lyx_mirror/lib/configure.py
Julien Rioux 69c17f11ae Convert odg to (cropped) eps without external script.
git-svn-id: svn://svn.lyx.org/lyx/lyx-devel/trunk@39986 a592a061-630c-0410-9148-cb99ea01b6c8
2011-10-25 23:33:35 +00:00

1382 lines
64 KiB
Python

#! /usr/bin/env python
# -*- coding: utf-8 -*-
#
# file configure.py
# This file is part of LyX, the document processor.
# Licence details can be found in the file COPYING.
# \author Bo Peng
# Full author contact details are available in file CREDITS.
import glob, logging, os, re, shutil, subprocess, sys
# set up logging
logging.basicConfig(level = logging.DEBUG,
format = '%(levelname)s: %(message)s', # ignore application name
filename = 'configure.log',
filemode = 'w')
#
# Add a handler to log to console
console = logging.StreamHandler()
console.setLevel(logging.INFO) # the console only print out general information
formatter = logging.Formatter('%(message)s') # only print out the message itself
console.setFormatter(formatter)
logger = logging.getLogger('LyX')
logger.addHandler(console)
def writeToFile(filename, lines, append = False):
" utility function: write or append lines to filename "
if append:
file = open(filename, 'a')
else:
file = open(filename, 'w')
file.write(lines)
file.close()
def addToRC(lines):
''' utility function: shortcut for appending lines to outfile
add newline at the end of lines.
'''
if lines.strip() != '':
writeToFile(outfile, lines + '\n', append = True)
logger.debug('Add to RC:\n' + lines + '\n\n')
def removeFiles(filenames):
'''utility function: 'rm -f'
ignore errors when file does not exist, or is a directory.
'''
for file in filenames:
try:
os.remove(file)
logger.debug('Removing file %s' % file)
except:
logger.debug('Failed to remove file %s' % file)
pass
def cmdOutput(cmd):
'''utility function: run a command and get its output as a string
cmd: command to run
'''
if os.name == 'nt':
b = False
cmd = 'cmd /d /c ' + cmd
else:
b = True
pipe = subprocess.Popen(cmd, shell=b, close_fds=b, stdin=subprocess.PIPE, \
stdout=subprocess.PIPE, universal_newlines=True)
pipe.stdin.close()
output = pipe.stdout.read()
pipe.stdout.close()
return output.strip()
def setEnviron():
''' I do not really know why this is useful, but we might as well keep it.
NLS nuisances.
Only set these to C if already set. These must not be set unconditionally
because not all systems understand e.g. LANG=C (notably SCO).
Fixing LC_MESSAGES prevents Solaris sh from translating var values in set!
Non-C LC_CTYPE values break the ctype check.
'''
os.environ['LANG'] = os.getenv('LANG', 'C')
os.environ['LC'] = os.getenv('LC_ALL', 'C')
os.environ['LC_MESSAGE'] = os.getenv('LC_MESSAGE', 'C')
os.environ['LC_CTYPE'] = os.getenv('LC_CTYPE', 'C')
def createDirectories():
''' Create the build directories if necessary '''
for dir in ['bind', 'clipart', 'doc', 'examples', 'images', 'kbd', \
'layouts', 'scripts', 'templates', 'ui' ]:
if not os.path.isdir( dir ):
try:
os.mkdir( dir)
logger.debug('Create directory %s.' % dir)
except:
logger.error('Failed to create directory %s.' % dir)
sys.exit(1)
def checkTeXPaths():
''' Determine the path-style needed by the TeX engine on Win32 (Cygwin) '''
windows_style_tex_paths = ''
if LATEX == '':
return windows_style_tex_paths
if os.name == 'nt' or sys.platform == 'cygwin':
from tempfile import mkstemp
fd, tmpfname = mkstemp(suffix='.ltx')
if os.name == 'nt':
from ctypes import windll, create_unicode_buffer
GetShortPathName = windll.kernel32.GetShortPathNameW
longname = unicode(tmpfname)
shortlen = GetShortPathName(longname, 0, 0)
shortname = create_unicode_buffer(shortlen)
if GetShortPathName(longname, shortname, shortlen):
inpname = shortname.value.replace('\\', '/')
else:
inpname = tmpfname.replace('\\', '/')
else:
inpname = cmdOutput('cygpath -m ' + tmpfname)
logname = os.path.basename(inpname.replace('.ltx', '.log'))
inpname = inpname.replace('~', '\\string~')
os.write(fd, r'\relax')
os.close(fd)
latex_out = cmdOutput(r'latex "\nonstopmode\input{%s}\makeatletter\@@end"' % inpname)
if 'Error' in latex_out:
latex_out = cmdOutput(r'latex "\nonstopmode\input{\"%s\"}\makeatletter\@@end"' % inpname)
if 'Error' in latex_out:
logger.warning("configure: TeX engine needs posix-style paths in latex files")
windows_style_tex_paths = 'false'
else:
logger.info("configure: TeX engine needs windows-style paths in latex files")
windows_style_tex_paths = 'true'
removeFiles([tmpfname, logname, 'texput.log'])
return windows_style_tex_paths
## Searching some useful programs
def checkProg(description, progs, rc_entry = [], path = [], not_found = ''):
'''
This function will search a program in $PATH plus given path
If found, return directory and program name (not the options).
description: description of the program
progs: check programs, for each prog, the first word is used
for searching but the whole string is used to replace
%% for a rc_entry. So, feel free to add '$$i' etc for programs.
path: additional pathes
rc_entry: entry to outfile, can be
1. emtpy: no rc entry will be added
2. one pattern: %% will be replaced by the first found program,
or '' if no program is found.
3. several patterns for each prog and not_found. This is used
when different programs have different usages. If you do not
want not_found entry to be added to the RC file, you can specify
an entry for each prog and use '' for the not_found entry.
not_found: the value that should be used instead of '' if no program
was found
'''
# one rc entry for each progs plus not_found entry
if len(rc_entry) > 1 and len(rc_entry) != len(progs) + 1:
logger.error("rc entry should have one item or item for each prog and not_found.")
sys.exit(2)
logger.info('checking for ' + description + '...')
## print '(' + ','.join(progs) + ')',
global java, perl
for idx in range(len(progs)):
# ac_prog may have options, ac_word is the command name
ac_prog = progs[idx]
ac_word = ac_prog.split(' ')[0]
if (ac_word.endswith('.class') or ac_word.endswith('.jar')) and java == '':
continue
if ac_word.endswith('.pl') and perl == '':
continue
msg = '+checking for "' + ac_word + '"... '
path = os.environ["PATH"].split(os.pathsep) + path
extlist = ['']
if "PATHEXT" in os.environ:
extlist = extlist + os.environ["PATHEXT"].split(os.pathsep)
for ac_dir in path:
for ext in extlist:
if os.path.isfile( os.path.join(ac_dir, ac_word + ext) ):
logger.info(msg + ' yes')
# deal with java and perl
if ac_word.endswith('.class'):
ac_prog = ac_prog.replace(ac_word, r'%s \"%s\"' % (java, os.path.join(ac_dir, ac_word[:-6])))
elif ac_word.endswith('.jar'):
ac_prog = ac_prog.replace(ac_word, r'%s -jar \"%s\"' % (java, os.path.join(ac_dir, ac_word)))
elif ac_word.endswith('.pl'):
ac_prog = ac_prog.replace(ac_word, r'%s -w \"%s\"' % (perl, os.path.join(ac_dir, ac_word)))
# write rc entries for this command
if len(rc_entry) == 1:
addToRC(rc_entry[0].replace('%%', ac_prog))
elif len(rc_entry) > 1:
addToRC(rc_entry[idx].replace('%%', ac_prog))
return [ac_dir, ac_word]
# if not successful
logger.info(msg + ' no')
# write rc entries for 'not found'
if len(rc_entry) > 0: # the last one.
addToRC(rc_entry[-1].replace('%%', not_found))
return ['', not_found]
def checkProgAlternatives(description, progs, rc_entry = [], alt_rc_entry = [], path = [], not_found = ''):
'''
The same as checkProg, but additionally, all found programs will be added
as alt_rc_entries
'''
# one rc entry for each progs plus not_found entry
if len(rc_entry) > 1 and len(rc_entry) != len(progs) + 1:
logger.error("rc entry should have one item or item for each prog and not_found.")
sys.exit(2)
logger.info('checking for ' + description + '...')
## print '(' + ','.join(progs) + ')',
found_prime = False
real_ac_dir = ''
real_ac_word = not_found
global java, perl
for idx in range(len(progs)):
# ac_prog may have options, ac_word is the command name
ac_prog = progs[idx]
ac_word = ac_prog.split(' ')[0]
if (ac_word.endswith('.class') or ac_word.endswith('.jar')) and java == '':
continue
if ac_word.endswith('.pl') and perl == '':
continue
msg = '+checking for "' + ac_word + '"... '
path = os.environ["PATH"].split(os.pathsep) + path
extlist = ['']
if "PATHEXT" in os.environ:
extlist = extlist + os.environ["PATHEXT"].split(os.pathsep)
found_alt = False
for ac_dir in path:
for ext in extlist:
if os.path.isfile( os.path.join(ac_dir, ac_word + ext) ):
logger.info(msg + ' yes')
pr = re.compile(r'(\\\S+)(.*)$')
m = None
# deal with java and perl
if ac_word.endswith('.class'):
ac_prog = ac_prog.replace(ac_word, r'%s \"%s\"' % (java, os.path.join(ac_dir, ac_word[:-6])))
elif ac_word.endswith('.jar'):
ac_prog = ac_prog.replace(ac_word, r'%s -jar \"%s\"' % (java, os.path.join(ac_dir, ac_word)))
elif ac_word.endswith('.pl'):
ac_prog = ac_prog.replace(ac_word, r'%s -w \"%s\"' % (perl, os.path.join(ac_dir, ac_word)))
# write rc entries for this command
if found_prime == False:
if len(rc_entry) == 1:
addToRC(rc_entry[0].replace('%%', ac_prog))
elif len(rc_entry) > 1:
addToRC(rc_entry[idx].replace('%%', ac_prog))
real_ac_dir = ac_dir
real_ac_word = ac_word
found_prime = True
if len(alt_rc_entry) == 1:
alt_rc = alt_rc_entry[0]
if alt_rc == "":
# if no explicit alt_rc is given, construct one
m = pr.match(rc_entry[0])
if m:
alt_rc = m.group(1) + "_alternatives" + m.group(2)
addToRC(alt_rc.replace('%%', ac_prog))
elif len(alt_rc_entry) > 1:
alt_rc = alt_rc_entry[idx]
if alt_rc == "":
# if no explicit alt_rc is given, construct one
m = pr.match(rc_entry[idx])
if m:
alt_rc = m.group(1) + "_alternatives" + m.group(2)
addToRC(alt_rc.replace('%%', ac_prog))
found_alt = True
break
if found_alt:
break
if found_alt == False:
# if not successful
logger.info(msg + ' no')
if found_prime:
return [real_ac_dir, real_ac_word]
# write rc entries for 'not found'
if len(rc_entry) > 0: # the last one.
addToRC(rc_entry[-1].replace('%%', not_found))
return ['', not_found]
def addAlternatives(rcs, alt_type):
'''
Returns a \\prog_alternatives string to be used as an alternative
rc entry. alt_type can be a string or a list of strings.
'''
r = re.compile(r'\\Format (\S+).*$')
m = None
alt = ''
alt_token = '\\%s_alternatives '
if isinstance(alt_type, str):
alt_tokens = [alt_token % alt_type]
else:
alt_tokens = map(lambda s: alt_token % s, alt_type)
for idxx in range(len(rcs)):
if len(rcs) == 1:
m = r.match(rcs[0])
if m:
alt = '\n'.join([s + m.group(1) + ' "%%"' for s in alt_tokens])
elif len(rcs) > 1:
m = r.match(rcs[idxx])
if m:
if idxx > 0:
alt += '\n'
alt += '\n'.join([s + m.group(1) + ' "%%"' for s in alt_tokens])
return alt
def listAlternatives(progs, alt_type, rc_entry = []):
'''
Returns a list of \\prog_alternatives strings to be used as alternative
rc entries. alt_type can be a string or a list of strings.
'''
if len(rc_entry) > 1 and len(rc_entry) != len(progs) + 1:
logger.error("rc entry should have one item or item for each prog and not_found.")
sys.exit(2)
alt_rc_entry = []
for idx in range(len(progs)):
if len(rc_entry) == 1:
rcs = rc_entry[0].split('\n')
alt = addAlternatives(rcs, alt_type)
alt_rc_entry.insert(0, alt)
elif len(rc_entry) > 1:
rcs = rc_entry[idx].split('\n')
alt = addAlternatives(rcs, alt_type)
alt_rc_entry.insert(idx, alt)
return alt_rc_entry
def checkViewer(description, progs, rc_entry = [], path = []):
''' The same as checkProgAlternatives, but for viewers '''
alt_rc_entry = listAlternatives(progs, 'viewer', rc_entry)
return checkProgAlternatives(description, progs, rc_entry, alt_rc_entry, path, not_found = 'auto')
def checkEditor(description, progs, rc_entry = [], path = []):
''' The same as checkProgAlternatives, but for editors '''
alt_rc_entry = listAlternatives(progs, 'editor', rc_entry)
return checkProgAlternatives(description, progs, rc_entry, alt_rc_entry, path, not_found = 'auto')
def checkViewerNoRC(description, progs, rc_entry = [], path = []):
''' The same as checkViewer, but do not add rc entry '''
alt_rc_entry = listAlternatives(progs, 'viewer', rc_entry)
rc_entry = []
return checkProgAlternatives(description, progs, rc_entry, alt_rc_entry, path, not_found = 'auto')
def checkEditorNoRC(description, progs, rc_entry = [], path = []):
''' The same as checkViewer, but do not add rc entry '''
alt_rc_entry = listAlternatives(progs, 'editor', rc_entry)
rc_entry = []
return checkProgAlternatives(description, progs, rc_entry, alt_rc_entry, path, not_found = 'auto')
def checkViewerEditor(description, progs, rc_entry = [], path = []):
''' The same as checkProgAlternatives, but for viewers and editors '''
alt_rc_entry = listAlternatives(progs, ['editor', 'viewer'], rc_entry)
return checkProgAlternatives(description, progs, rc_entry, alt_rc_entry, path, not_found = 'auto')
def checkDTLtools():
''' Check whether DTL tools are available (Windows only) '''
# Find programs! Returned path is not used now
if ((os.name == 'nt' or sys.platform == 'cygwin') and
checkProg('DVI to DTL converter', ['dv2dt']) != ['', ''] and
checkProg('DTL to DVI converter', ['dt2dv']) != ['', '']):
dtl_tools = True
else:
dtl_tools = False
return dtl_tools
def checkLatex(dtl_tools):
''' Check latex, return lyx_check_config '''
path, LATEX = checkProg('a Latex2e program', ['latex $$i', 'latex2e $$i'])
path, PPLATEX = checkProg('a DVI postprocessing program', ['pplatex $$i'])
#-----------------------------------------------------------------
path, PLATEX = checkProg('pLaTeX, the Japanese LaTeX', ['platex $$i'])
if PLATEX != '':
# check if PLATEX is pLaTeX2e
writeToFile('chklatex.ltx', r'\nonstopmode\makeatletter\@@end')
# run platex on chklatex.ltx and check result
if cmdOutput(PLATEX + ' chklatex.ltx').find('pLaTeX2e') != -1:
# We have the Japanese pLaTeX2e
addToRC(r'\converter platex dvi "%s -kanji=$$E $$i" "latex=platex"' % PLATEX)
else:
PLATEX = ''
removeFiles(['chklatex.ltx', 'chklatex.log'])
#-----------------------------------------------------------------
# use LATEX to convert from latex to dvi if PPLATEX is not available
if PPLATEX == '':
PPLATEX = LATEX
if dtl_tools:
# Windows only: DraftDVI
addToRC(r'''\converter latex dvi2 "%s" "latex"
\converter dvi2 dvi "python -tt $$s/scripts/clean_dvi.py $$i $$o" ""''' % PPLATEX)
else:
addToRC(r'\converter latex dvi "%s" "latex"' % PPLATEX)
# no latex
if LATEX != '':
# Check if latex is usable
writeToFile('chklatex.ltx', r'''
\nonstopmode
\ifx\undefined\documentclass\else
\message{ThisIsLaTeX2e}
\fi
\makeatletter
\@@end
''')
# run latex on chklatex.ltx and check result
if cmdOutput(LATEX + ' chklatex.ltx').find('ThisIsLaTeX2e') != -1:
# valid latex2e
return LATEX
else:
logger.warning("Latex not usable (not LaTeX2e) ")
# remove temporary files
removeFiles(['chklatex.ltx', 'chklatex.log'])
return ''
def checkLuatex():
''' Check if luatex is there and usable '''
path, LUATEX = checkProg('LuaTeX', ['lualatex $$i'])
path, DVILUATEX = checkProg('LuaTeX (DVI)', ['dvilualatex $$i'])
if LUATEX != '':
# luatex binary is there
msg = "checking if LuaTeX is usable ..."
# Check if luatex is usable
writeToFile('luatest.tex', r'''
\nonstopmode
\documentclass{minimal}
\usepackage{fontspec}
\begin{document}
.
\end{document}
''')
# run lualatex on luatest.tex and check result
luatest = cmdOutput(LUATEX + ' luatest.tex')
if luatest.find('XeTeX is required to compile this document') != -1:
# fontspec/luatex too old! We do not support this version.
logger.info(msg + ' no (probably not recent enough)')
elif luatest.find('! LaTeX Error: File `fontspec.sty\' not found') != -1:
# fontspec missing
logger.info(msg + ' no (missing fontspec)')
else:
# working luatex
logger.info(msg + ' yes')
addToRC(r'\converter luatex pdf5 "%s" "latex=lualatex"' % LUATEX)
if DVILUATEX != '':
addToRC(r'\converter luatex dvi3 "%s" "latex=dvilualatex"' % DVILUATEX)
# remove temporary files
removeFiles(['luatest.tex', 'luatest.log', 'luatest.aux', 'luatest.pdf'])
def checkModule(module):
''' Check for a Python module, return the status '''
msg = 'checking for "' + module + ' module"... '
try:
__import__(module)
logger.info(msg + ' yes')
return True
except ImportError:
logger.info(msg + ' no')
return False
def checkFormatEntries(dtl_tools):
''' Check all formats (\Format entries) '''
checkViewerEditor('a Tgif viewer and editor', ['tgif'],
rc_entry = [r'\Format tgif obj Tgif "" "%%" "%%" "vector"'])
#
checkViewerEditor('a FIG viewer and editor', ['xfig', 'jfig3-itext.jar', 'jfig3.jar'],
rc_entry = [r'\Format fig fig FIG "" "%%" "%%" "vector"'])
#
checkViewerEditor('a Dia viewer and editor', ['dia'],
rc_entry = [r'\Format dia dia DIA "" "%%" "%%" "vector,zipped=native"'])
#
checkViewerEditor('an OpenOffice drawing viewer and editor', ['libreoffice', 'ooffice', 'oodraw', 'soffice'],
rc_entry = [r'\Format odg "odg, sxd" "OpenOffice drawing" "" "%%" "%%" "vector,zipped=native"'])
#
checkViewerEditor('a Grace viewer and editor', ['xmgrace'],
rc_entry = [r'\Format agr agr Grace "" "%%" "%%" "vector"'])
#
checkViewerEditor('a FEN viewer and editor', ['xboard -lpf $$i -mode EditPosition'],
rc_entry = [r'\Format fen fen FEN "" "%%" "%%" ""'])
#
checkViewerEditor('a SVG viewer and editor', ['inkscape'],
rc_entry = [r'\Format svg svg SVG "" "%%" "%%" "vector"'])
#
path, iv = checkViewerNoRC('a raster image viewer', ['xv', 'kview', 'gimp-remote', 'gimp'],
rc_entry = [r'''\Format bmp bmp BMP "" "%s" "%s" ""
\Format gif gif GIF "" "%s" "%s" ""
\Format jpg "jpg, jpeg" JPEG "" "%s" "%s" ""
\Format pbm pbm PBM "" "%s" "%s" ""
\Format pgm pgm PGM "" "%s" "%s" ""
\Format png png PNG "" "%s" "%s" ""
\Format ppm ppm PPM "" "%s" "%s" ""
\Format tiff tif TIFF "" "%s" "%s" ""
\Format xbm xbm XBM "" "%s" "%s" ""
\Format xpm xpm XPM "" "%s" "%s" ""'''])
path, ie = checkEditorNoRC('a raster image editor', ['gimp-remote', 'gimp'],
rc_entry = [r'''\Format bmp bmp BMP "" "%s" "%s" ""
\Format gif gif GIF "" "%s" "%s" ""
\Format jpg "jpg, jpeg" JPEG "" "%s" "%s" ""
\Format pbm pbm PBM "" "%s" "%s" ""
\Format pgm pgm PGM "" "%s" "%s" ""
\Format png png PNG "" "%s" "%s" ""
\Format ppm ppm PPM "" "%s" "%s" ""
\Format tiff tif TIFF "" "%s" "%s" ""
\Format xbm xbm XBM "" "%s" "%s" ""
\Format xpm xpm XPM "" "%s" "%s" ""'''])
addToRC(r'''\Format bmp bmp BMP "" "%s" "%s" ""
\Format gif gif GIF "" "%s" "%s" ""
\Format jpg "jpg, jpeg" JPEG "" "%s" "%s" ""
\Format pbm pbm PBM "" "%s" "%s" ""
\Format pgm pgm PGM "" "%s" "%s" ""
\Format png png PNG "" "%s" "%s" ""
\Format ppm ppm PPM "" "%s" "%s" ""
\Format tiff tif TIFF "" "%s" "%s" ""
\Format xbm xbm XBM "" "%s" "%s" ""
\Format xpm xpm XPM "" "%s" "%s" ""''' % \
(iv, ie, iv, ie, iv, ie, iv, ie, iv, ie, iv, ie, iv, ie, iv, ie, iv, ie, iv, ie) )
#
checkViewerEditor('a text editor', ['xemacs', 'gvim', 'kedit', 'kwrite', 'kate', \
'nedit', 'gedit', 'notepad'],
rc_entry = [r'''\Format asciichess asc "Plain text (chess output)" "" "" "%%" ""
\Format asciiimage asc "Plain text (image)" "" "" "%%" ""
\Format asciixfig asc "Plain text (Xfig output)" "" "" "%%" ""
\Format dateout tmp "date (output)" "" "" "%%" ""
\Format docbook sgml DocBook B "" "%%" "document,menu=export"
\Format docbook-xml xml "DocBook (XML)" "" "" "%%" "document,menu=export"
\Format dot dot "Graphviz Dot" "" "" "%%" "vector"
\Format platex tex "LaTeX (pLaTeX)" "" "" "%%" "document,menu=export"
\Format literate nw NoWeb N "" "%%" "document,menu=export"
\Format sweave Rnw "Sweave" S "" "%%" "document,menu=export"
\Format r R "R/S code" "" "" "%%" "document,menu=export"
\Format lilypond ly "LilyPond music" "" "" "%%" "vector"
\Format lilypond-book lytex "LilyPond book (LaTeX)" "" "" "%%" "document,menu=export"
\Format latex tex "LaTeX (plain)" L "" "%%" "document,menu=export"
\Format luatex tex "LaTeX (LuaTeX)" "" "" "%%" "document,menu=export"
\Format pdflatex tex "LaTeX (pdflatex)" "" "" "%%" "document,menu=export"
\Format xetex tex "LaTeX (XeTeX)" "" "" "%%" "document,menu=export"
\Format text txt "Plain text" a "" "%%" "document,menu=export"
\Format text2 txt "Plain text (pstotext)" "" "" "%%" "document"
\Format text3 txt "Plain text (ps2ascii)" "" "" "%%" "document"
\Format text4 txt "Plain text (catdvi)" "" "" "%%" "document"
\Format textparagraph txt "Plain Text, Join Lines" "" "" "%%" "document"''' ])
#Spreadsheets using ssconvert from gnumeric
checkViewer('gnumeric spreadsheet software', ['gnumeric'],
rc_entry = [r'''\Format gnumeric gnumeric "Gnumeric spreadsheet" "" "" "%%" "document"
\Format excel xls "Excel spreadsheet" "" "" "%%" "document"
\Format oocalc ods "OpenOffice spreadsheet" "" "" "%%" "document"'''])
#
checkViewer('an HTML previewer', ['firefox', 'mozilla file://$$p$$i', 'netscape'],
rc_entry = [r'\Format xhtml xhtml "LyXHTML" y "%%" "" "document,menu=export"'])
#
checkEditor('a BibTeX editor', ['jabref', 'JabRef', \
'pybliographic', 'bibdesk', 'gbib', 'kbib', \
'kbibtex', 'sixpack', 'bibedit', 'tkbibtex' \
'xemacs', 'gvim', 'kedit', 'kwrite', 'kate', \
'nedit', 'gedit', 'notepad'],
rc_entry = [r'''\Format bibtex bib "BibTeX" "" "" "%%" ""''' ])
#
#checkProg('a Postscript interpreter', ['gs'],
# rc_entry = [ r'\ps_command "%%"' ])
checkViewer('a Postscript previewer', ['kghostview', 'okular', 'evince', 'gv', 'ghostview -swap'],
rc_entry = [r'''\Format eps eps EPS "" "%%" "" "vector"
\Format ps ps Postscript t "%%" "" "document,vector,menu=export"'''])
# for xdg-open issues look here: http://www.mail-archive.com/lyx-devel@lists.lyx.org/msg151818.html
checkViewer('a PDF previewer', ['kpdf', 'okular', 'evince', 'kghostview', 'xpdf', 'acrobat', 'acroread', \
'gv', 'ghostview'],
rc_entry = [r'''\Format pdf pdf "PDF (ps2pdf)" P "%%" "" "document,vector,menu=export"
\Format pdf2 pdf "PDF (pdflatex)" F "%%" "" "document,vector,menu=export"
\Format pdf3 pdf "PDF (dvipdfm)" m "%%" "" "document,vector,menu=export"
\Format pdf4 pdf "PDF (XeTeX)" X "%%" "" "document,vector,menu=export"
\Format pdf5 pdf "PDF (LuaTeX)" u "%%" "" "document,vector,menu=export"'''])
#
checkViewer('a DVI previewer', ['xdvi', 'kdvi', 'okular', 'yap', 'dviout -Set=!m'],
rc_entry = [r'''\Format dvi dvi DVI D "%%" "" "document,vector,menu=export"
\Format dvi3 dvi "DVI (LuaTeX)" V "%%" "" "document,vector,menu=export"'''])
if dtl_tools:
# Windows only: DraftDVI
addToRC(r'\Format dvi2 dvi DraftDVI "" "" "" "vector"')
#
checkViewer('an HTML previewer', ['firefox', 'mozilla file://$$p$$i', 'netscape'],
rc_entry = [r'\Format html "html, htm" HTML H "%%" "" "document,menu=export"'])
#
checkViewerEditor('Noteedit', ['noteedit'],
rc_entry = [r'\Format noteedit not Noteedit "" "%%" "%%" "vector"'])
#
checkViewerEditor('an OpenDocument/OpenOffice viewer', ['swriter', 'oowriter', 'abiword'],
rc_entry = [r'''\Format odt odt OpenDocument "" "%%" "%%" "document,vector,menu=export"
\Format sxw sxw "OpenOffice.Org (sxw)" "" "" "" "document,vector"'''])
#
checkViewerEditor('a Rich Text and Word viewer', ['swriter', 'oowriter', 'abiword'],
rc_entry = [r'''\Format rtf rtf "Rich Text Format" "" "%%" "%%" "document,vector,menu=export"
\Format word doc "MS Word" W "%%" "%%" "document,vector,menu=export"'''])
#
# entries that do not need checkProg
addToRC(r'''\Format date "" "date command" "" "" "" ""
\Format csv csv "Table (CSV)" "" "" "" "document"
\Format fax "" Fax "" "" "" "document"
\Format lyx lyx LyX "" "" "" ""
\Format lyx13x 13.lyx "LyX 1.3.x" "" "" "" "document"
\Format lyx14x 14.lyx "LyX 1.4.x" "" "" "" "document"
\Format lyx15x 15.lyx "LyX 1.5.x" "" "" "" "document"
\Format lyx16x 16.lyx "LyX 1.6.x" "" "" "" "document,menu=export"
\Format lyx20x 20.lyx "LyX 2.0.x" "" "" "" "document,menu=export"
\Format clyx cjklyx "CJK LyX 1.4.x (big5)" "" "" "" "document"
\Format jlyx cjklyx "CJK LyX 1.4.x (euc-jp)" "" "" "" "document"
\Format klyx cjklyx "CJK LyX 1.4.x (euc-kr)" "" "" "" "document"
\Format lyxpreview lyxpreview "LyX Preview" "" "" "" ""
\Format pdftex pdftex_t PDFTEX "" "" "" ""
\Format program "" Program "" "" "" ""
\Format pstex pstex_t PSTEX "" "" "" ""
\Format wmf wmf "Windows Metafile" "" "" "" "vector"
\Format emf emf "Enhanced Metafile" "" "" "" "vector"
\Format wordhtml "html, htm" "HTML (MS Word)" "" "" "" "document"
''')
def checkConverterEntries():
''' Check all converters (\converter entries) '''
checkProg('the pdflatex program', ['pdflatex $$i'],
rc_entry = [ r'\converter pdflatex pdf2 "%%" "latex=pdflatex"' ])
checkProg('XeTeX', ['xelatex $$i'],
rc_entry = [ r'\converter xetex pdf4 "%%" "latex=xelatex"' ])
checkLuatex()
# Look for tex2lyx in this order (see bugs #3308 and #6986):
# 1) If we're running LyX in-place then tex2lyx will be found
# in ../src/tex2lyx with respect to the srcdir.
# 2) If LyX was configured with a version suffix then tex2lyx
# will also have this version suffix.
# 3) Otherwise always use tex2lyx.
in_place = os.path.join(srcdir, '..', 'src', 'tex2lyx', 'tex2lyx')
in_place = os.path.abspath(in_place)
path, t2l = checkProg('a LaTeX/Noweb -> LyX converter', [in_place, 'tex2lyx' + version_suffix, 'tex2lyx'],
rc_entry = [r'''\converter latex lyx "%% -f $$i $$o" ""
\converter literate lyx "%% -n -f $$i $$o" ""'''], not_found = 'tex2lyx')
if path == '':
logger.warning("Failed to find tex2lyx on your system.")
#
checkProg('a Noweb -> LaTeX converter', ['noweave -delay -index $$i > $$o'],
rc_entry = [r'''\converter literate latex "%%" ""
\converter literate pdflatex "%%" ""'''])
#
checkProg('a Sweave -> LaTeX converter', ['Rscript --verbose --no-save --no-restore $$s/scripts/lyxsweave.R $$p$$i $$p$$o $$e $$r'],
rc_entry = [r'''\converter sweave latex "%%" ""
\converter sweave pdflatex "%%" ""
\converter sweave xetex "%%" ""
\converter sweave luatex "%%" ""'''])
#
checkProg('a Sweave -> R/S code converter', ['Rscript --verbose --no-save --no-restore $$s/scripts/lyxstangle.R $$i $$e $$r'],
rc_entry = [ r'\converter sweave r "%%" ""' ])
#
checkProg('an HTML -> LaTeX converter', ['html2latex $$i', 'gnuhtml2latex $$i',
'htmltolatex -input $$i -output $$o', 'htmltolatex.jar -input $$i -output $$o'],
rc_entry = [ r'\converter html latex "%%" ""' ])
#
checkProg('an MS Word -> LaTeX converter', ['wvCleanLatex $$i $$o'],
rc_entry = [ r'\converter word latex "%%" ""' ])
# eLyXer: search as an executable (elyxer.py, elyxer)
path, elyxer = checkProg('a LyX -> HTML converter',
['elyxer.py --directory $$r $$i $$o', 'elyxer --directory $$r $$i $$o'],
rc_entry = [ r'\converter lyx html "%%" ""' ])
path, elyxer = checkProg('a LyX -> HTML (MS Word) converter',
['elyxer.py --html --directory $$r $$i $$o', 'elyxer --html --directory $$r $$i $$o'],
rc_entry = [ r'\converter lyx wordhtml "%%" ""' ])
if elyxer.find('elyxer') >= 0:
addToRC(r'''\copier html "python -tt $$s/scripts/ext_copy.py -e html,png,jpg,jpeg,css $$i $$o"''')
addToRC(r'''\copier wordhtml "python -tt $$s/scripts/ext_copy.py -e html,png,jpg,jpeg,css $$i $$o"''')
else:
# search for HTML converters other than eLyXer
# On SuSE the scripts have a .sh suffix, and on debian they are in /usr/share/tex4ht/
path, htmlconv = checkProg('a LaTeX -> HTML converter', ['htlatex $$i', 'htlatex.sh $$i', \
'/usr/share/tex4ht/htlatex $$i', 'tth -t -e2 -L$$b < $$i > $$o', \
'latex2html -no_subdir -split 0 -show_section_numbers $$i', 'hevea -s $$i'],
rc_entry = [ r'\converter latex html "%%" "needaux"' ])
if htmlconv.find('htlatex') >= 0 or htmlconv == 'latex2html':
addToRC(r'''\copier html "python -tt $$s/scripts/ext_copy.py -e html,png,css $$i $$o"''')
else:
addToRC(r'''\copier html "python -tt $$s/scripts/ext_copy.py $$i $$o"''')
path, htmlconv = checkProg('a LaTeX -> HTML (MS Word) converter', ["htlatex $$i 'html,word' 'symbol/!' '-cvalidate'", \
"htlatex.sh $$i 'html,word' 'symbol/!' '-cvalidate'", \
"/usr/share/tex4ht/htlatex $$i 'html,word' 'symbol/!' '-cvalidate'"],
rc_entry = [ r'\converter latex wordhtml "%%" "needaux"' ])
if htmlconv.find('htlatex') >= 0:
addToRC(r'''\copier wordhtml "python -tt $$s/scripts/ext_copy.py -e html,png,css $$i $$o"''')
else:
addToRC(r'''\copier wordhtml "python -tt $$s/scripts/ext_copy.py $$i $$o"''')
# Check if LyXBlogger is installed
lyxblogger_found = checkModule('lyxblogger')
if lyxblogger_found:
addToRC(r'\Format blog blog "LyXBlogger" "" "" "" "document"')
addToRC(r'\converter xhtml blog "python -m lyxblogger $$i" ""')
#
checkProg('an OpenOffice.org -> LaTeX converter', ['w2l -clean $$i'],
rc_entry = [ r'\converter sxw latex "%%" ""' ])
#
checkProg('an OpenDocument -> LaTeX converter', ['w2l -clean $$i'],
rc_entry = [ r'\converter odt latex "%%" ""' ])
# According to http://www.tug.org/applications/tex4ht/mn-commands.html
# the command mk4ht oolatex $$i has to be used as default,
# but as this would require to have Perl installed, in MiKTeX oolatex is
# directly available as application.
# On SuSE the scripts have a .sh suffix, and on debian they are in /usr/share/tex4ht/
# Both SuSE and debian have oolatex
checkProg('a LaTeX -> Open Document converter', [
'oolatex $$i', 'mk4ht oolatex $$i', 'oolatex.sh $$i', '/usr/share/tex4ht/oolatex $$i',
'htlatex $$i \'xhtml,ooffice\' \'ooffice/! -cmozhtf\' \'-coo\' \'-cvalidate\''],
rc_entry = [ r'\converter latex odt "%%" "needaux"' ])
# On windows it is called latex2rt.exe
checkProg('a LaTeX -> RTF converter', ['latex2rtf -p -S -o $$o $$i', 'latex2rt -p -S -o $$o $$i'],
rc_entry = [ r'\converter latex rtf "%%" "needaux"' ])
#
checkProg('a RTF -> HTML converter', ['unrtf --html $$i > $$o'],
rc_entry = [ r'\converter rtf html "%%" ""' ])
#
checkProg('a PS to PDF converter', ['ps2pdf13 $$i $$o'],
rc_entry = [ r'\converter ps pdf "%%" ""' ])
#
checkProg('a PS to TXT converter', ['pstotext $$i > $$o'],
rc_entry = [ r'\converter ps text2 "%%" ""' ])
#
checkProg('a PS to TXT converter', ['ps2ascii $$i $$o'],
rc_entry = [ r'\converter ps text3 "%%" ""' ])
#
checkProg('a PS to EPS converter', ['ps2eps $$i'],
rc_entry = [ r'''\Format eps2 eps "EPS (ps2eps)" "" "" "" "vector"
\converter ps eps "%%" ""
\converter eps2 eps "%%" "resultfile=$$b.eps.eps"''' ])
#
checkProg('a PDF to PS converter', ['pdf2ps $$i $$o', 'pdftops $$i $$o'],
rc_entry = [ r'\converter pdf ps "%%" ""' ])
#
checkProg('a PDF to EPS converter', ['pdftops -eps -f 1 -l 1 $$i $$o'],
rc_entry = [ r'\converter pdf eps "%%" ""' ])
#
checkProg('a DVI to TXT converter', ['catdvi $$i > $$o'],
rc_entry = [ r'\converter dvi text4 "%%" ""' ])
#
checkProg('a DVI to PS converter', ['dvips -o $$o $$i'],
rc_entry = [ r'\converter dvi ps "%%" ""' ])
#
checkProg('a DVI to PDF converter', ['dvipdfmx -o $$o $$i', 'dvipdfm -o $$o $$i'],
rc_entry = [ r'\converter dvi pdf3 "%%" ""' ])
#
checkProg('a fax program', ['kdeprintfax $$i', 'ksendfax $$i', 'hylapex $$i'],
rc_entry = [ r'\converter ps fax "%%" ""'])
#
path, fig2dev = checkProg('a FIG -> Image converter', ['fig2dev'])
if fig2dev == "fig2dev":
addToRC(r'''\converter fig eps "fig2dev -L eps $$i $$o" ""
\converter fig ppm "fig2dev -L ppm $$i $$o" ""
\converter fig png "fig2dev -L png $$i $$o" ""
\converter fig pdftex "python -tt $$s/scripts/fig2pdftex.py $$i $$o" ""
\converter fig pstex "python -tt $$s/scripts/fig2pstex.py $$i $$o" ""''')
#
checkProg('a TIFF -> PS converter', ['tiff2ps $$i > $$o'],
rc_entry = [ r'\converter tiff eps "%%" ""', ''])
#
checkProg('a TGIF -> EPS/PPM converter', ['tgif'],
rc_entry = [
r'''\converter tgif eps "tgif -print -color -eps -stdout $$i > $$o" ""
\converter tgif png "tgif -print -color -png -o $$d $$i" ""
\converter tgif pdf "tgif -print -color -pdf -stdout $$i > $$o" ""''',
''])
#
checkProg('a WMF -> EPS converter', ['metafile2eps $$i $$o', 'wmf2eps -o $$o $$i'],
rc_entry = [ r'\converter wmf eps "%%" ""'])
#
checkProg('an EMF -> EPS converter', ['metafile2eps $$i $$o', 'wmf2eps -o $$o $$i'],
rc_entry = [ r'\converter emf eps "%%" ""'])
#
checkProg('an EPS -> PDF converter', ['epstopdf'],
rc_entry = [ r'\converter eps pdf "epstopdf --outfile=$$o $$i" ""', ''])
#
checkProg('an EPS -> PNG converter', ['convert $$i $$o'],
rc_entry = [ r'\converter eps png "%%" ""', ''])
#
# no agr -> pdf converter, since the pdf library used by gracebat is not
# free software and therefore not compiled in in many installations.
# Fortunately, this is not a big problem, because we will use epstopdf to
# convert from agr to pdf via eps without loss of quality.
checkProg('a Grace -> Image converter', ['gracebat'],
rc_entry = [
r'''\converter agr eps "gracebat -hardcopy -printfile $$o -hdevice EPS $$i 2>/dev/null" ""
\converter agr png "gracebat -hardcopy -printfile $$o -hdevice PNG $$i 2>/dev/null" ""
\converter agr jpg "gracebat -hardcopy -printfile $$o -hdevice JPEG $$i 2>/dev/null" ""
\converter agr ppm "gracebat -hardcopy -printfile $$o -hdevice PNM $$i 2>/dev/null" ""''',
''])
#
checkProg('a Dot -> Image converter', ['dot'],
rc_entry = [
r'''\converter dot eps "dot -Teps $$i -o $$o" ""
\converter dot png "dot -Tpng $$i -o $$o" ""''',
''])
#
checkProg('a Dia -> PNG converter', ['dia -e $$o -t png $$i'],
rc_entry = [ r'\converter dia png "%%" ""'])
#
checkProg('a Dia -> EPS converter', ['dia -e $$o -t eps $$i'],
rc_entry = [ r'\converter dia eps "%%" ""'])
#
checkProg('an OpenOffice -> EPS converter', ['libreoffice -headless -nologo -convert-to eps $$i'],
rc_entry = [ r'\converter odg eps2 "%%" ""'])
#
checkProg('a SVG -> PDF converter', ['rsvg-convert -f pdf -o $$o $$i', 'inkscape --file=$$i --export-area-drawing --without-gui --export-pdf=$$o'],
rc_entry = [ r'\converter svg pdf "%%" ""'])
#
checkProg('a SVG -> EPS converter', ['rsvg-convert -f ps -o $$o $$i', 'inkscape --file=$$i --export-area-drawing --without-gui --export-eps=$$o'],
rc_entry = [ r'\converter svg eps "%%" ""'])
#
checkProg('a SVG -> PNG converter', ['rsvg-convert -f png -o $$o $$i', 'inkscape --without-gui --file=$$i --export-png=$$o'],
rc_entry = [ r'\converter svg png "%%" ""'])
#
# gnumeric/xls/ods to tex
checkProg('a spreadsheet -> latex converter', ['ssconvert'],
rc_entry = [ r'''\converter gnumeric latex "ssconvert --export-type=Gnumeric_html:latex $$i $$o" ""
\converter oocalc latex "ssconvert --export-type=Gnumeric_html:latex $$i $$o" ""
\converter excel latex "ssconvert --export-type=Gnumeric_html:latex $$i $$o" ""''',
''])
path, lilypond = checkProg('a LilyPond -> EPS/PDF/PNG converter', ['lilypond'])
if (lilypond != ''):
version_string = cmdOutput("lilypond --version")
match = re.match('GNU LilyPond (\S+)', version_string)
if match:
version_number = match.groups()[0]
version = version_number.split('.')
if int(version[0]) > 2 or (len(version) > 1 and int(version[0]) == 2 and int(version[1]) >= 11):
addToRC(r'''\converter lilypond eps "lilypond -dbackend=eps -dsafe --ps $$i" ""
\converter lilypond png "lilypond -dbackend=eps -dsafe --png $$i" ""''')
addToRC(r'\converter lilypond pdf "lilypond -dbackend=eps -dsafe --pdf $$i" ""')
logger.info('+ found LilyPond version %s.' % version_number)
elif int(version[0]) > 2 or (len(version) > 1 and int(version[0]) == 2 and int(version[1]) >= 6):
addToRC(r'''\converter lilypond eps "lilypond -b eps --ps --safe $$i" ""
\converter lilypond png "lilypond -b eps --png $$i" ""''')
if int(version[0]) > 2 or (len(version) > 1 and int(version[0]) == 2 and int(version[1]) >= 9):
addToRC(r'\converter lilypond pdf "lilypond -b eps --pdf --safe $$i" ""')
logger.info('+ found LilyPond version %s.' % version_number)
else:
logger.info('+ found LilyPond, but version %s is too old.' % version_number)
else:
logger.info('+ found LilyPond, but could not extract version number.')
#
path, lilypond_book = checkProg('a LilyPond book (LaTeX) -> LaTeX converter', ['lilypond-book'])
if (lilypond_book != ''):
version_string = cmdOutput("lilypond-book --version")
match = re.match('^(\S+)$', version_string)
if match:
version_number = match.groups()[0]
version = version_number.split('.')
if int(version[0]) > 2 or (len(version) > 1 and int(version[0]) == 2 and int(version[1]) >= 13):
# Note: The --lily-output-dir flag is required because lilypond-book
# does not process input again unless the input has changed,
# even if the output format being requested is different. So
# once a .eps file exists, lilypond-book won't create a .pdf
# even when requested with --pdf. This is a problem if a user
# clicks View PDF after having done a View DVI. To circumvent
# this, use different output folders for eps and pdf outputs.
addToRC(r'\converter lilypond-book latex "lilypond-book --safe --lily-output-dir=ly-eps $$i" ""')
addToRC(r'\converter lilypond-book pdflatex "lilypond-book --safe --pdf --latex-program=pdflatex --lily-output-dir=ly-pdf $$i" ""')
addToRC(r'\converter lilypond-book xetex "lilypond-book --safe --pdf --latex-program=xelatex --lily-output-dir=ly-pdf $$i" ""')
addToRC(r'\converter lilypond-book luatex "lilypond-book --safe --pdf --latex-program=lualatex --lily-output-dir=ly-pdf $$i" ""')
logger.info('+ found LilyPond-book version %s.' % version_number)
else:
logger.info('+ found LilyPond-book, but version %s is too old.' % version_number)
else:
logger.info('+ found LilyPond-book, but could not extract version number.')
#
checkProg('a Noteedit -> LilyPond converter', ['noteedit --export-lilypond $$i'],
rc_entry = [ r'\converter noteedit lilypond "%%" ""', ''])
#
# Currently, lyxpak outputs a gzip compressed tar archive on *nix
# and a zip archive on Windows.
# So, we configure the appropriate version according to the platform.
cmd = r'\converter lyx %s "python -tt $$s/scripts/lyxpak.py $$r/$$i" ""'
if os.name == 'nt':
addToRC(r'\Format lyxzip zip "LyX Archive (zip)" "" "" "" "document,menu=export"')
addToRC(cmd % "lyxzip")
else:
addToRC(r'\Format lyxgz gz "LyX Archive (tar.gz)" "" "" "" "document,menu=export"')
addToRC(cmd % "lyxgz")
#
# FIXME: no rc_entry? comment it out
# checkProg('Image converter', ['convert $$i $$o'])
#
# Entries that do not need checkProg
addToRC(r'''
\converter csv lyx "python -tt $$s/scripts/csv2lyx.py $$i $$o" ""
\converter date dateout "python -tt $$s/scripts/date.py %d-%m-%Y > $$o" ""
\converter docbook docbook-xml "cp $$i $$o" "xml"
\converter fen asciichess "python -tt $$s/scripts/fen2ascii.py $$i $$o" ""
\converter lyx lyx13x "python -tt $$s/lyx2lyx/lyx2lyx -t 221 $$i > $$o" ""
\converter lyx lyx14x "python -tt $$s/lyx2lyx/lyx2lyx -t 245 $$i > $$o" ""
\converter lyx lyx15x "python -tt $$s/lyx2lyx/lyx2lyx -t 276 $$i > $$o" ""
\converter lyx lyx16x "python -tt $$s/lyx2lyx/lyx2lyx -t 345 $$i > $$o" ""
\converter lyx lyx20x "python -tt $$s/lyx2lyx/lyx2lyx -t 413 $$i > $$o" ""
\converter lyx clyx "python -tt $$s/lyx2lyx/lyx2lyx -c big5 -t 245 $$i > $$o" ""
\converter lyx jlyx "python -tt $$s/lyx2lyx/lyx2lyx -c euc_jp -t 245 $$i > $$o" ""
\converter lyx klyx "python -tt $$s/lyx2lyx/lyx2lyx -c euc_kr -t 245 $$i > $$o" ""
\converter clyx lyx "python -tt $$s/lyx2lyx/lyx2lyx -c big5 $$i > $$o" ""
\converter jlyx lyx "python -tt $$s/lyx2lyx/lyx2lyx -c euc_jp $$i > $$o" ""
\converter klyx lyx "python -tt $$s/lyx2lyx/lyx2lyx -c euc_kr $$i > $$o" ""
\converter lyxpreview png "python -tt $$s/scripts/lyxpreview2bitmap.py --png" ""
\converter lyxpreview ppm "python -tt $$s/scripts/lyxpreview2bitmap.py --ppm" ""
''')
def checkDocBook():
''' Check docbook '''
path, DOCBOOK = checkProg('SGML-tools 2.x (DocBook), db2x scripts or xsltproc', ['sgmltools', 'db2dvi', 'xsltproc'],
rc_entry = [
r'''\converter docbook dvi "sgmltools -b dvi $$i" ""
\converter docbook html "sgmltools -b html $$i" ""''',
r'''\converter docbook dvi "db2dvi $$i" ""
\converter docbook html "db2html $$i" ""''',
r'''\converter docbook dvi "" ""
\converter docbook html "" ""''',
r'''\converter docbook dvi "" ""
\converter docbook html "" ""'''])
#
if DOCBOOK != '':
return ('yes', 'true', '\\def\\hasdocbook{yes}')
else:
return ('no', 'false', '')
def checkOtherEntries():
''' entries other than Format and Converter '''
checkProg('ChkTeX', ['chktex -n1 -n3 -n6 -n9 -n22 -n25 -n30 -n38'],
rc_entry = [ r'\chktex_command "%%"' ])
checkProgAlternatives('BibTeX or alternative programs', ['bibtex', 'bibtex8', 'biber'],
rc_entry = [ r'\bibtex_command "%%"' ],
alt_rc_entry = [ r'\bibtex_alternatives "%%"' ])
checkProg('a specific Japanese BibTeX variant', ['pbibtex', 'jbibtex', 'bibtex'],
rc_entry = [ r'\jbibtex_command "%%"' ])
checkProgAlternatives('available index processors', ['texindy', 'makeindex -c -q', 'xindy'],
rc_entry = [ r'\index_command "%%"' ],
alt_rc_entry = [ r'\index_alternatives "%%"' ])
checkProg('an index processor appropriate to Japanese', ['mendex -c -q', 'jmakeindex -c -q', 'makeindex -c -q'],
rc_entry = [ r'\jindex_command "%%"' ])
checkProg('the splitindex processor', ['splitindex.pl', 'splitindex',
'splitindex.class'], rc_entry = [ r'\splitindex_command "%%"' ])
checkProg('a nomenclature processor', ['makeindex'],
rc_entry = [ r'\nomencl_command "makeindex -s nomencl.ist"' ])
## FIXME: OCTAVE is not used anywhere
# path, OCTAVE = checkProg('Octave', ['octave'])
## FIXME: MAPLE is not used anywhere
# path, MAPLE = checkProg('Maple', ['maple'])
checkProg('a spool command', ['lp', 'lpr'],
rc_entry = [
r'''\print_spool_printerprefix "-d "
\print_spool_command "lp"''',
r'''\print_spool_printerprefix "-P",
\print_spool_command "lpr"''',
''])
# Add the rest of the entries (no checkProg is required)
addToRC(r'''\copier fig "python -tt $$s/scripts/fig_copy.py $$i $$o"
\copier pstex "python -tt $$s/scripts/tex_copy.py $$i $$o $$l"
\copier pdftex "python -tt $$s/scripts/tex_copy.py $$i $$o $$l"
\copier program "python -tt $$s/scripts/ext_copy.py $$i $$o"
''')
def processLayoutFile(file, bool_docbook):
''' process layout file and get a line of result
Declare lines look like this: (article.layout, scrbook.layout, svjog.layout)
\DeclareLaTeXClass{article}
\DeclareLaTeXClass[scrbook]{book (koma-script)}
\DeclareLaTeXClass[svjour,svjog.clo]{article (Springer - svjour/jog)}
we expect output:
"article" "article" "article" "false" "article.cls"
"scrbook" "scrbook" "book (koma-script)" "false" "scrbook.cls"
"svjog" "svjour" "article (Springer - svjour/jog)" "false" "svjour.cls,svjog.clo"
'''
def checkForClassExtension(x):
'''if the extension for a latex class is not
provided, add .cls to the classname'''
if not '.' in x:
return x.strip() + '.cls'
else:
return x.strip()
classname = file.split(os.sep)[-1].split('.')[0]
# return ('LaTeX', '[a,b]', 'a', ',b,c', 'article') for \DeclareLaTeXClass[a,b,c]{article}
p = re.compile(r'\Declare(LaTeX|DocBook)Class\s*(\[([^,]*)(,.*)*\])*\s*{(.*)}')
for line in open(file).readlines():
res = p.search(line)
if res != None:
(classtype, optAll, opt, opt1, desc) = res.groups()
avai = {'LaTeX':'false', 'DocBook':bool_docbook}[classtype]
if opt == None:
opt = classname
prereq_latex = checkForClassExtension(classname)
else:
prereq_list = optAll[1:-1].split(',')
prereq_list = map(checkForClassExtension, prereq_list)
prereq_latex = ','.join(prereq_list)
prereq_docbook = {'true':'', 'false':'docbook'}[bool_docbook]
prereq = {'LaTeX':prereq_latex, 'DocBook':prereq_docbook}[classtype]
return '"%s" "%s" "%s" "%s" "%s"\n' % (classname, opt, desc, avai, prereq)
logger.warning("Layout file " + file + " has no \DeclareXXClass line. ")
return ""
def checkLatexConfig(check_config, bool_docbook):
''' Explore the LaTeX configuration
Return None (will be passed to sys.exit()) for success.
'''
msg = 'checking LaTeX configuration... '
# if --without-latex-config is forced, or if there is no previous
# version of textclass.lst, re-generate a default file.
if not os.path.isfile('textclass.lst') or not check_config:
# remove the files only if we want to regenerate
removeFiles(['textclass.lst', 'packages.lst'])
#
# Then, generate a default textclass.lst. In case configure.py
# fails, we still have something to start lyx.
logger.info(msg + ' default values')
logger.info('+checking list of textclasses... ')
tx = open('textclass.lst', 'w')
tx.write('''
# This file declares layouts and their associated definition files
# (include dir. relative to the place where this file is).
# It contains only default values, since chkconfig.ltx could not be run
# for some reason. Run ./configure.py if you need to update it after a
# configuration change.
''')
# build the list of available layout files and convert it to commands
# for chkconfig.ltx
foundClasses = []
for file in glob.glob( os.path.join('layouts', '*.layout') ) + \
glob.glob( os.path.join(srcdir, 'layouts', '*.layout' ) ) :
# valid file?
if not os.path.isfile(file):
continue
# get stuff between /xxxx.layout .
classname = file.split(os.sep)[-1].split('.')[0]
# tr ' -' '__'`
cleanclass = classname.replace(' ', '_')
cleanclass = cleanclass.replace('-', '_')
# make sure the same class is not considered twice
if foundClasses.count(cleanclass) == 0: # not found before
foundClasses.append(cleanclass)
retval = processLayoutFile(file, bool_docbook)
if retval != "":
tx.write(retval)
tx.close()
logger.info('\tdone')
if not check_config:
return None
# the following will generate textclass.lst.tmp, and packages.lst.tmp
logger.info(msg + '\tauto')
removeFiles(['chkconfig.classes', 'chkconfig.vars', 'chklayouts.tex',
'wrap_chkconfig.ltx'])
rmcopy = False
if not os.path.isfile( 'chkconfig.ltx' ):
shutil.copyfile( os.path.join(srcdir, 'chkconfig.ltx'), 'chkconfig.ltx' )
rmcopy = True
writeToFile('wrap_chkconfig.ltx', '%s\n\\input{chkconfig.ltx}\n' % docbook_cmd)
# Construct the list of classes to test for.
# build the list of available layout files and convert it to commands
# for chkconfig.ltx
declare = re.compile(r'\Declare(LaTeX|DocBook)Class\s*(\[([^,]*)(,.*)*\])*\s*{(.*)}')
empty = re.compile(r'^\s*$')
testclasses = list()
for file in glob.glob( os.path.join('layouts', '*.layout') ) + \
glob.glob( os.path.join(srcdir, 'layouts', '*.layout' ) ) :
nodeclaration = False
if not os.path.isfile(file):
continue
classname = file.split(os.sep)[-1].split('.')[0]
for line in open(file).readlines():
if not empty.match(line) and line[0] != '#':
logger.warning("Failed to find valid \Declare line for layout file `" + file + "'.\n\t=> Skipping this file!")
nodeclaration = True
break
if declare.search(line) == None:
continue
testclasses.append("\\TestDocClass{%s}{%s}" % (classname, line[1:].strip()))
break
if nodeclaration:
continue
testclasses.sort()
cl = open('chklayouts.tex', 'w')
for line in testclasses:
cl.write(line + '\n')
cl.close()
#
# we have chklayouts.tex, then process it
ret = 1
latex_out = cmdOutput(LATEX + ' wrap_chkconfig.ltx')
for line in latex_out.splitlines():
if re.match('^\+', line):
logger.info(line.strip())
# return None if the command succeeds
if line == "+Inspection done.":
ret = None
#
# remove the copied file
if rmcopy:
removeFiles( [ 'chkconfig.ltx' ] )
#
# currently, values in chhkconfig are only used to set
# \font_encoding
values = {}
for line in open('chkconfig.vars').readlines():
key, val = re.sub('-', '_', line).split('=')
val = val.strip()
values[key] = val.strip("'")
# chk_fontenc may not exist
try:
addToRC(r'\font_encoding "%s"' % values["chk_fontenc"])
except:
pass
# if configure successed, move textclass.lst.tmp to textclass.lst
# and packages.lst.tmp to packages.lst
if os.path.isfile('textclass.lst.tmp') and len(open('textclass.lst.tmp').read()) > 0 \
and os.path.isfile('packages.lst.tmp') and len(open('packages.lst.tmp').read()) > 0:
shutil.move('textclass.lst.tmp', 'textclass.lst')
shutil.move('packages.lst.tmp', 'packages.lst')
return ret
def checkModulesConfig():
removeFiles(['lyxmodules.lst', 'chkmodules.tex'])
logger.info('+checking list of modules... ')
tx = open('lyxmodules.lst', 'w')
tx.write('''## This file declares modules and their associated definition files.
## It has been automatically generated by configure
## Use "Options/Reconfigure" if you need to update it after a
## configuration change.
## "ModuleName" "filename" "Description" "Packages" "Requires" "Excludes" "Category"
''')
# build the list of available modules
seen = []
# note that this searches the local directory first, then the
# system directory. that way, we pick up the user's version first.
for file in glob.glob( os.path.join('layouts', '*.module') ) + \
glob.glob( os.path.join(srcdir, 'layouts', '*.module' ) ) :
# valid file?
logger.info(file)
if not os.path.isfile(file):
continue
filename = file.split(os.sep)[-1]
filename = filename[:-7]
if seen.count(filename):
continue
seen.append(filename)
retval = processModuleFile(file, filename, bool_docbook)
if retval != "":
tx.write(retval)
tx.close()
logger.info('\tdone')
def processModuleFile(file, filename, bool_docbook):
''' process module file and get a line of result
The top of a module file should look like this:
#\DeclareLyXModule[LaTeX Packages]{ModuleName}
#DescriptionBegin
#...body of description...
#DescriptionEnd
#Requires: [list of required modules]
#Excludes: [list of excluded modules]
#Category: [category name]
The last three lines are optional (though do give a category).
We expect output:
"ModuleName" "filename" "Description" "Packages" "Requires" "Excludes" "Category"
'''
remods = re.compile(r'\DeclareLyXModule\s*(?:\[([^]]*?)\])?{(.*)}')
rereqs = re.compile(r'#+\s*Requires: (.*)')
reexcs = re.compile(r'#+\s*Excludes: (.*)')
recaty = re.compile(r'#+\s*Category: (.*)')
redbeg = re.compile(r'#+\s*DescriptionBegin\s*$')
redend = re.compile(r'#+\s*DescriptionEnd\s*$')
modname = desc = pkgs = req = excl = catgy = ""
readingDescription = False
descLines = []
for line in open(file).readlines():
if readingDescription:
res = redend.search(line)
if res != None:
readingDescription = False
desc = " ".join(descLines)
# Escape quotes.
desc = desc.replace('"', '\\"')
continue
descLines.append(line[1:].strip())
continue
res = redbeg.search(line)
if res != None:
readingDescription = True
continue
res = remods.search(line)
if res != None:
(pkgs, modname) = res.groups()
if pkgs == None:
pkgs = ""
else:
tmp = [s.strip() for s in pkgs.split(",")]
pkgs = ",".join(tmp)
continue
res = rereqs.search(line)
if res != None:
req = res.group(1)
tmp = [s.strip() for s in req.split("|")]
req = "|".join(tmp)
continue
res = reexcs.search(line)
if res != None:
excl = res.group(1)
tmp = [s.strip() for s in excl.split("|")]
excl = "|".join(tmp)
continue
res = recaty.search(line)
if res != None:
catgy = res.group(1)
continue
if modname == "":
logger.warning("Module file without \DeclareLyXModule line. ")
return ""
if pkgs != "":
# this module has some latex dependencies:
# append the dependencies to chkmodules.tex,
# which is \input'ed by chkconfig.ltx
testpackages = list()
for pkg in pkgs.split(","):
if "->" in pkg:
# this is a converter dependency: skip
continue
if pkg.endswith(".sty"):
pkg = pkg[:-4]
testpackages.append("\\TestPackage{%s}" % (pkg,))
cm = open('chkmodules.tex', 'a')
for line in testpackages:
cm.write(line + '\n')
cm.close()
return '"%s" "%s" "%s" "%s" "%s" "%s" "%s"\n' % (modname, filename, desc, pkgs, req, excl, catgy)
def checkTeXAllowSpaces():
''' Let's check whether spaces are allowed in TeX file names '''
tex_allows_spaces = 'false'
if lyx_check_config:
msg = "Checking whether TeX allows spaces in file names... "
writeToFile('a b.tex', r'\message{working^^J}' )
if LATEX != '':
if os.name == 'nt' or sys.platform == 'cygwin':
latex_out = cmdOutput(LATEX + r""" "\nonstopmode\input{\"a b\"}\makeatletter\@@end" """)
else:
latex_out = cmdOutput(LATEX + r""" '\nonstopmode\input{"a b"}\makeatletter\@@end' """)
else:
latex_out = ''
if 'working' in latex_out:
logger.info(msg + 'yes')
tex_allows_spaces = 'true'
else:
logger.info(msg + 'no')
tex_allows_spaces = 'false'
addToRC(r'\tex_allows_spaces ' + tex_allows_spaces)
removeFiles( [ 'a b.tex', 'a b.log', 'texput.log' ])
def removeTempFiles():
# Final clean-up
if not lyx_keep_temps:
removeFiles(['chkconfig.vars', 'chklatex.ltx', 'chklatex.log',
'chklayouts.tex', 'chkmodules.tex', 'missfont.log',
'wrap_chkconfig.ltx', 'wrap_chkconfig.log'])
if __name__ == '__main__':
lyx_check_config = True
outfile = 'lyxrc.defaults'
lyxrc_fileformat = 3
rc_entries = ''
lyx_keep_temps = False
version_suffix = ''
## Parse the command line
for op in sys.argv[1:]: # default shell/for list is $*, the options
if op in [ '-help', '--help', '-h' ]:
print '''Usage: configure [options]
Options:
--help show this help lines
--keep-temps keep temporary files (for debug. purposes)
--without-latex-config do not run LaTeX to determine configuration
--with-version-suffix=suffix suffix of binary installed files
'''
sys.exit(0)
elif op == '--without-latex-config':
lyx_check_config = False
elif op == '--keep-temps':
lyx_keep_temps = True
elif op[0:22] == '--with-version-suffix=': # never mind if op is not long enough
version_suffix = op[22:]
else:
print "Unknown option", op
sys.exit(1)
#
# check if we run from the right directory
srcdir = os.path.dirname(sys.argv[0])
if srcdir == '':
srcdir = '.'
if not os.path.isfile( os.path.join(srcdir, 'chkconfig.ltx') ):
logger.error("configure: error: cannot find chkconfig.ltx script")
sys.exit(1)
setEnviron()
createDirectories()
dtl_tools = checkDTLtools()
## Write the first part of outfile
writeToFile(outfile, '''# This file has been automatically generated by LyX' lib/configure.py
# script. It contains default settings that have been determined by
# examining your system. PLEASE DO NOT MODIFY ANYTHING HERE! If you
# want to customize LyX, use LyX' Preferences dialog or modify directly
# the "preferences" file instead. Any setting in that file will
# override the values given here.
Format %i
''' % lyxrc_fileformat)
# check latex
LATEX = checkLatex(dtl_tools)
# check java and perl before any checkProg that may require them
java = checkProg('a java interpreter', ['java'])[1]
perl = checkProg('a perl interpreter', ['perl'])[1]
checkFormatEntries(dtl_tools)
checkConverterEntries()
(chk_docbook, bool_docbook, docbook_cmd) = checkDocBook()
checkTeXAllowSpaces()
windows_style_tex_paths = checkTeXPaths()
if windows_style_tex_paths != '':
addToRC(r'\tex_expects_windows_paths %s' % windows_style_tex_paths)
checkOtherEntries()
checkModulesConfig()
# --without-latex-config can disable lyx_check_config
ret = checkLatexConfig(lyx_check_config and LATEX != '', bool_docbook)
removeTempFiles()
# The return error code can be 256. Because most systems expect an error code
# in the range 0-127, 256 can be interpretted as 'success'. Because we expect
# a None for success, 'ret is not None' is used to exit.
sys.exit(ret is not None)