Beispiel #1
0
        void CheckHead()
        {
            Regex reg = new Regex("<head>[\\w\\W]*?</head>");
            Match m   = reg.Match(text);

            if (!m.Success)
            {
                Log.log("Warn:No head tag in xhtml"); return;
            }
            Regex reg_link   = new Regex("<link .*?>");
            Regex reg_script = new Regex("<script .*?>");
            var   ms         = reg_link.Matches(m.Value);

            foreach (Match link in ms)
            {
                XTag tag = new XTag(link.Value);
                if (tag.GetAttribute("type").ToLower() == "text/css")
                {
                    string url = tag.GetAttribute("href");
                    url = Util.ReferPath(filename, url);
                    css.Add(url);
                }
            }
            int   pos  = m.Index;
            Match scpt = reg_script.Match(text, pos);

            while (scpt.Success)
            {
                XTag tag = new XTag(scpt.Value);
                if (tag.GetAttribute("src").Contains("notereplace.js"))
                {
                    string scpt_end = "</script>";
                    int    sei      = text.IndexOf(scpt_end, scpt.Index);
                    if (sei < 0)
                    {
                        Log.log("Error:Unclosed script tag."); break;
                    }
                    text = text.Remove(scpt.Index, sei - scpt.Index + scpt_end.Length);
                    Log.log("Removed reference to notereplace.js");
                    break;
                }
                else
                {
                    pos = scpt.Index + scpt.Length;
                }
                scpt = reg_script.Match(text, pos);
            }
        }
Beispiel #2
0
        public ProcOPF(string root)
        {
            string container = Path.Combine(root, "META-INF/container.xml");

            if (!File.Exists(container))
            {
                Log.log("Error:Cannot find meta-inf"); return;
            }
            string metainf  = File.ReadAllText(container);
            Regex  reg      = new Regex("<rootfile .*?>");
            XTag   tag      = new XTag(reg.Match(metainf).Value);
            string opf_path = tag.GetAttribute("full-path");

            opf_path = Path.Combine(root, opf_path);

            string opf  = File.ReadAllText(opf_path);
            Regex  item = new Regex("<item .*?>");
            Match  m    = item.Match(opf);

            while (m.Success)
            {
                XTag t = new XTag(m.Value);
                if (Path.GetFileName(t.GetAttribute("href")) == "notereplace.js")
                {
                    opf = opf.Remove(m.Index, m.Length); break;
                }
                m = m.NextMatch();
            }

            File.WriteAllText(opf_path, opf);
        }
Beispiel #3
0
        void ProcNote(Match m, XTag tag)
        {
            string note_id = "", ref_id;

            //Link tag solve
            {
                var a = tag.GetClassNames();
                if (!Contains(a, "duokan-footnote"))
                {
                    string added = "duokan-footnote";
                    if (a.Length != 0)
                    {
                        added = " " + added;
                    }
                    tag.SetAttribute("class", tag.GetAttribute("class") + added);
                }
            }

            if (tag.GetAttribute("epub:type") != "noteref")
            {
                tag.SetAttribute("epub:type", "noteref");
            }
            {
                string href = tag.GetAttribute("href");
                int    pt   = href.IndexOf('#');
                if (pt < 0)
                {
                    Log.log("Error:Not a valid link :" + href + ""); return;
                }
                if (pt != 0)
                {
                    Log.log("Warn: href=\":" + href + "\"");
                }
                note_id = href.Substring(pt + 1);
            }

            ref_id = note_id + "_ref";
            tag.SetAttribute("id", ref_id);

            text = text.Remove(m.Index, m.Length);
            text = text.Insert(m.Index, tag.ToString());


            //Note content
            ProcNoteContent(note_id, ref_id);
        }
Beispiel #4
0
        void CheckFootnotes()
        {
            Regex reg_link = new Regex("<a .*?>");
            int   pos      = 0;
            Match link     = reg_link.Match(text);

            while (link.Success)
            {
                XTag tag = new XTag(link.Value);
                if (Contains(tag.GetClassNames(), "duokan-footnote") ||
                    tag.GetAttribute("epub:type") == "noteref")
                {
                    ProcNote(link, tag);
                }
                pos  = link.Index + 1;//假定注释本体都在链接后面
                link = reg_link.Match(text, pos);
            }
        }
Beispiel #5
0
        void ProcNoteContent(string note_id, string ref_id)
        {
            Regex  reg_tag = new Regex("<.*?>");
            Regex  reg_duokan = new Regex("<ol .*?>");
            Regex  reg_aside = new Regex("<aside .*?>");
            int    index = -1, length = 0;
            string note_content = null; string list_value = "1";

            Match m = reg_aside.Match(text);

            while (m.Success)
            {
                XTag tag = new XTag(m.Value);
                if (tag.GetAttribute("id") == note_id)
                {
                    index = m.Index;
                    XFragment frag = new XFragment(text, index);
                    if (frag.root != null)
                    {
                        var dk = frag.root.GetElementById(note_id);
                        if (dk != null)
                        {
                            //做过兼容,aside里套多看li
                            note_content = dk.innerXHTML;
                            list_value   = dk.tag.GetAttribute("value");
                        }
                        else
                        {
                            note_content = frag.root.innerXHTML;
                        }
                        length = frag.Length;
                    }
                    else
                    {
                        Log.log("Error:Found note but failure on parsing. id=" + note_id); return;
                    }
                    break;
                }
                m = m.NextMatch();
            }

            if (index < 0)//如果只对多看适配,没有aside
            {
                m = reg_duokan.Match(text);
                while (m.Success)
                {
                    XFragment frg = new XFragment(text, m.Index);
                    if (frg.root != null)
                    {
                        if (Contains(frg.root.tag.GetClassNames(), "duokan-footnote-content"))
                        {
                            var a = frg.root.GetElementById(note_id);
                            if (a != null)
                            {
                                index        = m.Index;
                                note_content = a.innerXHTML;
                                length       = frg.Length;
                                break;
                            }
                        }
                    }
                    m = m.NextMatch();
                }
            }

            if (note_content == null)
            {
                Log.log("Error:cannot find note"); return;
            }
            string template  = "<aside epub:type=\"footnote\" id=\"{0}\"><a href=\"#{1}\"></a><ol class=\"duokan-footnote-content\" style=\"list-style:none\"><li class=\"duokan-footnote-item\" id=\"{0}\">{2}</li></ol></aside>";
            string note_full = string.Format(template, note_id, ref_id, note_content);

            text = text.Remove(index, length);
            text = text.Insert(index, note_full);

            Log.log("Formated:" + note_id + ":" + note_content);
            contain_footnote = true;
        }