treerack/choice.go

package parse

type choiceDefinition struct {
	name     string
	commit   CommitType
	elements []string
}

type choiceParser struct {
	name      string
	commit    CommitType
	elements  []parser
	including []parser
}

func newChoice(name string, ct CommitType, elements []string) *choiceDefinition {
	return &choiceDefinition{
		name:     name,
		commit:   ct,
		elements: elements,
	}
}

func (d *choiceDefinition) nodeName() string { return d.name }

// could store and cache everything that it fulfils

func (d *choiceDefinition) parser(r *registry, path []string) (parser, error) {
	p, ok := r.parser(d.name)
	if ok {
		return p, nil
	}

	cp := &choiceParser{
		name:   d.name,
		commit: d.commit,
	}

	r.setParser(cp)

	var elements []parser
	path = append(path, d.name)
	for _, e := range d.elements {
		element, ok := r.parser(e)
		if ok {
			elements = append(elements, element)
			element.setIncludedBy(cp, path)
			continue
		}

		elementDefinition, ok := r.definition(e)
		if !ok {
			return nil, parserNotFound(e)
		}

		element, err := elementDefinition.parser(r, path)
		if err != nil {
			return nil, err
		}

		element.setIncludedBy(cp, path)
		elements = append(elements, element)
	}

	cp.elements = elements
	return cp, nil
}

func (d *choiceDefinition) commitType() CommitType {
	return d.commit
}

func (p *choiceParser) nodeName() string { return p.name }

func (p *choiceParser) setIncludedBy(i parser, path []string) {
	if stringsContain(path, p.name) {
		return
	}

	p.including = append(p.including, i)
}

func (p *choiceParser) cacheIncluded(c *context, n *Node) {
	if !c.excluded(n.from, p.name) {
		return
	}

	nc := newNode(p.name, p.commit, n.from, n.to)
	nc.append(n)
	c.cache.set(nc.from, p.name, nc)

	// maybe it is enough to cache only those that are on the path
	for _, i := range p.including {
		i.cacheIncluded(c, nc)
	}
}

func (p *choiceParser) parse(t Trace, c *context) {
	t = t.Extend(p.name)
	t.Out1("parsing choice", c.offset)

	if p.commit&Documentation != 0 {
		t.Out1("fail, doc")
		c.fail(c.offset)
		return
	}

	if m, ok := c.fromCache(p.name); ok {
		t.Out1("found in cache, match:", m)
		return
	}

	if c.excluded(c.offset, p.name) {
		t.Out1("excluded")
		c.fail(c.offset)
		return
	}

	c.exclude(c.offset, p.name)
	defer c.include(c.offset, p.name)

	node := newNode(p.name, p.commit, c.offset, c.offset)
	var match bool

	for {
		elements := p.elements
		var foundMatch bool

		// TODO: this can be the entry point for a transformation that enables the
		// processing of massive amounts of autogenerated rules in parallel in a
		// continously, dynamically cached way. E.g. teach a machine that learns
		// everything from a public library.

		t.Out2("elements again")
		for len(elements) > 0 {
			t.Out2("in the choice", c.offset, node.from, elements[0].nodeName())
			elements[0].parse(t, c)
			elements = elements[1:]
			c.offset = node.from

			if !c.match || match && c.node.tokenLength() <= node.tokenLength() {
				t.Out2("skipping")
				continue
			}

			t.Out2("appending", c.node.tokenLength(), node.tokenLength(),
				"\"", string(c.tokens[node.from:node.to]), "\"",
				"\"", string(c.tokens[c.node.from:c.node.to]), "\"",
				c.node.Name,
			)
			match = true
			foundMatch = true
			// node.clear()
			node = newNode(p.name, p.commit, c.offset, c.offset) // TODO: review caching conditions
			node.append(c.node)

			c.cache.set(node.from, p.name, node)
			for _, i := range p.including {
				i.cacheIncluded(c, node)
			}

			// TODO: a simple break here can force PEG-style "priority" choices
		}

		if !foundMatch {
			break
		}
	}

	if match {
		t.Out1("choice, success")
		t.Out2("choice done", node.nodeLength())
		c.success(node)
		return
	}

	t.Out1("fail")
	c.cache.set(node.from, p.name, nil)
	c.fail(node.from)
}