lyx_mirror/lib/scripts/lyxpreview2bitmap.py

571 lines
22 KiB
Python
Raw Normal View History

# -*- coding: utf-8 -*-
# file lyxpreview2bitmap.py
# This file is part of LyX, the document processor.
# Licence details can be found in the file COPYING.
# author Angus Leeming
# with much advice from members of the preview-latex project:
# David Kastrup, dak@gnu.org and
# Jan-Åke Larsson, jalar@mai.liu.se.
# Full author contact details are available in file CREDITS
# This script takes a LaTeX file and generates a collection of
# png or ppm image files, one per previewed snippet.
# Pre-requisites:
# * python 2.4 or later (subprocess module);
# * A latex executable;
# * preview.sty;
# * dvipng;
# * dv2dt;
# * pngtoppm (if outputing ppm format images).
# preview.sty and dvipng are part of the preview-latex project
# http://preview-latex.sourceforge.net/
# preview.sty can alternatively be obtained from
# CTAN/support/preview-latex/
# Example usage:
# lyxpreview2bitmap.py --bg=faf0e6 0lyxpreview.tex
# This script takes one obligatory argument:
#
# <input file>: The name of the .tex file to be converted.
#
# and these optional arguments:
#
# --png, --ppm: The desired output format. Either 'png' or 'ppm'.
# --dpi=<res>: A scale factor, used to ascertain the resolution of the
# generated image which is then passed to gs.
# --fg=<color>: The foreground color as a hexadecimal string, eg '000000'.
# --bg=<color>: The background color as a hexadecimal string, eg 'faf0e6'.
# --latex=<exe>: The converter for latex files. Default is latex.
# --bibtex=<exe>: The converter for bibtex files. Default is bibtex.
# --lilypond: Preprocess through lilypond-book. Default is false.
# --lilypond-book=<exe>:
# The converter for lytex files. Default is lilypond-book.
#
# -d, --debug Show the output from external commands.
# -h, --help Show an help screen and exit.
# -v, --verbose Show progress messages.
# Decomposing TEXFILE's name as DIR/BASE.tex, this script will,
# if executed successfully, leave in DIR:
# * a (possibly large) number of image files with names
# like BASE[0-9]+.png
# * a file BASE.metrics, containing info needed by LyX to position
# the images correctly on the screen.
# What does this script do?
# 1) Call latex/pdflatex/xelatex/whatever (CONVERTER parameter)
# 2) If the output is a PDF fallback to legacy
# 3) Otherwise check each page of the DVI (with dv2dt) looking for
# PostScript literals, not well supported by dvipng. Pages
# containing them are passed to the legacy method in a new LaTeX file.
# 4) Call dvipng on the pages without PS literals
# 5) Join metrics info coming from both methods (legacy and dvipng)
# and write them to file
# dvipng is fast but gives problem in several cases, like with
# PSTricks, TikZ and other packages using PostScript literals
# for all these cases the legacy route is taken (step 3).
# Moreover dvipng can't work with PDF files, so, if the CONVERTER
# paramter is pdflatex we have to fallback to legacy route (step 2).
import getopt, glob, os, re, shutil, string, sys
from legacy_lyxpreview2ppm import extract_resolution, legacy_conversion_step1
from lyxpreview_tools import bibtex_commands, check_latex_log, copyfileobj, \
error, filter_pages, find_exe, find_exe_or_terminate, \
join_metrics_and_rename, latex_commands, latex_file_re, make_texcolor, \
mkstemp, pdflatex_commands, progress, run_command, run_latex, run_tex, \
warning, write_metrics_info
def usage(prog_name):
msg = """
Usage: %s <options> <input file>
Options:
--dpi=<res>: Resolution per inch (default: 128)
--png, --ppm: Select the output format (default: png)
--fg=<color>: Foreground color (default: black, ie '000000')
--bg=<color>: Background color (default: white, ie 'ffffff')
--latex=<exe>: Specify the executable for latex (default: latex)
--bibtex=<exe>: Specify the executable for bibtex (default: bibtex)
--lilypond: Preprocess through lilypond-book (default: false)
--lilypond-book=<exe>:
The executable for lilypond-book (default: lilypond-book)
-d, --debug: Show the output from external commands
-h, --help: Show this help screen and exit
-v, --verbose: Show progress messages
The colors are hexadecimal strings, eg 'faf0e6'."""
return msg % prog_name
# Returns a list of tuples containing page number and ascent fraction
# extracted from dvipng output.
# Use write_metrics_info to create the .metrics file with this info
def extract_metrics_info(dvipng_stdout):
# "\[[0-9]+" can match two kinds of numbers: page numbers from dvipng
# and glyph numbers from mktexpk. The glyph numbers always match
# "\[[0-9]+\]" while the page number never is followed by "\]". Thus:
page_re = re.compile("\[([0-9]+)[^]]");
metrics_re = re.compile("depth=(-?[0-9]+) height=(-?[0-9]+)")
success = 0
page = ""
pos = 0
results = []
while 1:
match = page_re.search(dvipng_stdout, pos)
if match == None:
break
page = match.group(1)
pos = match.end()
match = metrics_re.search(dvipng_stdout, pos)
if match == None:
break
success = 1
# Calculate the 'ascent fraction'.
descent = string.atof(match.group(1))
ascent = string.atof(match.group(2))
frac = 0.5
if ascent < 0:
# This is an empty image, forbid its display
frac = -1.0
elif ascent >= 0 or descent >= 0:
if abs(ascent + descent) > 0.1:
frac = ascent / (ascent + descent)
# Sanity check
if frac < 0:
frac = 0.5
results.append((int(page), frac))
pos = match.end() + 2
if success == 0:
error("Failed to extract metrics info from dvipng")
return results
def fix_latex_file(latex_file, pdf_output):
def_re = re.compile(r"(\\newcommandx|\\global\\long\\def)(\\[a-zA-Z]+)")
tmp = mkstemp()
changed = False
macros = []
for line in open(latex_file, 'r').readlines():
if not pdf_output and line.startswith("\\documentclass"):
changed = True
line += "\\PassOptionsToPackage{draft}{microtype}\n"
else:
match = def_re.match(line)
if match != None:
macroname = match.group(2)
if macroname in macros:
definecmd = match.group(1)
if definecmd == "\\newcommandx":
changed = True
line = line.replace(definecmd, "\\renewcommandx")
else:
macros.append(macroname)
tmp.write(line)
if changed:
copyfileobj(tmp, open(latex_file,"wb"), 1)
return changed
def convert_to_ppm_format(pngtopnm, basename):
png_file_re = re.compile("\.png$")
for png_file in glob.glob("%s*.png" % basename):
ppm_file = png_file_re.sub(".ppm", png_file)
p2p_cmd = '%s "%s"' % (pngtopnm, png_file)
p2p_status, p2p_stdout = run_command(p2p_cmd)
if p2p_status:
error("Unable to convert %s to ppm format" % png_file)
ppm = open(ppm_file, 'w')
ppm.write(p2p_stdout)
os.remove(png_file)
# Returns a tuple of:
# ps_pages: list of page indexes of pages containing PS literals
# pdf_pages: list of page indexes of pages requiring running pdflatex
# page_count: total number of pages
# pages_parameter: parameter for dvipng to exclude pages with PostScript/PDF
def find_ps_pages(dvi_file):
# latex failed
# FIXME: try with pdflatex
if not os.path.isfile(dvi_file):
error("No DVI output.")
# Check for PostScript specials in the dvi, badly supported by dvipng,
# and inclusion of PDF/PNG/JPG files.
# This is required for correct rendering of PSTricks and TikZ
dv2dt = find_exe_or_terminate(["dv2dt"])
dv2dt_call = '%s "%s"' % (dv2dt, dvi_file)
# The output from dv2dt goes to stdout
dv2dt_status, dv2dt_output = run_command(dv2dt_call)
psliteral_re = re.compile("^special[1-4] [0-9]+ '(\"|ps:)")
hyperref_re = re.compile("^special[1-4] [0-9]+ 'ps:.*/DEST pdfmark")
pdffile_re = re.compile("^special[1-4] [0-9]+ 'PSfile=.*\\.(pdf|png|jpg|jpeg|PDF|PNG|JPG|JPEG)")
# Parse the dtl file looking for PostScript specials and pdflatex files.
# Pages using PostScript specials or pdflatex files are recorded in
# ps_pages or pdf_pages, respectively, and then used to create a
# different LaTeX file for processing in legacy mode.
# If hyperref is detected, the corresponding page is recorded in pdf_pages.
page_has_ps = False
page_has_pdf = False
page_index = 0
ps_pages = []
pdf_pages = []
ps_or_pdf_pages = []
for line in dv2dt_output.split("\n"):
# New page
if line.startswith("bop"):
page_has_ps = False
page_has_pdf = False
page_index += 1
# End of page
if line.startswith("eop") and (page_has_ps or page_has_pdf):
# We save in a list all the PostScript/PDF pages
if page_has_ps:
ps_pages.append(page_index)
else:
pdf_pages.append(page_index)
ps_or_pdf_pages.append(page_index)
if psliteral_re.match(line) != None:
# Literal PostScript special detected!
# If hyperref is detected, put this page on the pdf pages list
if hyperref_re.match(line) != None:
page_has_ps = False
page_has_pdf = True
else:
page_has_ps = True
elif pdffile_re.match(line) != None:
# Inclusion of pdflatex image file detected!
page_has_pdf = True
# Create the -pp parameter for dvipng
pages_parameter = ""
if len(ps_or_pdf_pages) > 0 and len(ps_or_pdf_pages) < page_index:
# Don't process Postscript/PDF pages with dvipng by selecting the
# wanted pages through the -pp parameter. E.g., dvipng -pp 4-12,14,64
pages_parameter = " -pp "
skip = True
last = -1
# Use page ranges, as a list of pages could exceed command line
# maximum length (especially under Win32)
for index in xrange(1, page_index + 1):
if (not index in ps_or_pdf_pages) and skip:
# We were skipping pages but current page shouldn't be skipped.
# Add this page to -pp, it could stay alone or become the
# start of a range.
pages_parameter += str(index)
# Save the starting index to avoid things such as "11-11"
last = index
# We're not skipping anymore
skip = False
elif (index in ps_or_pdf_pages) and (not skip):
# We weren't skipping but current page should be skipped
if last != index - 1:
# If the start index of the range is the previous page
# then it's not a range
pages_parameter += "-" + str(index - 1)
# Add a separator
pages_parameter += ","
# Now we're skipping
skip = True
# Remove the trailing separator
pages_parameter = pages_parameter.rstrip(",")
# We've to manage the case in which the last page is closing a range
if (not index in ps_or_pdf_pages) and (not skip) and (last != index):
pages_parameter += "-" + str(index)
return (ps_pages, pdf_pages, page_index, pages_parameter)
def main(argv):
# Set defaults.
dpi = 128
fg_color = "000000"
bg_color = "ffffff"
bibtex = None
latex = None
lilypond = False
lilypond_book = None
output_format = "png"
script_name = argv[0]
# Parse and manipulate the command line arguments.
try:
(opts, args) = getopt.gnu_getopt(argv[1:], "dhv", ["bibtex=", "bg=",
"debug", "dpi=", "fg=", "help", "latex=", "lilypond",
"lilypond-book=", "png", "ppm", "verbose"])
except getopt.GetoptError, err:
error("%s\n%s" % (err, usage(script_name)))
opts.reverse()
for opt, val in opts:
if opt in ("-h", "--help"):
print usage(script_name)
sys.exit(0)
elif opt == "--bibtex":
bibtex = [val]
elif opt == "--bg":
bg_color = val
elif opt in ("-d", "--debug"):
import lyxpreview_tools
lyxpreview_tools.debug = True
elif opt == "--dpi":
try:
dpi = string.atoi(val)
except:
error("Cannot convert %s to an integer value" % val)
elif opt == "--fg":
fg_color = val
elif opt == "--latex":
latex = [val]
elif opt == "--lilypond":
lilypond = True
elif opt == "--lilypond-book":
lilypond_book = [val]
elif opt in ("--png", "--ppm"):
output_format = opt[2:]
elif opt in ("-v", "--verbose"):
import lyxpreview_tools
lyxpreview_tools.verbose = True
# Determine input file
if len(args) != 1:
err = "A single input file is required, %s given" % (len(args) or "none")
error("%s\n%s" % (err, usage(script_name)))
input_path = args[0]
dir, latex_file = os.path.split(input_path)
# Echo the settings
progress("Starting %s..." % script_name)
progress("Output format: %s" % output_format)
progress("Foreground color: %s" % fg_color)
progress("Background color: %s" % bg_color)
progress("Resolution (dpi): %s" % dpi)
progress("File to process: %s" % input_path)
# Check for the input file
if not os.path.exists(input_path):
error('File "%s" not found.' % input_path)
if len(dir) != 0:
os.chdir(dir)
fg_color_dvipng = make_texcolor(fg_color, False)
bg_color_dvipng = make_texcolor(bg_color, False)
# External programs used by the script.
latex = find_exe_or_terminate(latex or latex_commands)
bibtex = find_exe(bibtex or bibtex_commands)
if lilypond:
lilypond_book = find_exe_or_terminate(lilypond_book or
["lilypond-book --safe"])
# These flavors of latex are known to produce pdf output
pdf_output = latex in pdflatex_commands
progress("Latex command: %s" % latex)
progress("Latex produces pdf output: %s" % pdf_output)
progress("Bibtex command: %s" % bibtex)
progress("Lilypond-book command: %s" % lilypond_book)
progress("Preprocess through lilypond-book: %s" % lilypond)
progress("Altering the latex file for font size and colors")
# Make sure that multiple defined macros and the microtype package
# don't cause issues in the latex file.
fix_latex_file(latex_file, pdf_output)
if lilypond:
progress("Preprocess the latex file through %s" % lilypond_book)
if pdf_output:
lilypond_book += " --pdf"
lilypond_book += " --latex-program=%s" % latex.split()[0]
# Make a copy of the latex file
lytex_file = latex_file_re.sub(".lytex", latex_file)
shutil.copyfile(latex_file, lytex_file)
# Preprocess the latex file through lilypond-book.
lytex_status, lytex_stdout = run_tex(lilypond_book, lytex_file)
if pdf_output:
progress("Using the legacy conversion method (PDF support)")
return legacy_conversion_step1(latex_file, dpi, output_format, fg_color,
bg_color, latex, pdf_output)
# This can go once dvipng becomes widespread.
dvipng = find_exe(["dvipng"])
if dvipng == None:
progress("Using the legacy conversion method (dvipng not found)")
return legacy_conversion_step1(latex_file, dpi, output_format, fg_color,
bg_color, latex, pdf_output)
dv2dt = find_exe(["dv2dt"])
if dv2dt == None:
progress("Using the legacy conversion method (dv2dt not found)")
return legacy_conversion_step1(latex_file, dpi, output_format, fg_color,
bg_color, latex, pdf_output)
pngtopnm = ""
if output_format == "ppm":
pngtopnm = find_exe(["pngtopnm"])
if pngtopnm == None:
progress("Using the legacy conversion method (pngtopnm not found)")
return legacy_conversion_step1(latex_file, dpi, output_format,
fg_color, bg_color, latex, pdf_output)
# Compile the latex file.
error_pages = []
latex_status, latex_stdout = run_latex(latex, latex_file, bibtex)
latex_log = latex_file_re.sub(".log", latex_file)
if latex_status:
progress("Will try to recover from %s failure" % latex)
error_pages = check_latex_log(latex_log)
# The dvi output file name
dvi_file = latex_file_re.sub(".dvi", latex_file)
# If there's no DVI output, look for PDF and go to legacy or fail
if not os.path.isfile(dvi_file):
# No DVI, is there a PDF?
pdf_file = latex_file_re.sub(".pdf", latex_file)
if os.path.isfile(pdf_file):
progress("%s produced a PDF output, fallback to legacy." \
% (os.path.basename(latex)))
progress("Using the legacy conversion method (PDF support)")
return legacy_conversion_step1(latex_file, dpi, output_format,
fg_color, bg_color, latex, True)
else:
error("No DVI or PDF output. %s failed." \
% (os.path.basename(latex)))
# Look for PS literals or inclusion of pdflatex files in DVI pages
# ps_pages: list of indexes of pages containing PS literals
# pdf_pages: list of indexes of pages requiring running pdflatex
# page_count: total number of pages
# pages_parameter: parameter for dvipng to exclude pages with PostScript
(ps_pages, pdf_pages, page_count, pages_parameter) = find_ps_pages(dvi_file)
# If all pages need PostScript or pdflatex, directly use the legacy method.
if len(ps_pages) == page_count:
progress("Using the legacy conversion method (PostScript support)")
return legacy_conversion_step1(latex_file, dpi, output_format, fg_color,
bg_color, latex, pdf_output)
elif len(pdf_pages) == page_count:
progress("Using the legacy conversion method (PDF support)")
return legacy_conversion_step1(latex_file, dpi, output_format, fg_color,
bg_color, "pdflatex", True)
# Retrieve resolution
resolution = extract_resolution(latex_log, dpi)
# Run the dvi file through dvipng.
dvipng_call = '%s -Ttight -depth -height -D %d -fg "%s" -bg "%s" %s "%s"' \
% (dvipng, resolution, fg_color_dvipng, bg_color_dvipng, pages_parameter, dvi_file)
dvipng_status, dvipng_stdout = run_command(dvipng_call)
if dvipng_status:
warning("%s failed to generate images from %s... fallback to legacy method" \
% (os.path.basename(dvipng), dvi_file))
progress("Using the legacy conversion method (dvipng failed)")
return legacy_conversion_step1(latex_file, dpi, output_format, fg_color,
bg_color, latex, pdf_output)
# Extract metrics info from dvipng_stdout.
metrics_file = latex_file_re.sub(".metrics", latex_file)
dvipng_metrics = extract_metrics_info(dvipng_stdout)
# If some pages require PostScript pass them to legacy method
if len(ps_pages) > 0:
# Create a new LaTeX file just for the snippets needing
# the legacy method
legacy_latex_file = latex_file_re.sub("_legacy.tex", latex_file)
filter_pages(latex_file, legacy_latex_file, ps_pages)
# Pass the new LaTeX file to the legacy method
progress("Pages %s include postscript specials" % ps_pages)
progress("Using the legacy conversion method (PostScript support)")
legacy_status, legacy_metrics = legacy_conversion_step1(legacy_latex_file,
dpi, output_format, fg_color, bg_color, latex, pdf_output, True)
# Now we need to mix metrics data from dvipng and the legacy method
original_bitmap = latex_file_re.sub("%d." + output_format, legacy_latex_file)
destination_bitmap = latex_file_re.sub("%d." + output_format, latex_file)
# Join metrics from dvipng and legacy, and rename legacy bitmaps
join_metrics_and_rename(dvipng_metrics, legacy_metrics, ps_pages,
original_bitmap, destination_bitmap)
# If some pages require running pdflatex pass them to legacy method
if len(pdf_pages) > 0:
# Create a new LaTeX file just for the snippets needing
# the legacy method
legacy_latex_file = latex_file_re.sub("_legacy.tex", latex_file)
filter_pages(latex_file, legacy_latex_file, pdf_pages)
# Pass the new LaTeX file to the legacy method
progress("Pages %s require processing with pdflatex" % pdf_pages)
progress("Using the legacy conversion method (PDF support)")
legacy_status, legacy_metrics = legacy_conversion_step1(legacy_latex_file,
dpi, output_format, fg_color, bg_color, "pdflatex", True, True)
# Now we need to mix metrics data from dvipng and the legacy method
original_bitmap = latex_file_re.sub("%d." + output_format, legacy_latex_file)
destination_bitmap = latex_file_re.sub("%d." + output_format, latex_file)
# Join metrics from dvipng and legacy, and rename legacy bitmaps
join_metrics_and_rename(dvipng_metrics, legacy_metrics, pdf_pages,
original_bitmap, destination_bitmap)
# Invalidate metrics for pages that produced errors
if len(error_pages) > 0:
error_count = 0
for index in error_pages:
if index not in ps_pages and index not in pdf_pages:
dvipng_metrics.pop(index - 1)
dvipng_metrics.insert(index - 1, (index, -1.0))
error_count += 1
if error_count:
warning("Failed to produce %d preview snippet(s)" % error_count)
# Convert images to ppm format if necessary.
if output_format == "ppm":
convert_to_ppm_format(pngtopnm, latex_file_re.sub("", latex_file))
# Actually create the .metrics file
write_metrics_info(dvipng_metrics, metrics_file)
return (0, dvipng_metrics)
if __name__ == "__main__":
sys.exit(main(sys.argv)[0])