#!/usr/bin/env python # -*- coding: iso-8859-1 -*- # Copyright (C) 2012-2014 Bastian Kleineidam from __future__ import print_function import sys import os import time import cgi import codecs import json sys.path.append(os.path.join(os.path.dirname(__file__), "..")) from dosagelib.scraper import get_scraperclasses from dosagelib.util import getLangName class Status: """Status of a comic strip.""" ok = u"ok" error = u"error" comicdata_template = u""" /* generated on %(date)s */ $(document).ready(function() { $('#comics').html('
'); $('#comictable').dataTable( { "aaData": [ %(content)s ], "aoColumns": [ { "sTitle": "Name" }, { "sTitle": "Genre" }, { "sTitle": "Language" }, { "sTitle": "Status" }, { "sTitle": "Votes" } ] } ); } ); """ comic_template = u"""\ title: %(name)s url: "/comics/%(pageurl)s" --- Dosage comic %(name)s -----------------------------------------

Description%(description)s
Website%(url)s
Genre%(genre)s
Language%(language)s
Adult content%(adult)s
Status%(status)s on %(date)s
Votes%(vote)s
[Edit this info](%(editurl)s) or go back to the [comic list](../comic-index.html). """ comic_edit_template = u"""\ title: Edit %(name)s url: "/comics/%(editurl)s" --- Edit info for comic %(name)s
Description
Website
Genre
Language
Adult content
Back to the [comic](%(pageurl)s). """ def get_mtime (filename): """Return modification time of filename.""" return os.path.getmtime(filename) def strdate(t): """Get formatted date string.""" return time.strftime("%d.%m.%Y", time.localtime(t)) def get_testscraper(line): """Get scraper from test output line.""" classname = line.split('::')[1][4:] for scraperclass in get_scraperclasses(): if scraperclass.__name__ == classname: return scraperclass raise ValueError("Scraper %r not found" % classname) def get_testinfo(filename, modified): """Maintains a static list of comics which users can vote on. @return: {name -> { "status": Status.*, "url": string, "description": string, "error": string or None, "genre": string, "language": string, "adult": bool, } } """ testinfo = {} with open(filename, "r") as f: print("Tests parsed: 0", end=" ", file=sys.stderr) num_tests = 0 add_error = False keys = [] for line in f: try: if line.startswith((". ", "F ")) and "test_comics" in line: add_error = line.startswith("F ") num_tests += 1 key, entry = get_testentry(line) keys.append(key) update_testentry(key, entry, testinfo) if num_tests % 5 == 0: print(num_tests, end=" ", file=sys.stderr) elif add_error and line.startswith(" E "): entry["error"] = line[3:].strip() except ValueError as msg: print(msg) return testinfo def get_testentry(line): """Get one test entry.""" scraper = get_testscraper(line) key = scraper.__name__ name = scraper.getName() if len(name) > 40: name = name[:37] + u"..." genres = u",".join(sorted(scraper.genres)) if isinstance(scraper.description, unicode): desc = scraper.description else: desc = scraper.description.decode("utf-8", "ignore") if not desc: desc = u'-' entry = { "status": Status.ok if line.startswith(u". ") else Status.error, "name": name, "url": scraper.url, "description": desc, "genre": genres, "language": getLangName(scraper.lang), "error": None, "adult": scraper.adult, "vote": get_vote(name), } return key, entry def update_testentry(key, entry, testinfo): """Update one entry with testinfo information.""" testinfo[key] = entry _votes = None def get_vote(name): global _votes if _votes is None: fname = os.path.join('data', 'votes.json') with open(fname, 'rb') as f: _votes = json.load(f) key = name.replace('/', '_') return _votes.get(key, 0) def get_comicdata(testinfo): """Get comic data for table listing.""" rows = [] for key in sorted(testinfo.keys()): entry = testinfo[key] url = "comics/%s.html" % key args = { "url": quote(url), "status": quote(entry["status"]), "name": quote(entry["name"]), "language": quote(entry["language"]), "genre": quote(entry["genre"]), "vote": entry["vote"], } row = u'["%(name)s", "%(genre)s", "%(language)s", "%(status)s", "%(vote)d"]' % args rows.append(row) return u",\n".join(rows) def write_html(testinfo, outputdir, modified): """Write index page and all comic pages.""" content = get_comicdata(testinfo) date = unicode(strdate(modified)) args = {"date": quote(date), "content": content} fname = os.path.join("media", "js", "comicdata.js") with codecs.open(fname, 'w', 'utf-8') as fp: fp.write(comicdata_template % args) comicdir = os.path.join(outputdir, "comics") if not os.path.isdir(comicdir): os.mkdir(comicdir) for key, entry in testinfo.items(): write_html_comic(key, entry, comicdir, date) def write_html_comic(key, entry, outputdir, date): """Write a comic page.""" args = { "url": quote(entry["url"]), "pageurl": quote("%s.html" % key), "editurl": quote("%s_edit.html" % key), "name": quote(entry["name"]), "comic": quote(key), "adult": quote(u"yes" if entry["adult"] else u"no"), "adultchecked": 'checked="checked"' if entry["adult"] else '', "genre": quote(entry["genre"]), "language": quote(entry["language"]), "description": quote(entry["description"]), "status": quote(entry["status"]), "date": quote(date), "vote": entry["vote"], "votename": key, } fname = os.path.join(outputdir, key+".md") with codecs.open(fname, 'w', 'utf-8') as fp: fp.write(comic_template % args) fname = os.path.join(outputdir, key+"_edit.md") with codecs.open(fname, 'w', 'utf-8') as fp: fp.write(comic_edit_template % args) def quote(arg): """CGI-escape and jinja-escape the argument.""" return cgi.escape(arg.replace(u'{', u'').replace(u'}', u''), quote=True) def main(args): """Generate HTML output for test result.""" filename = args[0] outputdir = args[1] modified = get_mtime(filename) testinfo = get_testinfo(filename, modified) write_html(testinfo, outputdir, modified) return 0 if __name__ == '__main__': sys.exit(main(sys.argv[1:]))