properties in LilyPond are properties that do not have any
@q{side effects}.
That is, looking up a pure property should never result in calls to the
following functions:
@itemize
@item @code{set_property}
@item @code{set_object}
@item @code{suicide}
@end itemize
This means that, if the property is calculated via a callback, this callback
must not only avoid the functions above but make sure that any functions
it calls also avoid the functions above.  Also, to date in LilyPond, a pure
function will always return the same value before line breaking (or, more
precisely, before any version of @code{break_into_pieces} is called).  This
convention makes it possible to cache pure functions and be more flexible
about the order in which functions are called. For example; Stem #'length has
a pure property that will @emph{never} trigger one of the functions listed
above and will @emph{always} return the same value before line breaking,
independent of where it is called.  Sometimes, this will be the actual length
of the Stem.  But sometimes it will not.  For example; stem that links up
with a beam will need its end set to the Y position of the beam at the stem's
X position.  However, the beam's Y positions can only be known after the score
is broken up in to several systems (a beam that has a shallow slope on a
compressed line of music, for example, may have a steeper one on an
uncompressed line).  Thus, we only call the impure version of the properties
once we are @emph{absolutely certain} that all of the parameters needed to
calculate their final value have been calculated.  The pure version provides a
useful estimate of what this Stem length (or any property) will be, and
the art of creating good pure properties is trying to get the estimation
as close to the actual value as possible.

Of course, like Gregory Peck and Tintin, some Grobs will have properties
that will always be pure.  For example, the height of a note-head in
not-crazy music will never depend on line breaking or other parameters
decided late in the typesetting process.  Inversely, in rare cases,
certain properties are difficult to estimate with pure values.  For
example, the height of a Hairpin at a certain cross-section of its
horizontal span is difficult to know without knowing the horizontal
distance that the hairpin spans, and LilyPond provides an
over-estimation by reporting the pure height as the entire height of the
Hairpin.

Purity, like for those living in a convent, is more like a contract than
an @emph{a priori}.  If you write a pure-function, you are promising
the user (and the developer who may have to clean up after you) that
your function will not be dependent on factors that change at different
stages of the compilation process (compilation of a score, not of
LilyPond).

One last oddity is that purity, in LilyPond, is currently limited
exclusively to things that have to do with Y-extent and positioning.
There is no concept of @q{pure X} as, by design, X is always the
independent variable (i.e. from column X1 to column X2, what will be the
Y height of a given grob).  Furthermore, there is no purity for
properties like color, text, and other things for which a meaningful notion
of estimation is either not necessary or has not yet been found.  For example,
even if a color were susceptible to change at different points of the
compilation process, it is not clear what a pure estimate of this color
would be or how this pure color could be used.  Thus, in this document and
in the source, you will see purity discussed almost interchangeably with
Y-axis positioning issues.


@node Writing a pure function
@subsection Writing a pure function
Pure functions take, at a minimum, three arguments: the @var{grob}, the
starting column at which the function is being evaluated (hereafter
referred to as @var{start}), and the end column at which the grob is
being evaluated (hereafter referred to as @var{end}).  For items,
@var{start} and @var{end} must be provided (meaning they are not optional)
but will not have a meaningful impact on the result, as items only occupy
one column and will thus yield a value or not (if they are not in the range
from @var{start} to @var{end}).  For spanners however, @var{start} and
@var{end} are important, as we may can get a better pure estimation of a
slice of the spanner than considering it on the whole.  This is useful
during line breaking, for example, when we want to estimate the Y-extent
of a spanner broken at given starting and ending columns.


@node How purity is defined and stored
@subsection How purity is defined and stored
Purity is defined in LilyPond with the creation of an unpure-pure container
(unpure is not a word, but hey, neither was LilyPond until the 90s).  For example:

@example
#(define (foo grob)
  '(-1 . 1))

#(define (bar grob start end)
  '(-2 . 2))

\override Stem #'length = #(ly:make-unpure-pure-container foo bar)
@end example

Note that items can only ever have two pure heights: their actual pure height
if they are between @q{start} and @q{end}, or an empty interval if they are
not.  Thus, their pure property is cached to speed LilyPond up.  Pure
heights for spanners are generally not cached as they change depending
on the start and end values.  They are only cached in certain particular
cases.  Before writing a lot of caching code, make sure that it is a
value that will be reused a lot.


@node Where purity is used
@subsection Where purity is used
Pure Y values must be used in any functions that are called before
line breaking.  Examples of this can be seen in
@code{Separation_items::boxes} to construct horizontal skylines and in
@code{Note_spacing::stem_dir_correction} to correct for optical
illusions in spacing.  Pure properties are also used in the calculation
of other pure properties.  For example, the @code{Axis_group_interface}
has pure functions that look up other pure functions.

Purity is also implicitly used in any functions that should only ever
return pure values.  For example, extra-spacing-height is only ever used
before line-breaking and thus should never use values that would only be
available after line breaking.  In this case, there is no need to create
callbacks with pure equivalents because these functions, by design, need
to be pure.

To know if a property will be called before and/or after line-breaking
is sometimes tricky and can, like all things in coding, be found by
using a debugger and/or adding @var{printf} statements to see where they
are called in various circumstances.


@node Case studies
@subsection Case studies
In each of these case studies, we expose a problem in pure properties, a
solution, and the pros and cons of this solution.

@subheading Time signatures
A time signature needs to prevent accidentals from passing over or under
it, but its extent does not necessarily extend to the Y-position of
accidentals.  LilyPond's horizontal spacing sometimes makes a line of
music compact and, when doing so, allows certain columns to pass over
each other if they will not collide.  This type of passing over is not
desirable with time signatures in traditional engraving.  But how do we
know if this passing over will happen before line breaking, as we are
not sure what the X positions will be?  We need a pure estimation of how
much extra spacing height the time signatures would need to prevent this
form of passing over without making this height so large as to
overly-distort the Y-extent of an system, which could result in a very
@q{loose} looking score with lots of horizontal space between columns.
So, to approximate this extra spacing height, we use the Y-extent of a
time signature's next-door-neighbor grobs via the pure-from-neighbor
interface.

@itemize
@item pros: By extending the extra spacing height of a time signature to
that of its next-door-neighbors, we make sure that grobs to the right of
it that could pass above or below it do not.

@item cons: This over-estimation of the vertical height could prevent
snug vertical spacing of systems, as the system will be registered as
being taller at the point of the time signature than it actually is.
This approach can be used for clefs and bar lines as well.
@end itemize

@subheading Stems
As described above, Stems need pure height approximations when they are
beamed, as we do not know the beam positions before line breaking.  To
estimate this pure height, we take all the stems in a beam and find
their pure heights as if they were not beamed.  Then, we find the union
of all these pure heights and take the intersection between this
interval (which is large) and an interval going from the note-head of a
stem to infinity in the direction of the stem so that the interval stops
at the note head.

@itemize
@item pros: This is guaranteed to be at least as long as the beamed
stem, as a beamed stem will never go over the ideal length of the
extremal beam of a stem.

@item cons: Certain stems will be estimated as being too long, which
leads to the same problem of too-much-vertical-height as described
above.

@end itemize


@node Debugging tips
@subsection Debugging tips
A few questions to ask yourself when working with pure properties:

@itemize
@item Is the property really pure?  Are you sure that its value could
not be changed later in the compiling process due to other changes?

@item Can the property be made to correspond even more exactly with the
eventual impure property?

@item For a spanner, is the pure property changing correctly depending
on the starting and ending points of the spanner?

@item For an Item, will the item's pure height need to act in horizontal
spacing but not in vertical spacing?  If so, use extra-spacing-height
instead of pure height.

@end itemize


@node LilyPond scoping
@section LilyPond scoping

The LilyPond language has a concept of scoping, i.e. you can do:

@example
foo = 1

#(begin
   (display (+ foo 2)))
@end example

@noindent with @code{\paper}, @code{\midi} and @code{\header} being
nested scope inside the @file{.ly} file-level scope.  @w{@code{foo = 1}}
is translated in to a scheme variable definition.

This implemented using modules, with each scope being an anonymous
module that imports its enclosing scope's module.

LilyPond's core, loaded from @file{.scm} files, is usually placed in the
@code{lily} module, outside the @file{.ly} level.  In the case of

@example
lilypond a.ly b.ly
@end example

@noindent
we want to reuse the built-in definitions, without changes effected in
user-level @file{a.ly} leaking into the processing of @file{b.ly}.

The user-accessible definition commands have to take care to avoid
memory leaks that could occur when running multiple files.  All
information belonging to user-defined commands and markups is stored in
a manner that allows it to be garbage-collected when the module is
dispersed, either by being stored module-locally, or in weak hash
tables.


@node Scheme->C interface
@section Scheme->C interface

Most of the C functions interfacing with Guile/Scheme used in LilyPond
are described in the API Reference of the
@uref{http://www.gnu.org/software/guile/manual/html_node/index.html,
GUILE Reference Manual}.

The remaining functions are defined in @file{lily/lily-guile.cc},
@file{lily/include/lily-guile.hh} and
@file{lily/include/lily-guile-macros.hh}.
Although their names are meaningful there's a few things you should know
about them.

@menu
* Comparison::
* Conversion::
@end menu

@node Comparison
@subsection Comparison

This is the trickiest part of the interface.

Mixing Scheme values with C comparison operators won't produce any crash
or warning when compiling but must be avoided:

@example
scm_string_p (scm_value) == SCM_BOOL_T
@end example

As we can read in the reference, @code{scm_string_p} returns a Scheme
value: either @code{#t} or @code{#f} which are written @code{SCM_BOOL_T}
and @code{SCM_BOOL_F} in C.  This will work, but it is not following
to the API guidelines.  For further information, read this discussion:

@smallexample
@uref{http://lists.gnu.org/archive/html/lilypond-devel/2011-08/msg00646.html}
@end smallexample

There are functions in the Guile reference that returns C values
instead of Scheme values.  In our example, a function called
@code{scm_is_string} (described after @code{string?} and @code{scm_string_p})
returns the C value 0 or 1.

So the best solution was simply:

@example
scm_is_string (scm_value)
@end example

There a simple solution for almost every common comparison.  Another example:
we want to know if a Scheme value is a non-empty list.  Instead of:

@example
(scm_is_true (scm_list_p (scm_value)) && scm_value != SCM_EOL)
@end example

one can usually use:

@example
scm_is_pair (scm_value)
@end example

since a list of at least one member is a pair.  This test is
cheap; @code{scm_list_p} is actually quite more complex since it makes
sure that its argument is neither a `dotted list' where the last pair
has a non-null @code{cdr}, nor a circular list.  There are few
situations where the complexity of those tests make sense.

Unfortunately, there is not a @code{scm_is_[something]} function for
everything.  That's one of the reasons why LilyPond has its own Scheme
interface.  As a rule of thumb, tests that are cheap enough to be
worth inlining tend to have such a C interface.  So there is
@code{scm_is_pair} but not @code{scm_is_list}, and @code{scm_is_eq}
but not @code{scm_is_equal}.

@subheading General definitions

@subsubheading bool to_boolean (SCM b)

Return @code{true} if @var{b} is @code{SCM_BOOL_T}, else return @code{false}.

This should be used instead of @code{scm_is_true} and
@code{scm_is_false} for properties since in LilyPond, unset properties
are read as an empty list, and by convention unset Boolean properties
default to false.  Since both @code{scm_is_true} and
@code{scm_is_false} only compare with @code{##f} in line with what
Scheme's conditionals do, they are not really useful for checking the
state of a Boolean property.

@subsubheading bool ly_is_[something] (args)

Behave the same as scm_is_[something] would do if it existed.

@subsubheading bool is_[type] (SCM s)

Test whether the type of @var{s} is [type].
[type] is a LilyPond-only set of values (direction, axis...).  More
often than not, the code checks LilyPond specific C++-implemented
types using

@subsubheading [Type *] unsmob<Type> (SCM s)

This tries converting a Scheme object to a pointer of the desired
kind.  If the Scheme object is of the wrong type, a pointer value
of@w{ }@code{0} is returned, making this suitable for a Boolean test.

@node Conversion
@subsection Conversion

@subheading General definitions

@subsubheading bool to_boolean (SCM b)

Return @code{true} if @var{b} is @code{SCM_BOOL_T}, else return @code{false}.

This should be used instead of @code{scm_is_true} and @code{scm_is_false}
for properties since empty lists are sometimes used to unset them.

@subsubheading [C type] ly_scm2[C type] (SCM s)

Behave the same as scm_to_[C type] would do if it existed.

@subsubheading [C type] robust_scm2[C type] (SCM s, [C type] d)

Behave the same as scm_to_[C type] would do if it existed.
Return @var{d} if type verification fails.


@node LilyPond miscellany
@section LilyPond miscellany

This is a place to dump information that may be of use to developers
but doesn't yet have a proper home.  Ideally, the length of this section
would become zero as items are moved to other homes.


@menu
* Spacing algorithms::
* Info from Han-Wen email::
* Music functions and GUILE debugging::
* Articulations on EventChord::
@end menu

@node Spacing algorithms
@subsection Spacing algorithms

Here is information from an email exchange about spacing algorithms.

On Thu, 2010-02-04 at 15:33 -0500, Boris Shingarov wrote:
I am experimenting with some modifications to the line breaking code,
and I am stuck trying to understand how some of it works.  So far my
understanding is that Simple_spacer operates on a vector of Grobs, and
it is a well-known Constrained-QP problem (rods = constraints, springs
= quadratic function to minimize).  What I don't understand is, if the
spacer operates at the level of Grobs, which are built at an earlier
stage in the pipeline, how are the changes necessitated by differences
in line breaking, taken into account?  in other words, if I take the
last measure of a line and place it on the next line, it is not just a
matter of literally moving that graphic to where the start of the next
line is, but I also need to draw a clef, key signature, and possibly
other fundamental things -- but at that stage in the rendering
pipeline, is it not too late??

Joe Neeman answered:

We create lots of extra grobs (eg. a BarNumber at every bar line) but
most of them are not drawn.  See the break-visibility property in
item-interface.

Here is another e-mail exchange.  Janek Warchoł asked for a starting point
to fixing 1301 (change clef colliding with notes).  Neil Puttock replied:

The clef is on a loose column (it floats before the head), so the
first place I'd look would be lily/spacing-loose-columns.cc (and
possibly lily/spacing-determine-loose-columns.cc).
I'd guess the problem is the way loose columns are spaced between
other columns: in this snippet, the columns for the quaver and tuplet
minim are so close together that the clef's column gets dumped on top
of the quaver (since it's loose, it doesn't influence the spacing).

@node Info from Han-Wen email
@subsection Info from Han-Wen email

In 2004, Douglas Linhardt decided to try starting a document that would
explain LilyPond architecture and design principles.  The material below
is extracted from that email, which can be found at
@uref{http://thread.gmane.org/gmane.comp.gnu.lilypond.devel/2992}.
The headings reflect questions from Doug or comments from Han-Wen;
the body text are Han-Wen's answers.

@subheading Figuring out how things work.

I must admit that when I want to know how a program works, I use grep
and emacs and dive into the source code.  The comments and the code
itself are usually more revealing than technical documents.

@subheading What's a grob, and how is one used?

Graphical object - they are created from within engravers, either as
Spanners (derived class) -slurs, beams- or Items (also a derived
class) -notes, clefs, etc.

There are two other derived classes System (derived from Spanner,
containing a "line of music") and Paper_column (derived from Item, it
contains all items that happen at the same moment).  They are separate
classes because they play a special role in the linebreaking process.

@subheading What's a smob, and how is one used?

A C(++) object that is encapsulated so it can be used as a Scheme
object.  See GUILE info, "19.3 Defining New Types (Smobs)"

@subheading When is each C++ class constructed and used?

@itemize

@item
Music classes

In the parser.yy see the macro calls MAKE_MUSIC_BY_NAME().

@item
Contexts

Constructed during "interpreting" phase.

@item
Engravers

Executive branch of Contexts, plugins that create grobs, usually one
engraver per grob type.  Created  together with context.

@item
Layout Objects

= grobs

@item
Grob Interfaces

These are not C++ classes per se.  The idea of a Grob interface hasn't
crystallized well.  ATM, an interface is a symbol, with a bunch of grob
properties.  They are not objects that are created or destroyed.

@item
Iterators

Objects that walk through different music classes, and deliver events
in a synchronized way, so that notes that play together are processed
at the same moment and (as a result) end up on the same horizontal position.

Created during interpreting phase.

BTW, the entry point for interpreting is ly:run-translator
(ly_run_translator on the C++ side)

@end itemize

@subheading Can you get to Context properties from a Music object?

You can create music object with a Scheme function that reads context
properties (the \applycontext syntax).  However, that function is
executed during Interpreting, so you can not really get Context
properties from Music objects, since music objects are not directly
connected to Contexts.  That connection is made by the  Music_iterators

@subheading Can you get to Music properties from a Context object?

Yes, if you are given the music object within a Context
object.  Normally, the music objects enter Contexts in synchronized
fashion, and the synchronization is done by Music_iterators.

@subheading What is the relationship between C++ classes and Scheme objects?

Smobs are C++ objects in Scheme.  Scheme objects (lists, functions) are
manipulated from C++ as well using the GUILE C function interface
(prefix: scm_)

@subheading How do Scheme procedures get called from C++ functions?

scm_call_*, where * is an integer from 0 to 4.
Also scm_c_eval_string (), scm_eval ()

@subheading How do C++ functions get called from Scheme procedures?

Export a C++ function to Scheme with LY_DEFINE.

@subheading What is the flow of control in the program?

Good question.  Things used to be clear-cut, but we have Scheme
and SMOBs now, which means that interactions do not follow a very
rigid format anymore.  See below for an overview, though.

@subheading Does the parser make Scheme procedure calls or C++ function calls?

Both.  And the Scheme calls can call C++ and vice versa.  It's nested,
with the SCM datatype as lubrication between the interactions

(I think the word "lubrication" describes the process better than the
traditional word "glue")

@subheading How do the front-end and back-end get started?

Front-end: a file is parsed, the rest follows from that.  Specifically,

Parsing leads to a Music + Music_output_def object (see parser.yy,
definition of toplevel_expression )

A Music + Music_output_def object leads to a Global_context object (see
ly_run_translator ())

During interpreting, Global_context + Music leads to a bunch of
Contexts (see Global_translator::run_iterator_on_me ()).

After interpreting, Global_context contains a Score_context (which
contains staves, lyrics etc.) as a child.  Score_context::get_output ()
spews a Music_output object (either a Paper_score object for notation
or Performance object for MIDI).

The Music_output object is the entry point for the backend (see
ly_render_output ()).

The main steps of the backend itself are in

@itemize

@item
@file{paper-score.cc} , Paper_score::process_

@item
@file{system.cc} , System::get_lines()

@item
The step, where things go from grobs to output, is in
System::get_line(): each grob delivers a Stencil (a Device
independent output description), which is interpreted by our
outputting backends (@file{scm/output-tex.scm} and
@file{scm/output-ps.scm}) to produce TeX and PS.

@end itemize

Interactions between grobs and putting things into .tex and .ps files
have gotten a little more complex lately.  Jan has implemented
page-breaking, so now the backend also involves Paper_book,
Paper_lines and other things.  This area is still heavily in flux, and
perhaps not something you should want to look at.

@subheading How do the front-end and back-end communicate?

There is no communication from backend to front-end.  From front-end to
backend is simply the program flow: music + definitions gives
contexts, contexts yield output, after processing, output is written
to disk.

@subheading Where is the functionality associated with KEYWORDs?

See @file{my-lily-lexer.cc} (keywords, there aren't that many)
and @file{ly/*.ly} (most of the other backslashed @code{/\words} are identifiers)

@subheading What Contexts/Properties/Music/etc. are available when they are processed?

What do you mean exactly with this question?

See @file{ly/engraver-init.ly} for contexts,
see @file{scm/define-*.scm} for other objects.

@subheading How do you decide if something is a Music, Context, or Grob property?
Why is part-combine-status a Music property when it seems (IMO)
to be related to the Staff context?

The Music_iterators and Context communicate through two channels

Music_iterators can set and read context properties, idem for
Engravers and Contexts

Music_iterators can send "synthetic" music events (which aren't in
the input) to a context.  These are caught by Engravers.  This is
mostly a one way communication channel.

part-combine-status is part of such a synthetic event, used by
Part_combine_iterator to communicate with Part_combine_engraver.


@subheading Deciding between context and music properties

I'm adding a property to affect how \autochange works.  It seems to
me that it should be a context property, but the Scheme autochange
procedure has a Music argument.  Does this mean I should use
a Music property?

\autochange is one of these extra strange beasts: it requires
look-ahead to decide when to change staves.  This is achieved by
running the interpreting step twice (see
@file{scm/part-combiner.scm} , at the bottom), and
storing the result of the first step (where to switch
staves) in a Music property.  Since you want to influence that
where-to-switch list, your must affect the code in
make-autochange-music (@file{scm/part-combiner.scm}).
That code is called directly from the parser and there are no
official "parsing properties" yet, so there is no generic way
to tune \autochange.  We would have to invent something new
for this, or add a separate argument,

@example
    \autochange #around-central-C ..music..
@end example

@noindent
where around-central-C is some function that is called from
make-autochange-music.

@subheading More on context and music properties

From Neil Puttock, in response to a question about transposition:

Context properties (using \set & \unset) are tied to engravers: they
provide information relevant to the generation of graphical objects.

Since transposition occurs at the music interpretation stage, it has
no direct connection with engravers: the pitch of a note is fixed
before a notehead is created.  Consider the following minimal snippet:

@example
@{ c' @}
@end example

This generates (simplified) a NoteEvent, with its pitch and duration
as event properties,

@example
(make-music
  'NoteEvent
  'duration
  (ly:make-duration 2 0 1 1)
  'pitch
  (ly:make-pitch 0 0 0)
@end example

which the Note_heads_engraver hears.  It passes this information on to
the NoteHead grob it creates from the event, so the head's correct
position and duration-log can be determined once it's ready for
printing.

If we transpose the snippet,

@example
\transpose c d @{ c' @}
@end example

the pitch is changed before it reaches the engraver (in fact, it
happens just after the parsing stage with the creation of a
TransposedMusic music object):

@example
(make-music
 'NoteEvent
 'duration
 (ly:make-duration 2 0 1 1)
 'pitch
 (ly:make-pitch 0 1 0)
@end example

You can see an example of a music property relevant to transposition:
untransposable.

@example
\transpose c d @{ c'2 \withMusicProperty #'untransposable ##t c' @}
@end example

-> the second c' remains untransposed.

Take a look at @file{lily/music.cc} to see where the transposition takes place.


@subheading How do I tell about the execution environment?

I get lost figuring out what environment the code I'm looking at is in when it
executes.  I found both the C++ and Scheme autochange code.  Then I was trying
to figure out where the code got called from.  I finally figured out that the
Scheme procedure was called before the C++ iterator code, but it took me a
while to figure that out, and I still didn't know who did the calling in the
first place.  I only know a little bit about Flex and Bison, so reading those
files helped only a little bit.

@emph{Han-Wen:} GDB can be of help here.  Set a breakpoint in C++, and run.  When you
hit the breakpoint, do a backtrace.  You can inspect Scheme objects
along the way by doing

@example
p ly_display_scm(obj)
@end example

this will display OBJ through GUILE.

@node Music functions and GUILE debugging
@subsection Music functions and GUILE debugging

Ian Hulin was trying to do some debugging in music functions, and
came up with the following question (edited and adapted to current
versions):

HI all,
I'm working on the Guile Debugger Stuff, and would like to try
debugging a music function definition such as:

@example
conditionalMark =
#(define-music-function () ()
  #@{ \tag instrumental-part @{\mark \default@} #@} )
@end example

It appears @code{conditionalMark} does not get set up as an
equivalent of a Scheme

@example
(define conditionalMark = define-music-function () () ...
@end example

@noindent
although something gets defined because Scheme apparently recognizes

@example
#(set-break! conditionalMark)
@end example

@noindent
later on in the file without signalling any Guile errors.

However the breakpoint trap is never encountered as
@code{define-music-function} passed things on to
@code{ly:make-music-function}, which is really C++ code
@code{ly_make_music_function}, so Guile never finds out about the
breakpoint.


The answer in the mailing list archive at that time was less than
helpful.  The question already misidentifies the purpose of
@code{ly:make-music-function} which is only called once at the
time of @emph{defining} @code{conditionalMark} but is not involved
in its later @emph{execution}.

Here is the real deal:

A music function is not the same as a GUILE function.  It boxes
both a proper Scheme function (with argument list and body from
the @code{define-music-function} definition) along with a call
signature representing the @emph{types} of both function and
arguments.

Those components can be reextracted using
@code{ly:music-function-extract} and
@code{ly:music-function-signature}, respectively.

When LilyPond's parser encounters a music function call in its
input, it reads, interprets, and verifies the arguments
individually according to the call signature and @emph{then} calls
the proper Scheme function.

While it is actually possible these days to call a music function
@emph{as if} it were a Scheme function itself, this pseudo-call
uses its own wrapping code matching the argument list @emph{as a
whole} to the call signature, substituting omitted optional
arguments with defaults and verifying the result type.

So putting a breakpoint on the music function itself will still
not help with debugging uses of the function using LilyPond
syntax.

However, either calling mechanism ultimately calls the proper
Scheme function stored as part of the music function, and that is
where the breakpoint belongs:

@example
#(set-break! (ly:music-function-extract conditionalMark))
@end example

will work for either calling mechanism.

@node Articulations on EventChord
@subsection Articulations on EventChord

From David Kastrup's email
@uref{http://lists.gnu.org/archive/html/lilypond-devel/2012-02/msg00189.html}:

LilyPond's typesetting does not act on music expressions and music
events.  It acts exclusively on stream events.  It is the act of
iterators to convert a music expression into a sequence of stream events
played in time order.

The EventChord iterator is pretty simple: it just takes its "elements"
field when its time comes up, turns every member into a StreamEvent and
plays that through the typesetting process.  The parser currently
appends all postevents belonging to a chord at the end of "elements",
and thus they get played at the same point of time as the elements of
the chord.  Due to this design, you can add per-chord articulations or
postevents or even assemble chords with a common stem by using parallel
music providing additional notes/events: the typesetter does not see a
chord structure or postevents belonging to a chord, it just sees a
number of events occuring at the same point of time in a Voice context.

So all one needs to do is let the EventChord iterator play articulations
after elements, and then adding to articulations in EventChord is
equivalent to adding them to elements (except in cases where the order
of events matters).