biskuteri-cafe-JKomasto2/BasicHTMLParser.java

import java.util.List;
import java.util.ArrayList;
import java.util.Deque;
import java.util.LinkedList;
import java.util.Locale;
import java.text.BreakIterator;
import cafe.biskuteri.hinoki.Tree;

interface
BasicHTMLParser {

	public static Tree<String>
	parse(String html)
	{
		List<String> segments;
		segments = distinguishTagsFromPcdata(html);

		Tree<String> document;
		document = toNodes(segments);
		document = splitText(document);
		document = evaluateHtmlEscapes(document);
		document = hierarchise(document);

		return document;
	}

//	 -	-%-	 -

	private static List<String>
	distinguishTagsFromPcdata(String html)
	{
		List<String> returnee = new ArrayList<>();
		StringBuilder segment = new StringBuilder();
		boolean inTag = false;
		for (char c: html.toCharArray())
		{
			if (c == '<')
			{
				String addee = empty(segment);
				if (!addee.isEmpty()) returnee.add(addee);
				inTag = true;
				segment.append(c);
			}
			else if (c == '>')
			{
				assert inTag;
				assert segment.length() > 0;
				segment.append(c);
				returnee.add(empty(segment));
				inTag = false;
			}
			else
			{
				segment.append(c);
			}
		}
		String addee = empty(segment);
		if (!addee.isEmpty()) returnee.add(addee);

		return returnee;
	}

	private static Tree<String>
	toNodes(List<String> segments)
	{
		Tree<String> returnee = new Tree<String>();

		for (String segment: segments)
		{
			boolean isTag = segment.startsWith("<");
			Tree<String> node = new Tree<String>();

			if (!isTag)
			{
				node.key = "text";
				node.value = segment;
				returnee.add(node);
				continue;
			}

			node.key = "tag";

			String key = null, value = null;
			StringBuilder b = new StringBuilder();
			boolean inQuotes = false, inValue = false;
			char[] chars = segment.toCharArray();
			for (int o = 1; o < chars.length - 1; ++o)
			{
				char c = chars[o];
				if (c == '"')
				{
					inQuotes = !inQuotes;
				}
				else if (inQuotes)
				{
					b.append(c);
				}
				else if (c == '=')
				{
					assert b.length() > 0;
					key = empty(b);
					inValue = true;
				}
				else if (Character.isWhitespace(c))
				{
					if (b.length() > 0)
					{
						if (inValue) value = empty(b);
						else key = empty(b);
						Tree<String> attr = new Tree<String>();
						attr.key = key;
						attr.value = value;
						node.add(attr);
					}
					inValue = false;
				}
				else
				{
					b.append(c);
				}
			}
			if (b.length() > 0)
			{
				if (inValue) value = empty(b);
				else key = empty(b);
				Tree<String> attr = new Tree<String>();
				attr.key = key;
				attr.value = value;
				node.add(attr);
			}

			returnee.add(node);
		}

		return returnee;
	}

	private static Tree<String>
	splitText(Tree<String> nodes)
	{
		Tree<String> returnee = new Tree<>();

		for (Tree<String> node: nodes)
		{
			if (node.key.equals("tag"))
			{
				returnee.add(node);
				continue;
			}
			assert node.key.equals("text");

			StringBuilder b = new StringBuilder();
			boolean alnum = false, calnum;
			boolean space = false, cspace;
			boolean emoji = false;
			for (char c: node.value.toCharArray())
			{
				calnum = isMastodonAlnum(c);
				cspace = Character.isWhitespace(c);

				if (c == ':' && !emoji)
				{
                    // See note on #isMastodonAlnum.

					if (b.length() > 0)
					{
						Tree<String> addee = new Tree<>();
						addee.key = space ? "space" : "text";
						addee.value = empty(b);
						returnee.add(addee);
					}
					emoji = true;
					b.append(c);
				}
				else if (c == ':' && emoji)
				{
					assert !space;
					b.append(c);
					Tree<String> addee = new Tree<>();
					addee.key = "emoji";
					addee.value = empty(b);
					returnee.add(addee);
					/*
					* Technically, addee.value.length()
					* could be zero, which probably means
					* someone just put two colons in a row,
					* maybe for Haskell source code. I'd
					* be surprised if Mastodon didn't escape
					* it. (If they did, the next step will
					* handle them.) Anyways treating it as
					* an empty emoji is the correct action.
					*/
					emoji = false;
					calnum = false;
				}
				else if (cspace != space)
				{
                    if (b.length() > 0)
					{
						Tree<String> addee = new Tree<>();
						addee.key = space ? "space" : "text";
						addee.value = empty(b);
						returnee.add(addee);
					}
					b.append(c);
				}
				else
				{
					b.append(c);
				}
				/*
				* We can specially handle special
				* characters like \n, but I'll opt not to.
				*/

				alnum = calnum;
				space = cspace;
			}
			if (b.length() > 0)
			{
				Tree<String> addee = new Tree<>();
				addee.key = space ? "space" : "text";
				addee.value = empty(b);
				returnee.add(addee);
			}
		}

		return returnee;
	}

	private static Tree<String>
	evaluateHtmlEscapes(Tree<String> nodes)
	{
		for (Tree<String> node: nodes)
		{
			node.value = evaluateHtmlEscapes(node.value);
			for (Tree<String> attr: node)
			{
				attr.key = evaluateHtmlEscapes(attr.key);
				attr.value = evaluateHtmlEscapes(attr.value);
			}
		}

		return nodes;
	}

	private static Tree<String>
	hierarchise(Tree<String> nodes)
	{
		Tree<String> root = new Tree<String>();
		root.key = "tag";
		root.add(new Tree<>("html", null));
		root.add(new Tree<>("children", null));

		Deque<Tree<String>> parents = new LinkedList<>();
		parents.push(root);
		for (Tree<String> node: nodes)
		{
			if (node.key.equals("tag"))
			{
				assert node.size() > 0;
				String tagName = node.get(0).key;

				assert node.get("children") == null;
				node.add(new Tree<>("children", null));

				boolean isClosing, selfClosing;
				isClosing = tagName.startsWith("/");
				selfClosing = node.get("/") != null;
				selfClosing |= tagName.equals("br");
				if (isClosing)
				{
					assert parents.size() > 1;

					Tree<String> parent, grandparent;
					parent = parents.pop();
					grandparent = parents.peek();

					String pTagName = parent.get(0).key;
					assert tagName.equals("/" + pTagName);

					grandparent.get("children").add(parent);
				}
				else if (selfClosing)
				{
					parents.peek().get("children").add(node);
				}
				else
				{
					parents.push(node);
				}
			}
			else
			{
				parents.peek().get("children").add(node);
			}
		}

		assert parents.size() == 1;
		return parents.pop();
	}

	private static String
	empty(StringBuilder b)
	{
		String s = b.toString();
		b.delete(0, b.length());
		return s;
	}

	private static boolean
	isMastodonAlnum(char c)
	{
        return Character.isLetterOrDigit(c);
        /*
        * Not joking. Mastodon is using the POSIX :alnum: regex
        * character class here (/app/lib/emoji_formatter.rb;
        * ruby-doc§Regexp). It prevents emojis preceeded by
        * Japanese like さ too, but not punctuation like tildes
        * or full stops. This is server-enforced, the web client
        * does string substitution and supports anything.
        * (To see this, make a post with an emoji preceeded
        * by text, then try again with the same emoji also
        * present elsewhere in the post at a valid position.)
        */
	}

	private static String
	evaluateHtmlEscapes(String string)
	{
		if (string == null) return string;

		StringBuilder whole = new StringBuilder();
		StringBuilder part = new StringBuilder();
		boolean inEscape = false;
		for (char c: string.toCharArray())
		{
			if (inEscape && c == ';')
			{
				part.append(c);
				inEscape = false;
				String v = empty(part);
				if (v.equals("&lt;")) part.append('<');
				if (v.equals("&gt;")) part.append('>');
				if (v.equals("&amp;")) part.append('&');
				if (v.equals("&quot;")) part.append('"');
				if (v.equals("&apos;")) part.append('\'');
				if (v.equals("&#39;")) part.append('\'');
			}
			else if (!inEscape && c == '&')
			{
				String v = empty(part);
				if (!v.isEmpty()) whole.append(v);
				part.append(c);
				inEscape = true;
			}
			else
			{
				part.append(c);
			}
		}
		String v = empty(part);
		if (!v.isEmpty()) whole.append(v);
		return whole.toString();
	}
}