X-Git-Url: https://git.mdrn.pl/librarian.git/blobdiff_plain/a67b6e3ad2751083a433228454665c212c9a5a1e..e868b7e6511dfbb89e5a9dd69bbb7ef13c42620a:/librarian/pdf.py diff --git a/librarian/pdf.py b/librarian/pdf.py index b9ead15..476fbee 100644 --- a/librarian/pdf.py +++ b/librarian/pdf.py @@ -11,33 +11,33 @@ from StringIO import StringIO from tempfile import mkdtemp import re from copy import deepcopy +from subprocess import call, PIPE import sys -sys.path.append('..') # for running from working copy from Texml.processor import process from lxml import etree from lxml.etree import XMLSyntaxError, XSLTApplyError +from librarian.dcparser import Person from librarian.parser import WLDocument -from librarian import ParseError +from librarian import ParseError, DCNS, get_resource from librarian import functions - functions.reg_substitute_entities() -functions.reg_person_name() functions.reg_strip() functions.reg_starts_white() functions.reg_ends_white() +functions.reg_texcommand() STYLESHEETS = { - 'wl2tex': 'xslt/wl2tex.xslt', + 'wl2tex': 'pdf/wl2tex.xslt', } -def insert_tags(doc, split_re, tagname): - """ inserts for every occurence of `split_re' in text nodes in the `doc' tree +def insert_tags(doc, split_re, tagname, exclude=None): + """ inserts for every occurence of `split_re' in text nodes in the `doc' tree >>> t = etree.fromstring('A-B-CX-Y-Z'); >>> insert_tags(t, re.compile('-'), 'd'); @@ -45,128 +45,250 @@ def insert_tags(doc, split_re, tagname): ABCXYZ """ - for elem in doc.iter(): + for elem in doc.iter(tag=etree.Element): + if exclude and elem.tag in exclude: + continue if elem.text: chunks = split_re.split(elem.text) - elem.text = chunks.pop(0) - while chunks: + while len(chunks) > 1: ins = etree.Element(tagname) ins.tail = chunks.pop() elem.insert(0, ins) + elem.text = chunks.pop(0) if elem.tail: chunks = split_re.split(elem.tail) parent = elem.getparent() ins_index = parent.index(elem) + 1 - elem.tail = chunks.pop(0) - while chunks: + while len(chunks) > 1: ins = etree.Element(tagname) ins.tail = chunks.pop() parent.insert(ins_index, ins) + elem.tail = chunks.pop(0) def substitute_hyphens(doc): - insert_tags(doc, + insert_tags(doc, re.compile("(?<=[^-\s])-(?=[^-\s])"), - "dywiz") + "dywiz", + exclude=[DCNS("identifier.url"), DCNS("rights.license")] + ) def fix_hanging(doc): - insert_tags(doc, + insert_tags(doc, re.compile("(?<=\s\w)\s+"), - "nbsp") + "nbsp", + exclude=[DCNS("identifier.url"), DCNS("rights.license")] + ) + + +def move_motifs_inside(doc): + """ moves motifs to be into block elements """ + for master in doc.xpath('//powiesc|//opowiadanie|//liryka_l|//liryka_lp|//dramat_wierszowany_l|//dramat_wierszowany_lp|//dramat_wspolczesny'): + for motif in master.xpath('motyw'): + for sib in motif.itersiblings(): + if sib.tag not in ('sekcja_swiatlo', 'sekcja_asterysk', 'separator_linia', 'begin', 'end', 'motyw', 'extra', 'uwaga'): + # motif shouldn't have a tail - it would be untagged text + motif.tail = None + motif.getparent().remove(motif) + sib.insert(0, motif) + break + +def hack_motifs(doc): + """ dirty hack for the marginpar-creates-orphans LaTeX problem + see http://www.latex-project.org/cgi-bin/ltxbugs2html?pr=latex/2304 + + moves motifs in stanzas from first verse to second + and from next to last to last, then inserts negative vspace before them + """ + for motif in doc.findall('//strofa//motyw'): + # find relevant verse-level tag + verse, stanza = motif, motif.getparent() + while stanza is not None and stanza.tag != 'strofa': + verse, stanza = stanza, stanza.getparent() + breaks_before = sum(1 for i in verse.itersiblings('br', preceding=True)) + breaks_after = sum(1 for i in verse.itersiblings('br')) + if (breaks_before == 0 and breaks_after > 0) or breaks_after == 1: + move_by = 1 + if breaks_after == 2: + move_by += 1 + moved_motif = deepcopy(motif) + motif.tag = 'span' + motif.text = None + moved_motif.tail = None + moved_motif.set('moved', str(move_by)) + + for br in verse.itersiblings('br'): + if move_by > 1: + move_by -= 1 + continue + br.addnext(moved_motif) + break + + +def parse_creator(doc): + """ find all dc:creator and dc.contributor tags and add *_parsed versions with forenames first """ + for person in doc.xpath("|".join('//dc:'+(tag) for tag in ( + 'creator', 'contributor.translator', 'contributor.editor', 'contributor.technical_editor')), + namespaces = {'dc': str(DCNS)})[::-1]: + if not person.text: + continue + p = Person.from_text(person.text) + person_parsed = deepcopy(person) + person_parsed.tag = person.tag + '_parsed' + person_parsed.set('sortkey', person.text) + person_parsed.text = p.readable() + person.getparent().insert(0, person_parsed) -def get_resource(path): - return os.path.join(os.path.dirname(__file__), path) def get_stylesheet(name): return get_resource(STYLESHEETS[name]) -def transform(provider, slug, output_file=None, output_dir=None): - """ produces a pdf file - provider is a DocProvider - either output_file (a file-like object) or output_dir (path to file/dir) should be specified - if output_dir is specified, file will be written to //.pdf +def package_available(package, args='', verbose=False): + """ check if a verion of a latex package accepting given args is available """ + tempdir = mkdtemp('-wl2pdf-test') + fpath = os.path.join(tempdir, 'test.tex') + f = open(fpath, 'w') + f.write(r""" + \documentclass{book} + \usepackage[%s]{%s} + \begin{document} + \end{document} + """ % (args, package)) + f.close() + if verbose: + p = call(['xelatex', '-output-directory', tempdir, fpath]) + else: + p = call(['xelatex', '-interaction=batchmode', '-output-directory', tempdir, fpath], stdout=PIPE, stderr=PIPE) + shutil.rmtree(tempdir) + return p == 0 + + +def transform(provider, slug=None, file_path=None, + output_file=None, output_dir=None, make_dir=False, verbose=False, save_tex=None, morefloats=None, + cover=None, flags=None): + """ produces a PDF file with XeLaTeX + + provider: a DocProvider + slug: slug of file to process, available by provider + file_path can be provided instead of a slug + output_file: file-like object or path to output file + output_dir: path to directory to save output file to; either this or output_file must be present + make_dir: writes output to //.pdf istead of /.pdf + verbose: prints all output from LaTeX + save_tex: path to save the intermediary LaTeX file to + morefloats (old/new/none): force specific morefloats + cover: a cover.Cover object + flags: less-advertising, """ # Parse XSLT try: - style_filename = get_stylesheet("wl2tex") - style = etree.parse(style_filename) - - document = load_including_children(provider, slug) - - # dirty hack for the marginpar-creates-orphans LaTeX problem - # see http://www.latex-project.org/cgi-bin/ltxbugs2html?pr=latex/2304 - for motif in document.edoc.findall('//strofa//motyw'): - # find relevant verse-level tag - verse, stanza = motif, motif.getparent() - while stanza is not None and stanza.tag != 'strofa': - verse, stanza = stanza, stanza.getparent() - breaks_before = sum(1 for i in verse.itersiblings('br', preceding=True)) - breaks_after = sum(1 for i in verse.itersiblings('br')) - if (breaks_before == 0 and breaks_after > 0) or breaks_after == 1: - move_by = 1 - if breaks_after == 2: - move_by += 1 - moved_motif = deepcopy(motif) - motif.tag = 'span' - motif.text = None - moved_motif.tail = None - moved_motif.set('moved', str(move_by)) - - for br in verse.itersiblings(tag='br'): - if move_by > 1: - move_by -= 1 - continue - br.addnext(moved_motif) - break - + if file_path: + if slug: + raise ValueError('slug or file_path should be specified, not both') + document = load_including_children(provider, file_path=file_path) + else: + if not slug: + raise ValueError('either slug or file_path should be specified') + document = load_including_children(provider, slug=slug) + + if cover: + document.edoc.getroot().set('data-cover-width', str(cover.width)) + document.edoc.getroot().set('data-cover-height', str(cover.height)) + if flags: + for flag in flags: + document.edoc.getroot().set('flag-' + flag, 'yes') + + # check for LaTeX packages + if morefloats: + document.edoc.getroot().set('morefloats', morefloats.lower()) + elif package_available('morefloats', 'maxfloats=19'): + document.edoc.getroot().set('morefloats', 'new') + + # hack the tree + move_motifs_inside(document.edoc) + hack_motifs(document.edoc) + parse_creator(document.edoc) substitute_hyphens(document.edoc) fix_hanging(document.edoc) - # if output to dir, create the file - if output_dir is not None: + # find output dir + if make_dir and output_dir is not None: author = unicode(document.book_info.author) output_dir = os.path.join(output_dir, author) + # wl -> TeXML + style_filename = get_stylesheet("wl2tex") + style = etree.parse(style_filename) texml = document.transform(style) + + # TeXML -> LaTeX + temp = mkdtemp('-wl2pdf') + + if cover: + c = cover(document.book_info.author.readable(), document.book_info.title) + with open(os.path.join(temp, 'cover.png'), 'w') as f: + c.save(f) + del document # no longer needed large object :) - temp = mkdtemp('wl2pdf-') tex_path = os.path.join(temp, 'doc.tex') fout = open(tex_path, 'w') - process(StringIO(texml), fout, 'utf8', 255, 0, 0) + process(StringIO(texml), fout, 'utf-8') fout.close() del texml + if save_tex: + shutil.copy(tex_path, save_tex) + + # LaTeX -> PDF shutil.copy(get_resource('pdf/wl.sty'), temp) - shutil.copy(get_resource('pdf/wl-logo.png'), temp) - print "pdflatex -output-directory %s %s" % (temp, os.path.join(temp, 'doc.tex')) - if os.system("pdflatex -output-directory %s %s" % (temp, os.path.join(temp, 'doc.tex'))): + shutil.copy(get_resource('res/wl-logo.png'), temp) + + cwd = os.getcwd() + os.chdir(temp) + + if verbose: + p = call(['xelatex', tex_path]) + else: + p = call(['xelatex', '-interaction=batchmode', tex_path], stdout=PIPE, stderr=PIPE) + if p: raise ParseError("Error parsing .tex file") + os.chdir(cwd) + + # save the PDF pdf_path = os.path.join(temp, 'doc.pdf') if output_dir is not None: try: os.makedirs(output_dir) except OSError: pass - output_path = os.path.join(output_dir, '%s.pdf' % slug) + if slug: + output_path = os.path.join(output_dir, '%s.pdf' % slug) + else: + output_path = os.path.join(output_dir, os.path.splitext(os.path.basename(file_path))[0] + '.pdf') shutil.move(pdf_path, output_path) else: - with open(pdf_path) as f: - output_file.write(f.read()) - output_file.close() + if hasattr(output_file, 'write'): + # file-like object + with open(pdf_path) as f: + output_file.write(f.read()) + output_file.close() + else: + # path to output file + shutil.copy(pdf_path, output_file) + shutil.rmtree(temp) - return True except (XMLSyntaxError, XSLTApplyError), e: raise ParseError(e) -def load_including_children(provider, slug=None, uri=None): - """ makes one big xml file with children inserted at end +def load_including_children(provider, slug=None, uri=None, file_path=None): + """ makes one big xml file with children inserted at end either slug or uri must be provided """ @@ -174,31 +296,20 @@ def load_including_children(provider, slug=None, uri=None): f = provider.by_uri(uri) elif slug: f = provider[slug] + elif file_path: + f = open(file_path, 'r') else: - raise ValueError('Neither slug nor URI provided for a book.') + raise ValueError('Neither slug, URI nor file path provided for a book.') + + text = f.read().decode('utf-8') + text = re.sub(ur"([\u0400-\u04ff]+)", ur"\1", text) - document = WLDocument.from_file(f, True, - parse_dublincore=True, - preserve_lines=False) + document = WLDocument.from_string(text, True, + parse_dublincore=True) + f.close() for child_uri in document.book_info.parts: + print child_uri child = load_including_children(provider, uri=child_uri) document.edoc.getroot().append(child.edoc.getroot()) - return document - - -if __name__ == '__main__': - import sys - from librarian import DirDocProvider - - if len(sys.argv) < 2: - print >> sys.stderr, 'Usage: python pdf.py ' - sys.exit(1) - - main_input = sys.argv[1] - basepath, ext = os.path.splitext(main_input) - path, slug = os.path.realpath(basepath).rsplit('/', 1) - provider = DirDocProvider(path) - transform(provider, slug, output_dir=path) -