-# executes os.system(command) if infile is newer than
-# outfile or outfile don't exist
-#
-def compile (command, workingdir, infile, outfile):
- "Test if infile is newer than outfile. If so, cd to workingdir"
- "and execute command"
- indate = file_mtime (workingdir+infile)
- try:
- outdate = file_mtime (workingdir+outfile)
- recompile = indate > outdate
-
- except os.error:
- recompile = 1
-
- if recompile:
- sys.stderr.write ('invoking `%s\'\n' % command)
- if workingdir == '':
- status = os.system (command)
- else:
- status = os.system ('cd %s; %s' %(workingdir, command))
- if status:
- raise CompileStatus
-
-class Properties:
- #
- # init
- #
- def __init__(self):
- self.__linewidth = {
- 1: {'a4':{10: 345, 11: 360, 12: 390},
- 'a5':{10: 276, 11: 276, 12: 276},
- 'b5':{10: 345, 11: 356, 12: 356},
- 'letter':{10: 345, 11: 360, 12: 390},
- 'legal': {10: 345, 11: 360, 12: 390},
- 'executive':{10: 345, 11: 360, 12: 379}},
- 2: {'a4':{10: 167, 11: 175, 12: 190},
- 'a5':{10: 133, 11: 133, 12: 133},
- 'b5':{10: 167, 11: 173, 12: 173},
- 'letter':{10: 167, 11: 175, 12: 190},
- 'legal':{10: 167, 11: 175, 12: 190},
- 'executive':{10: 167, 11: 175, 12: 184}}}
- # >0 --> force all mudela to this pt size
- self.force_mudela_fontsize = 0
- self.force_verbatim_b = 0
- self.__data = {
- 'mudela-fontsize' : {'init': 16},
- 'papersize' : {'init': 'a4'},
- 'num-column' : {'init': 1},
- 'tex-fontsize' : {'init': 10}
- }
- def clear_for_new_file(self):
- for var in self.__data.keys():
- self.__data[var] = {'init': self.__data[var]['init']}
- def clear_for_new_block(self):
- for var in self.__data.keys():
- if self.__data[var].has_key('block'):
- del self.__data[var]['block']
- def __get_variable(self, var):
- if self.__data[var].has_key('block'):
- return self.__data[var]['block']
- elif self.__data[var].has_key('file'):
- return self.__data[var]['file']
- else:
- return self.__data[var]['init']
- def setPapersize(self, size, requester):
- self.__data['papersize'][requester] = size
- def getPapersize(self):
- return self.__get_variable('papersize')
- def setMudelaFontsize(self, size, requester):
- self.__data['mudela-fontsize'][requester] = size
- def getMudelaFontsize(self):
- if self.force_mudela_fontsize:
- return self.force_mudela_fontsize
- return self.__get_variable('mudela-fontsize')
- def setTexFontsize(self, size, requester):
- self.__data['tex-fontsize'][requester] = size
- def getTexFontsize(self):
- return self.__get_variable('tex-fontsize')
- def setNumColumn(self, num, requester):
- self.__data['num-column'][requester] = num
- def getNumColumn(self):
- return self.__get_variable('num-column')
- def getLineWidth(self):
- return self.__linewidth[self.getNumColumn()][self.getPapersize()][self.getTexFontsize()]
-
-
-class Mudela_output:
- def __init__ (self, basename):
- self.basename = basename
- self.temp_filename = "%s/%s" %(outdir, 'mudela-temp.ly')
- self.file = open (self.temp_filename, 'w')
- self.__lines = []
- # 'tex' or 'eps'
- self.graphic_type = 'tex'
- self.code_type = 'unknown'
- def write (self, line):
- # match only if there is nothing but whitespace before \begin HACK
- if re.search('^\s*\\\\begin{mudela}', line):
- self.scan_begin_statement(line)
- else:
- if self.code_type == 'unknown':
- if re.search('^\s*\\\\score', line) or \
- re.search('^\s*\\\\paper', line) or \
- re.search('^\s*\\\\header', line) or \
- re.search('^\s*[A-Za-z]*\s*=', line):
- self.code_type = 'ly'
- self.__lines.append(line)
- def scan_begin_statement(self, line):
- r = begin_mudela_opts_re.search(line)
- if r:
- o = r.group()[1:-1]
- optlist = re.compile('[\s,]*').split(o)
- else:
- optlist = []
- if 'fragment' in optlist:
- print "warning: obsolete option: fragment"
- if 'floating' in optlist:
- print "warning: obsolete option: floating, change to eps"
- if 'eps' in optlist:
- self.graphic_type = 'eps'
- for pt in fontsize_pt2i.keys():
- if pt in optlist:
- Props.setMudelaFontsize(fontsize_pt2i[pt], 'block')
- def write_red_tape(self):
- self.file.write ('\\include \"paper%d.ly\"\n' \
- % Props.getMudelaFontsize())
-
- s = fontsize_i2a[Props.getMudelaFontsize()]
- if self.code_type == 'fly':
- linewidth_str = 'linewidth = -1.\cm;'
- else:
- linewidth_str = 'linewidth = %i.\\pt;' % Props.getLineWidth()
- self.file.write("\\paper {"
- + "\\paper_%s " % s
- + linewidth_str
- + "castingalgorithm = \Gourlay; \n}")
- #+ "castingalgorithm = \Wordwrap; indent = 2.\cm; \n}")
- if self.code_type == 'fly':
- self.file.write('\\score{\n\\notes{')
- def close (self):
- if self.code_type == 'unknown':
- self.code_type = 'fly'
- if self.code_type == 'ly':
- self.write_red_tape()
- for l in self.__lines:
- self.file.write(l)
- else:
- self.write_red_tape()
- for l in self.__lines:
- self.file.write(l)
- self.file.write('}}')
-
- self.file.close()
-
- inf = outdir + self.basename + '.ly'
- outf = outdir + self.basename + '.tex'
- if not os.path.isfile(inf):
- status = 1
- else:
- status = os.system ('diff -q %s %s' % (self.temp_filename, inf))
- if status:
- os.rename (self.temp_filename, inf)
- if need_recompile_b(inf, outf):
- out_files.append((self.graphic_type, inf))
- def insert_me_string(self):
- "Returns a string that can be used directly in latex."
- if self.graphic_type == 'tex':
- return ['tex', self.basename]
- elif self.graphic_type == 'eps':
- return ['eps', self.basename]
- else:
- raise SomethingIsSeriouslyBroken
-
-class Tex_output:
- def __init__ (self, name):
- self.output_fn = '%s/%s' % (outdir, name)
- self.__lines = []
- def open_verbatim (self, line):
- self.__lines.append('\\begin{verbatim}\n')
- s = re.sub('[\s,]*verbatim[\s]*', '', line)
- s = re.sub('\[\]', '', s)
- self.__lines.append(s);
- def close_verbatim (self):
- self.__lines.append('\\end{verbatim}\n')
- def write (self, s):
- self.__lines.append(s)
- def create_graphics(self):
- s = ''
- g_vec = []
- for line in self.__lines:
- if type(line)==type([]):
- g_vec.append(line)
- for g in g_vec:
- if need_recompile_b(outdir+g[1]+'.ly', outdir+g[1]+'.tex'):
- s = s + ' ' + g[1]+'.ly'
- if s != '':
- e = os.system('cd %s; lilypond %s' %(outdir, s))
- if e:
- print "error: lilypond exited with value", e
- sys.exit(e)
- for g in g_vec:
- if g[0] == 'eps':
- compile('tex %s' % g[1]+'.tex', outdir, g[1]+'.tex', g[1]+'.dvi')
- compile('dvips -E -o %s %s' %(g[1]+'.eps', g[1]+'.dvi'), outdir,
- g[1]+'.dvi', g[1]+'.eps')
- def write_outfile(self):
- file = open(self.output_fn+'.latex', 'w')
- file.write('% Created by mudela-book\n')
- for line in self.__lines:
- if type(line)==type([]):
- if line[0] == 'tex':
- file.write('\\preMudelaExample\\input %s\n\postMudelaExample '\
- # TeX applies the prefix of the main source automatically.
- % (line[1]+'.tex'))
-# % (outdir+line[1]+'.tex'))
- if line[0] == 'eps':
- ps_dim = ps_dimention(outdir+line[1]+'.eps')
- file.write('\\parbox{%ipt}{\includegraphics{%s}}\n' \
- % (ps_dim[0], line[1]+'.eps'))
-# % (ps_dim[0], outdir+line[1]+'.eps'))
- else:
- file.write(line)
- file.close()
-class Tex_input:
- def __init__ (self, filename):
- for fn in [filename, filename+'.tex', filename+'.doc']:
- try:
- self.infile = open (fn)
- self.filename = fn
- return
- except:
- continue
- raise IOError
- def get_lines (self):
- lines = self.infile.readlines ()
- (retlines, retdeps) = ([],[self.filename])
- for line in lines:
- r_inp = input_re.search (line)
- r_inc = include_re.search (line)
-
- # Filename rules for \input :
- # input: no .tex ext
- # 1. will search for file with exact that name (tex-input.my will be found)
- # 2. will search for file with .tex ext, (tex-input.my
- # will find tex-input.my.tex)
- # input: with .tex ext
- # 1. will find exact match
-
- # Filename rules for \include :
- # 1. will only find files with name given to \include + .tex ext
- if r_inp:
- try:
- t = Tex_input (r_inp.groups()[0])
- ls = t.get_lines ()
- retlines = retlines + ls[0]
- retdeps = retdeps + ls[1]
- except:
- print "warning: can't find %s, let's hope latex will" \
- % r_inp.groups()[0]
- retlines.append (line)
- elif r_inc:
- try:
- t = Tex_input (r_inc.groups()[0]+'.tex')
- ls =t.get_lines ()
- ls[0].insert(0, '\\newpage\n')
- ls[0].append('\\newpage\n')
- retlines = retlines + ls[0]
- retdeps = retdeps + ls[1]
- except:
- print "warning: can't find %s, let's hope latex will" \
- % r_inc.groups()[0]
- retlines.append (line)
- else:
- r_mud = defined_mudela_cmd_re.search(line)
- if r_mud:
- ss = "\\\\verb(?P<xx>[^a-zA-Z])\s*\\\\%s\s*(?P=xx)" \
- % re.escape(r_mud.group()[1:])
- if re.search(ss, line):
- retlines.append(line)
- continue
- while 1:
- opts = r_mud.groups()[2]
- if opts == None:
- opts = ''
- else:
- opts = ', '+opts
- (start, rest) = string.split(line, r_mud.group(), 1)
- retlines.append(start+'\n')
- v = string.split(defined_mudela_cmd[r_mud.groups()[0]], '\n')
- for l in v[1:-1]:
- l = re.sub(r'\\fontoptions', opts, l)
- l = re.sub(r'\\maininput', r_mud.groups()[3], l)
- retlines.append(l)
- r_mud = defined_mudela_cmd_re.search(rest)
- if not r_mud:
- retlines.append(rest)
- break;
- line = rest
- else:
- retlines.append (line)
- else:
- return (retlines, retdeps)
-
-
-class Main_tex_input(Tex_input):
- def __init__ (self, name, outname):
-
- Tex_input.__init__ (self, name) # ugh
- self.outname = outname
- self.chapter = 0
- self.section = 0
- self.fine_count =0
- self.mudtex = Tex_output (self.outname)
- self.mudela = None
- self.deps = []
- self.verbatim = 0
- # set to 'mudela' when we are processing mudela code,
- # both verbatim and graphic-to-be
- self.mode = 'latex'
- def set_sections (self, l):
- if section_re.search (l):
- self.section = self.section + 1
- if chapter_re.search (l):
- self.section = 0
- self.chapter = self.chapter + 1
-
- def gen_basename (self):
- return '%s-%d.%d.%d' % (self.outname, self.chapter,
- self.section, self.fine_count)
- def extract_papersize_from_documentclass(self, line):
- pre = extract_papersize_re.search(line)
- if not pre:
- return None
- return pre.groups()[0]
- def extract_fontsize_from_documentclass(self, line):
- r = extract_fontsize_re.search(line)
- if r:
- return int(r.groups()[0])
- def do_it(self):
- preMudelaDef = postMudelaDef = 0
- (lines, self.deps) = self.get_lines ()
- for line in lines:
- if documentclass_re.search (line):
- p = self.extract_papersize_from_documentclass (line)
- if p:
- Props.setPapersize(p, 'file')
- f = self.extract_fontsize_from_documentclass (line)
- if f:
- Props.setTexFontsize (f, 'file')
- elif twocolumn_re.search (line):
- Props.setNumColumn (2, 'file')
- elif onecolumn_re.search (line):
- Props.setNumColumn (1, 'file')
- elif preMudelaExample_re.search (line):
- preMudelaDef = 1
- elif postMudelaExample_re.search (line):
- postMudelaDef = 1
- elif begin_document_re.search (line):
- if not preMudelaDef:
- self.mudtex.write ('\\def\\preMudelaExample{}\n')
- if not postMudelaDef:
- self.mudtex.write ('\\def\\postMudelaExample{}\n')
- elif begin_mudela_re.search (line):
- Props.clear_for_new_block()
- if __debug__:
- if self.mode == 'mudela':
- raise AssertionError
- self.mode = 'mudela'
- r = begin_mudela_opts_re.search (line)
- if r:
- o = r.group()[1:][:-1]
- optlist = re.compile('[ ,]*').split(o)
- else:
- optlist = []
- if ('verbatim' in optlist) or (Props.force_verbatim_b):
- self.verbatim = 1
- self.mudtex.open_verbatim (line)
- continue
- else:
- self.verbatim = 0
- self.mudela = Mudela_output (self.gen_basename ())
- elif end_mudela_re.search (line):
- if __debug__:
- if self.mode != 'mudela':
- raise AssertionError
- if self.mudela:
- self.mudela.close ()
- self.mudtex.write (self.mudela.insert_me_string())
- del self.mudela
- self.mudela = None
- self.fine_count = self.fine_count + 1
- else:
- self.mudtex.write (line)
- self.mudtex.close_verbatim ()
- self.mode = 'latex'
- continue
-
- if self.mode == 'mudela' and not self.verbatim:
- self.mudela.write (line)
- else:
- self.mudtex.write (line)
- self.set_sections(line)
- self.mudtex.create_graphics()
- self.mudtex.write_outfile()
- del self.mudtex
+
+default_music_fontsize = 16
+default_text_fontsize = 12
+
+# latex linewidths:
+# indices are no. of columns, papersize, fontsize
+# Why can't this be calculated?
+latex_linewidths = {
+ 1: {'a4':{10: 345, 11: 360, 12: 390},
+ 'a5':{10: 276, 11: 276, 12: 276},
+ 'b5':{10: 345, 11: 356, 12: 356},
+ 'letter':{10: 345, 11: 360, 12: 390},
+ 'legal': {10: 345, 11: 360, 12: 390},
+ 'executive':{10: 345, 11: 360, 12: 379}},
+ 2: {'a4':{10: 167, 11: 175, 12: 190},
+ 'a5':{10: 133, 11: 133, 12: 133},
+ 'b5':{10: 167, 11: 173, 12: 173},
+ 'letter':{10: 167, 11: 175, 12: 190},
+ 'legal':{10: 167, 11: 175, 12: 190},
+ 'executive':{10: 167, 11: 175, 12: 184}}}
+
+
+options = [
+ ('DIM', '', 'default-mudela-fontsize', 'default fontsize for music. DIM is assumed to in points'),
+ ('EXT', 'f', 'format', 'set format. EXT is one of texi and latex.'),
+ ('', 'h', 'help', 'print help'),
+ ('DIR', 'I', 'include', 'include path'),
+ ('', '', 'init', 'mudela-book initfile'),
+# ('DIM', '', 'force-mudela-fontsize', 'force fontsize for all inline mudela. DIM is assumed to in points'),
+ ('', '', 'force-verbatim', 'make all mudela verbatim'),
+ ('', 'M', 'dependencies', 'write dependencies'),
+ ('', 'n', 'no-lily', 'don\'t run lilypond'),
+ ('FILE', 'o', 'outname', 'prefix for filenames'),
+ ('', 'v', 'version', 'print version information' )
+ ]
+
+format = ''
+run_lilypond = 1
+use_hash = 1
+no_match = 'a\ba'
+
+# format specific strings, ie. regex-es for input, and % strings for output
+output_dict= {
+ 'latex': {
+ 'output-mudela-fragment' : r"""\begin[eps,fragment%s]{mudela}
+ \context Staff <
+ \context Voice{
+ %s
+ }
+ >
+\end{mudela}""",
+ 'output-mudela':r"""\begin[%s]{mudela}
+%s
+\end{mudela}""",
+ 'output-verbatim': r"""\begin{verbatim}%s\end{verbatim}""",
+ 'output-default-post': r"""\def\postMudelaExample{}""",
+ 'output-default-pre': r"""\def\preMudelaExample{}""",
+ 'output-eps': '\\noindent\\parbox{\\mudelaepswidth{%s.eps}}{\includegraphics{%s.eps}}',
+ 'output-tex': '\\preMudelaExample \\input %s.tex \\postMudelaExample\n'
+ },
+ 'texi' : {'output-mudela': """@mudela[%s]
+%s
+@end mudela
+""",
+ 'output-verbatim': r"""@example
+%s
+@end example
+""",
+
+# do some tweaking: @ is needed in some ps stuff.
+# override EndLilyPondOutput, since @tex is done
+# in a sandbox, you can't do \input lilyponddefs at the
+# top of the document.
+ 'output-all': r"""@tex
+\catcode`\@=12
+\input lilyponddefs
+\def\EndLilyPondOutput{}
+\input %s.tex
+\catcode`\@=0
+@end tex
+@html
+<img src=%s.png>
+@end html
+""",
+ }
+ }
+
+def output_verbatim (body):
+ if __main__.format == 'texi':
+ body = re.sub ('([@{}])', '@\\1', body)
+ return get_output ('output-verbatim') % body
+
+re_dict = {
+ 'latex': {'input': '\\\\input{?([^}\t \n}]*)',
+ 'include': '\\\\include{([^}]+)}',
+
+ 'comma-sep' : ', *',
+ 'header': r"""\\documentclass(\[.*?\])?""",
+ 'preamble-end': '\\\\begin{document}',
+ 'verbatim': r"""(?s)\\begin{verbatim}(.*?)\\end{verbatim}""",
+ 'verb': r"""\\verb(.)(.*?)\1""",
+ 'mudela-file': '\\\\mudelafile(\[[^\\]]+\])?{([^}]+)}',
+ 'mudela' : '\\\\mudela(\[.*?\])?{(.*?)}',
+ 'mudela-block': r"""(?s)\\begin(\[.*?\])?{mudela}(.*?)\\end{mudela}""",
+ 'interesting-cs': '\\\\(chapter|section|twocolumn|onecolumn)',
+ 'def-post-re': r"""\\def\\postMudelaExample""",
+ 'def-pre-re': r"""\\def\\preMudelaExample""",
+ },
+ 'texi': {'input': '@include[ \n\t]+([^\t \n]*)',
+ 'include': no_match,
+ 'header': no_match,
+ 'preamble-end': no_match,
+ 'verbatim': r"""(?s)@example(.*?)@end example$""",
+ 'verb': r"""@code{(.*?)}""",
+ 'mudela-file': '@mudelafile(\[[^\\]]+\])?{([^}]+)}',
+ 'mudela' : '@mudela(\[.*?\])?{(.*?)}',
+ 'mudela-block': r"""(?s)@mudela(\[.*?\])?(.*?)@end mudela""",
+ 'interesting-cs': r"""[\\@](chapter|section)""",
+ 'comma-sep' : ', *',
+ }
+ }
+
+
+for r in re_dict.keys ():
+ olddict = re_dict[r]
+ newdict = {}
+ for k in olddict.keys ():
+ newdict[k] = re.compile (olddict[k])
+ re_dict[r] = newdict
+
+
+def uniq (list):
+ list.sort ()
+ s = list
+ list = []
+ for x in s:
+ if x not in list:
+ list.append (x)
+ return list
+
+
+def get_output (name):
+ return output_dict[format][name]
+
+def get_re (name):
+ return re_dict[format][name]
+
+
+def bounding_box_dimensions(fname):
+ try:
+ fd = open(fname)
+ except IOError:
+ error ("Error opening `%s'" % fname)
+ str = fd.read (-1)
+ s = re.search('%%BoundingBox: ([0-9]+) ([0-9]+) ([0-9]+) ([0-9]+)', str)
+ if s:
+ return (int(s.group(3))-int(s.group(1)),
+ int(s.group(4))-int(s.group(2)))
+ else:
+ return (0,0)
+
+
+def find_file (name):
+ for a in include_path:
+ try:
+ nm = os.path.join (a, name)
+ f = open (nm)
+ return nm
+ except IOError:
+ pass
+ return ''
+
+def error (str):
+ sys.stderr.write (str + "\n Exiting ... \n\n")
+ raise 'Exiting.'
+
+
+def compose_full_body (body, opts):
+ "Construct the text of an input file: add stuff to BODY using OPTS as options."
+ paper = 'a4'
+ music_size = default_music_fontsize
+ latex_size = default_text_fontsize
+
+ cols = 1
+ for o in opts:
+ m = re.search ('^(.*)paper$', o)
+ if m:
+ paper = m.group (1)
+
+
+ m = re.match ('([0-9]+)pt', o)
+ if m:
+ music_size = string.atoi(m.group (1))
+
+ m = re.match ('latexfontsize=([0-9]+)pt', o)
+ if m:
+ latex_size = string.atoi (m.group (1))
+
+
+ if 'twocolumn' in opts:
+ cols = 2