💾 Archived View for runjimmyrunrunyoufuckerrun.com › src › foreign › abcmidi › doc › programming › a… captured on 2021-12-17 at 13:26:06.
-=-=-=-=-=-=-
Some Notes on the abc2midi Code
-------------------------------
written by Seymour Shlien
Abc2midi.txt - last updated 29 November 1999.
This file provides an algorithmic description of the program
abc2midi which converts an abc file into a midi file.
The sources of abc2midi now comprising of 6700 lines of C code are
contained in the following five files.
parseabc.c is the front end which scans the abc file and invokes
the appropriate event handler for each element it encounters
(eg. bar lines, notes, chords etc.) It happens to be the
front end for other programs such as abc2ps, yaps and
abc2abc.
store.c contains all the event handlers which are invoked from
parseabc. It converts the abc file into an internal
representation described later in this file.
genmidi.c converts this internal representation in an output midi
file.
queues.c are some utilities needed by genmidi.
midifile.c is a library of public domain routines to read and write
midifiles.
In order to handle the multiple voices, parts, repeats and accompaniments
which occur in abc files, the program performs multiple passes. The
midi file stores the different voices, accompaniment and lyrics in
individual tracks, so it is necessary to separates these parts prior
to writing it to the midi file. If there are repeats in the music,
something which appears only once in the abc file may be invoked twice
when creating the output midi file.
Parseabc.c
----------
The parser has been written so that it should be possible to write
your own utility to handle the interpreted abc and link it with
the parser code. This is very similar to the way that the midifilelib
utilities work. The parser is designed so that it can write out
almost exactly what it reads in. This means that in some cases
the level of parsing is fairly primitive and it may be necessary to make
use of routines in store.c which perform further processing on an item.
In the first phase of parsing, the abc file is read and when, say, a note
is encountered, the routine event_note() is called. The code for event_note
is in the file store.c. Encountering an X in the abc generally causes a
routine called event_X() to be called. An internal representation of the tune
is built up and when the end of an abc tune is reached, a little bit of
processing is done on the internal representation before the routine
mywritetrack() is called to actually write out the MIDI file.
The main program to abc2midi is contained in this file. Since this
main program is used to start other programs such as yaps, it is
fairly simple. It calls event_init defined in store.c which obtains
all the user parameters (eg. input filename) and then executes the
top level procedure, parsefile.
The procedure parsefile (filename), opens the specified file (or stdin)
and processes it one character at a time. The characters fill a line
buffer which is passed to the procedure parseline whenever a linebreak
character (eg. linefeed) is encountered. By doing it in this fashion,
abc2midi is able to eliminate the extra carriage return which occurs
in DOS files.
The procedure parseline first checks for blank lines and comments. A
blank line signals the end of the tune and the event_blankline is called
to initiate the next stage in processing. If it is neither a comment or
blank line, the procedure must decide whether the line is a a field line
(eg."X:2") or part of a music body (eg. "|:C>DGz|..."). This is decided
using the following rule. If the first letter begins with one of the field
letter commands and is immediately followed by a ":", then it is treated
as a field line and parsefield is called. Otherwise if a K: field line
was already encountered (variable inbody set to 1), then it is treated
as a line of music. If neither case is satisfied, then the line is
intepreted as plain text.
The procedure parsefield identifies the particular field line type
and invokes the appropriate support function or event handler. If the
field command occurs after the first K: line, then only certain field
lines are legal. (For example it is illegal to place a T: (title) line
after the K: (key signature) line.)
The procedure parsemusic is the most complex procedure in the file and
recognizes the musical notes, chords, guitar chord names, bar lines,
repeats, slurs, ties, graces etc. Appropriate event handlers and
support functions are called to complete the tasks of parsing the
information. The parsenote function, for example, must check for
decorations, accidentals, octave shifts as well as identifying the
note and its determining its duration.
Unlike the other software modules, parseabc.c contains few global
variables. The global variables linenum, inhead, inbody, parsing,
slur, parserinchord indicate which lexical component of the abc
tune is being processed.
store.c
-------
This is the most complex module consisting of almost 3000 lines
of source code. This module contains all of the event handlers which
may be invoked in the file parseabc.c. The main purpose of these
handlers are to build up an internal representation of the abc file
in the global memory arrays. This is a complete representation
which allows the regeneration of the abc file (eg. abc2abc) or
the creation of a midi file. The internal representation is used
by the genmidi.c software to create the midi file.
Global variables
The internal representation is stored in global variables defined
at the beginning of the file. A description of the most important
variables is given here.
The music is stored in the four arrays, feature, pitch, num, denom.
These arrays contains a list of the lexical components in the order
that they have been encountered or created.
The array "feature" identifies the type of component which can be one
of the following enum features.
SINGLE_BAR, DOUBLE_BAR, BAR_REP, REP_BAR, REP1, REP2, BAR1,
REP_BAR2, DOUBLE_REP, THICK_THIN, THIN_THICK, PART, TEMPO,
TIME, KEY, REST, TUPLE, NOTE, NONOTE, OLDTIE, TEXT, SLUR_ON,
SLUR_OFF, TIE, TITLE, CHANNEL, TRANSPOSE, RTRANSPOSE, GRACEON,
GRACEOFF, SETGRACE, SETC, GCHORD, GCHORDON, GCHORDOFF, VOICE,
CHORDON, CHORDOFF, SLUR_TIE, TNOTE, LT, GT, DYNAMIC, LINENUM,
MUSICLINE, MUSICSTOP, WORDLINE, WORDSTOP
The array pitch[] mainly stores the pitch of a musical note,
represented in the same manner as in a midi file. Thus middle
C has a value of 60. The array has other uses, for example
in a LINENUM feature it stores the line number relative to
the X: reference command where the lexical feature has been
detected. The LINENUM feature is useful when reporting warnings
or errors in the input file. Duration of the notes are represented
as a fraction, where the standard unit is defined by the L:
command. If feature[n] is a NOTE, then num[n] and denom[n]
contain the numerator and denominator of this fraction.
Here is an example of the contents of these arrays for the
following small file in the same order that they are stored in
these arrays.
X: 1
T: sample
M: 2/4
L: 1/8
K: G
|: C>D[EF]G |C4:|
Feature Pitch Num Denom
LINENUM 2 0 0
TITLE 0 0 0
LINENUM 3 0 0
LINENUM 4 0 0
LINENUM 5 0 0
DOUBLE_BAR 0 0 0
LINENUM 6 0 0
MUSICLINE 0 0 0
BAR_REP 0 0 0
NOTE 60 2 3
NOTE 62 1 3
CHORDON 0 0 0
NOTE 64 1 2
NOTE 66 1 2
CHORDOFF 0 1 2
NOTE 67 1 2
SINGLE_BAR 0 0 0
NOTE 60 2 1
REP_BAR 0 0 0
MUSIC_STOP 0 0 0
LINENUM 7 0 0
In order to support multivoice abc files in the form
V:1
| ...| ...|....
V:2
| ...| ...|....
%
V:1
| ...| ...|....
V:2
| ...| ...|....
etc.
abc2midi maintains a voicecontext structure for each voice.
This allows each voice to define its own key signature, default
note length using internal field commands. There is a head
voicecontext which is used when no voice field commands are
defined in the abc file. The v[] array maintains a set of
all voices active in the abc file and voicecount keeps track
of the number of voices.
Similarly, abcmidi maintains a list of all the part stored
in the feature[], pitch[], num[] and denom[] arrays.
All text items (eg. title and some other fields) are stored
in char atext[][] arrays, so that they can be included in
the midi output file. The textual information is repeated
in each track of the output midi file.
Following the conventions in the midi file, the program uses
the quarter note as the standard unit of time instead of the
whole note. In contrast, the standard length in the abc file
is based on the whole note. For example in L:1/8, the fraction
refers to a whole note. In order to convert time units to
quarter notes, the numerator of the time specifications
for note lengths is multiplied by 4 when it is added to
the feature list.
Table of contents of procedures in store.c
getarg(option, argc, argv) - gets run time arguments.
newvoice(n) - creates a new voice context.
getvoicecontext() - finds or makes a voice context.
clearvoicecontexts() - frees up the memory of the voice context
event_init() - called by main program
event_text(s) - called when a line of text is encountered.
event_reserved(p) - handles reserved character codes H-Z.
event_tex(s) - called whenever a TeX command is encountered.
event_linebreak() - called whenever a newline character is encountered.
event_startmusicline() - called for each new music line.
event_endmusicline() - called at the end of each music line.
event_eof() - end of abc file encountered.
event_fatal_error(s) - reports fatal error.
event_error(s) - reports an error.
event_warning(s) - reports a potential problem.
event_comment(s) - called whenever a comment is encountered.
event_specific(package, s) - recognizes %%package s.
event_startinline() - beginning of an in-music field command.
event_closeinline() - finishes an in-music field command.
event_field(k,f) - for R:, T: and any other unprocessed field commands.
event_words(p) - processes a w: field command.
char_out(list,out,ch) - for building up a parts list in a P: command.
read_spec() - converts P:A(AB)3 to P:AABABAB etc.
event_part(s) - handles a P: field.
char_break() - reports error for improper voice change.
event_voice(n,s) - processes a V: field.
event_length(n) - recognizes L: field
event_blankline - starts finishfile() procedure.
event_refno(n) - recognizes X:
event_tempo(n, a, b, rel) - recognizes Q:
event_timesig(n, m) - recognizes M:
event_key(sharps, s, minor, modmap, modmul) - recognizes K:
event_graceon() - start of grace notes, "{" encountered.
event_graceoff() - end of grace notes, "}" encountered.
event_rep1() - handles first repeat indicated by [1.
event_rep2() - handles second repeat indicated by [2.
event_slur(t) -called when s encountered in abc.
event_sluron(t) called when "(" is encountered in abc.
event_sluroff(t) called when ")" is encountered in abc.
slurtotie() - converts slur into tied notes.
event_tie() - handles tie indication "-".
event_rest(n,m) - processes rest indication Z or z.
event_bar(type) - processes various types of bar lines.
event_space() - space character encountered. Ignored here.
event_linend(ch,n) - handles line continuation at end of line (eg. \).
event_broken(type, mult) - handles >, <, >> etc. in abc.
event_tuple(n,q,r) - handles triplets and general tuplets.
event_chord() - called whenever + is encountered.
marknotestart() - remembers last few notes in voice context.
marknoteend() - this is used to process broken rhythms (eg. >).
marknote() - called for single note (as opposed to chord).
event_chordon() - called whenever [ is encountered.
event_chordoff() - called whenever ] is encountered.
splitstring(s,sep,handler) - splits string with separator sep.
event_instuction(s) - handles !...! event.
getchordnumber(s) - searches known chords for chord s.
addchordname(s, len, notes) - adds chord name to known chord list.
event_gchord(s) - handles guitar chords.
event_handle_gchord(s) - handler for guitar chords.
event_handle_instruction(s) - handles dynamic indications (eg. !ppp!).
event_finger(p) - handles 1,2,...5 in guitar chord field (does nothing).
hornp(num,denom) - modifies rhythm to hornpipe.
event_note(roll, staccato, updown, accidental, mult, note, octave, n, m)
doroll(note,octave,n,m,pitch) - applies roll to note.
dotrill(note,octave,n,m,pitch) - applies trill to note.
pitchof(note,accidental,mult,octave) -finds MIDI pitch value
setmap(sf,map,mult) - converts key signature to accidental map.
altermap(v,modmap,modmul) - modifies accidental map.
copymap(v) - sets up working accidental map at beginning of bar.
addfeature(f,p,n,d) - places feature in internal tables.
autoextend(maxnotes) - increase memory limits for feature arrays.
textextend(maxstrings, stringarray) - resize array pointers to strings.
myputc(c) - workaround for problems with PCC compiler.
tiefix() - connect up tied notes.
dotie(j,xinchord) - called in preprocessing stage to handle ties.
addfrac(xnum,xdenom,a,b) - add a/b to the number of units in bar.
applybroken(place, type, n) - adjust length of broken notes.
brokenadjust() -adjust length of broken notes.
applygrace() - assign lengths to grace notes.
dograce() - assign lengths to grace notes.
lenmul(n, a, b) - multiply num(n),denom(n) by a/b.
zerobar() - start a new count of beats in the bar.
delendrep() - remove bogus repeat.
placeendrep(j) - patch up missing repeat.
placestartrep(j) - patch up missing repeat.
fixreps() - find and correct missing repeats in music.
startfile() - initialization performed after an event_refno.
tempounits(t_num, t_denom) - interprets Q: field.
setbeat() - sets default gchord command for time signature.
headerprocess() - called after the first K: field in tune.
finishfile() - starts next stage of processing when end of tune
is encountered.
All the functions in this file respond to event calls from parseabc.
Once the internal representation of the abc file is completed, the
procedure finishfile is called to perform some clean up and create
the midi file. An internal representation of the midi file is
first created and then it is written onto the designated output file.
As finishfile provides the link to the next module, genmidi.c, here
is a brief description of what it does.
proc finishfile performs several passes through the internal
representation to clean up the graces (dograce), the tied notes
(tiefix) and any unbalanced repeats. It then calls writetrack(i)
for each midi track to create the internal midi representation
and finally the midi representation is recorded in the output
file.
genmidi.c
---------
The procedure finishfile described above, creates each midi track
by calling the function writetrack which is defined here. To create
a track from the internal representation, the program must find all
the parts and put them in order with all the repeats. In addition, if
it contains karaoke text, this too must be placed in a separate track.
Any chordal accompaniment is generated from the guitar chord indications
and placed in another track. For multivoice and multipart music, a voice
may be missing in a particular part. If the voice is missing, the
procedure fillvoice ensures that all voices remain properly aligned when
the voice reappears in another part.
Here is a simple road map to the important procedures included in this
file.
dodeferred is here used to handle any dynamic indications (eg. !ppp!)
which may be contained in the file. The %%MIDI messages are stored
in a atext string which is pointed to by the contents of the pitch[]
array.
checkbar is called each time a bar line is encountered and reports
a warning if the wrong number of beats occur.
Transitions between parts are handled by the procedure partbreak.
There are a number of procedures for handling karoake text --
karaokestarttrack(), findwline(startline), getword(place,w),
write_syllable(place) and checksyllables().
For the first track, the meter, tempo and key signature are recorded
using the functions set_meter(n,m), write_meter(n,m), write_keysig(sf,mi).
Chordal accompaniment is produced by the procedure dogchords(i).
queues.c
--------
For each midi note, it is necessary to send a note-on and a note-off
instruction. When polyphonic music is played on the same track, keeping
track of the time to issue a note-off instruction may get complicated.
The procedures in this file are used to maintain a linked list for the
notes to be turned off. The notes are put into the list in the order
that they are encountered but the order in which to issue note-off
commands is maintained by the links. As many as 50 notes playing
simultaneously can be handled by the list.
Addendum
--------
The following section contains clarifications on various components
of abc2midi.
29 June 2003
Treatment of octave shifts.
The key signature field command K: has provision for shifting
a note using either transpose= or octave= subcommands. Both
of these functions operate quite differently and deserve some
description especially for multivoiced tunes.
The octave shift, is performed in event_note in store.c, using
the cached value v.octaveshift which is stored in the global
voicecontext structure, v. There is a structure for each voice
in the abc file. Whenever a new V: command is encountered,
event_voice (store.c) is invoked, which swaps the appropriate
voices structure into the global v array using the function
getvoicecontext(n). If getvoicecontext cannot find a cached
structure for that voice, then a new voice structure is created
and added to the linked list of voice structures. The v.octaveshift
variable is updated by event_octave which is called by event_key
(store.c) which is called by parsekey in parseabc.c
(Comment: it is not too clear how an octave switch is
handled in the middle of a repeat section. i.e. does the old
value get restored when repeating.)
(Description of transpose shift is in CHANGES July 1 2003.)