summercms/internal/docsite/render.go

package docsite

import (
	"bytes"
	"cmp"
	"fmt"
	"html"
	"path"
	"slices"
	"strings"
	"unicode"
	"unicode/utf8"

	"github.com/yuin/goldmark"
	"github.com/yuin/goldmark/ast"
	"github.com/yuin/goldmark/extension"
	"github.com/yuin/goldmark/parser"
	"github.com/yuin/goldmark/renderer"
	"github.com/yuin/goldmark/text"
	"github.com/yuin/goldmark/util"
)

// newMarkdown returns the one goldmark pipeline every page goes through:
// CommonMark plus GFM, heading IDs, and the page transformers. The html
// renderer keeps its default safe mode, so raw HTML in Markdown is never
// passed through.
func newMarkdown() goldmark.Markdown {
	return goldmark.New(
		goldmark.WithExtensions(extension.GFM),
		goldmark.WithParserOptions(
			parser.WithAutoHeadingID(),
			parser.WithASTTransformers(
				util.Prioritized(h1Stripper{}, 100),
				util.Prioritized(linkRewriter{}, 200),
				util.Prioritized(fenceAnnotator{}, 300),
			),
		),
		// goldmark registers lower priority values last, so 100 overrides the
		// default html renderer (1000) for fenced code blocks.
		goldmark.WithRendererOptions(renderer.WithNodeRenderers(util.Prioritized(codeRenderer{}, 100))),
	)
}

// fenceAnnotator marks fences that carry src= with the reference and its
// source_url link, for codeRenderer's caption.
type fenceAnnotator struct{}

func (fenceAnnotator) Transform(doc *ast.Document, reader text.Reader, pc parser.Context) {
	pctx, _ := pc.Get(pageKey).(*pageContext)
	src := reader.Source()
	_ = ast.Walk(doc, func(n ast.Node, entering bool) (ast.WalkStatus, error) {
		fc, ok := n.(*ast.FencedCodeBlock)
		if !entering || !ok || fc.Info == nil {
			return ast.WalkContinue, nil
		}
		ref, ok := ParseSrc(string(fc.Info.Segment.Value(src)))
		if !ok {
			return ast.WalkContinue, nil
		}
		fc.SetAttributeString("data-src", []byte(ref.String()))
		if pctx != nil && pctx.site.cfg.SourceURL != "" {
			fc.SetAttributeString("data-href", []byte(strings.ReplaceAll(pctx.site.cfg.SourceURL, "{path}", ref.Path)))
		}
		return ast.WalkSkipChildren, nil
	})
}

// codeRenderer renders every fenced code block inside <figure class="code">;
// a src= fence gets a <figcaption> naming its source, linked to source_url.
type codeRenderer struct{}

func (codeRenderer) RegisterFuncs(r renderer.NodeRendererFuncRegisterer) {
	r.Register(ast.KindFencedCodeBlock, renderFence)
}

func attrString(n ast.Node, name string) string {
	v, ok := n.AttributeString(name)
	if !ok {
		return ""
	}
	b, _ := v.([]byte)
	return string(b)
}

func renderFence(w util.BufWriter, src []byte, node ast.Node, entering bool) (ast.WalkStatus, error) {
	if !entering {
		return ast.WalkContinue, nil
	}
	n := node.(*ast.FencedCodeBlock)
	_, _ = w.WriteString(`<figure class="code">`)
	if ref := attrString(n, "data-src"); ref != "" {
		_, _ = w.WriteString("<figcaption>")
		if href := attrString(n, "data-href"); href != "" {
			_, _ = w.WriteString(`<a href="` + html.EscapeString(href) + `">` + html.EscapeString(ref) + "</a>")
		} else {
			_, _ = w.WriteString(html.EscapeString(ref))
		}
		_, _ = w.WriteString("</figcaption>")
	}
	_, _ = w.WriteString("<pre><code")
	if lang := n.Language(src); lang != nil {
		_, _ = w.WriteString(` class="language-` + html.EscapeString(string(lang)) + `"`)
	}
	_ = w.WriteByte('>')
	for i := 0; i < n.Lines().Len(); i++ {
		line := n.Lines().At(i)
		_, _ = w.WriteString(html.EscapeString(string(line.Value(src))))
	}
	_, _ = w.WriteString("</code></pre></figure>\n")
	return ast.WalkSkipChildren, nil
}

// h1Stripper removes the page's leading "# Title" heading; the template
// renders the title once.
type h1Stripper struct{}

func (h1Stripper) Transform(doc *ast.Document, _ text.Reader, _ parser.Context) {
	if h, ok := doc.FirstChild().(*ast.Heading); ok && h.Level == 1 {
		doc.RemoveChild(doc, h)
	}
}

// slugIDs is the one heading-ID algorithm of the site, shared by the
// renderer, the search index and the link checker. It is GitHub-compatible:
// lowercase, Unicode letters, digits, "_" and "-" kept, spaces mapped to
// "-", other punctuation dropped, and duplicates suffixed -1, -2.
type slugIDs struct {
	used map[string]bool
}

func newSlugIDs() *slugIDs {
	return &slugIDs{used: map[string]bool{}}
}

// Generate implements parser.IDs.
func (s *slugIDs) Generate(value []byte, _ ast.NodeKind) []byte {
	base := slugify(string(value))
	if base == "" {
		base = "section"
	}
	id := base
	for i := 1; s.used[id]; i++ {
		id = fmt.Sprintf("%s-%d", base, i)
	}
	s.used[id] = true
	return []byte(id)
}

// Put implements parser.IDs.
func (s *slugIDs) Put(value []byte) {
	s.used[string(value)] = true
}

func slugify(v string) string {
	var b strings.Builder
	for _, r := range strings.ToLower(strings.TrimSpace(v)) {
		switch {
		case unicode.IsLetter(r) || unicode.IsDigit(r) || r == '_' || r == '-':
			b.WriteRune(r)
		case unicode.IsSpace(r):
			b.WriteByte('-')
		}
	}
	return b.String()
}

var pageKey = parser.NewContextKey()

// pageContext carries per-page state into the AST transformers.
type pageContext struct {
	site *site
	page *Page
	// mdLinks maps a rewritten link destination to its .md form, for the
	// raw Markdown output.
	mdLinks map[string]string
}

// linkRewriter points relative links at other pages to their site URLs:
// a guide's link to another page's .md, or to modules/<m>/README.md, and a
// README's ../<m>/README.md all become the target page. Fragments are kept;
// external and mailto links are untouched.
type linkRewriter struct{}

func (linkRewriter) Transform(doc *ast.Document, _ text.Reader, pc parser.Context) {
	pctx, _ := pc.Get(pageKey).(*pageContext)
	if pctx == nil {
		return
	}
	_ = ast.Walk(doc, func(n ast.Node, entering bool) (ast.WalkStatus, error) {
		link, ok := n.(*ast.Link)
		if !entering || !ok {
			return ast.WalkContinue, nil
		}
		dest := string(link.Destination)
		target, frag, ok := pctx.site.resolveLink(pctx.page, dest)
		if !ok {
			return ast.WalkContinue, nil
		}
		link.Destination = []byte(pctx.site.url(target.URL+".html") + frag)
		pctx.mdLinks[dest] = pctx.site.url(target.URL+".md") + frag
		return ast.WalkContinue, nil
	})
}

// resolveLink maps a relative link destination in page from to the page it
// names, with its "#fragment" (or "").
func (s *site) resolveLink(from *Page, dest string) (*Page, string, bool) {
	if dest == "" || strings.HasPrefix(dest, "#") || strings.HasPrefix(dest, "/") || hasScheme(dest) {
		return nil, "", false
	}
	target, frag, _ := strings.Cut(dest, "#")
	if frag != "" {
		frag = "#" + frag
	}
	if !strings.HasSuffix(target, ".md") {
		return nil, "", false
	}
	p, ok := s.bySource[path.Clean(path.Join(path.Dir(from.Source), target))]
	if !ok {
		return nil, "", false
	}
	return p, frag, true
}

func hasScheme(dest string) bool {
	i := strings.IndexByte(dest, ':')
	return i > 0 && !strings.ContainsAny(dest[:i], "/?#")
}

// heading is one heading of a rendered page.
type heading struct {
	Level int
	ID    string
	Text  string
	node  *ast.Heading
}

// renderedPage is one page after parsing and rendering.
type renderedPage struct {
	html     []byte
	doc      ast.Node
	headings []heading
	mdLinks  map[string]string
}

// parsePage parses a page body with a fresh slug-ID table and the page's
// transformer context.
func (s *site) parsePage(md goldmark.Markdown, p *Page) (ast.Node, *pageContext) {
	pctx := &pageContext{site: s, page: p, mdLinks: map[string]string{}}
	ctx := parser.NewContext(parser.WithIDs(newSlugIDs()))
	ctx.Set(pageKey, pctx)
	doc := md.Parser().Parse(text.NewReader(p.Body), parser.WithContext(ctx))
	return doc, pctx
}

// renderPage parses and renders a page body to HTML.
func (s *site) renderPage(md goldmark.Markdown, p *Page) (renderedPage, error) {
	doc, pctx := s.parsePage(md, p)
	var buf bytes.Buffer
	if err := md.Renderer().Render(&buf, p.Body, doc); err != nil {
		return renderedPage{}, fmt.Errorf("docsite: render %s: %w", p.Source, err)
	}
	return renderedPage{html: buf.Bytes(), doc: doc, headings: pageHeadings(doc, p.Body), mdLinks: pctx.mdLinks}, nil
}

// headingIDs returns the heading IDs of a page, in document order, from the
// same parse and slug algorithm the renderer uses.
func (s *site) headingIDs(md goldmark.Markdown, p *Page) []string {
	doc, _ := s.parsePage(md, p)
	var ids []string
	for _, h := range pageHeadings(doc, p.Body) {
		ids = append(ids, h.ID)
	}
	return ids
}

func pageHeadings(doc ast.Node, src []byte) []heading {
	var out []heading
	_ = ast.Walk(doc, func(n ast.Node, entering bool) (ast.WalkStatus, error) {
		h, ok := n.(*ast.Heading)
		if !entering || !ok {
			return ast.WalkContinue, nil
		}
		id, _ := h.AttributeString("id")
		idb, _ := id.([]byte)
		out = append(out, heading{Level: h.Level, ID: string(idb), Text: plainText(h, src), node: h})
		return ast.WalkSkipChildren, nil
	})
	return out
}

// plainText extracts the text of a node from its text nodes only (never
// from rendered HTML), with whitespace collapsed.
func plainText(n ast.Node, src []byte) string {
	var b strings.Builder
	_ = ast.Walk(n, func(c ast.Node, entering bool) (ast.WalkStatus, error) {
		if !entering {
			return ast.WalkContinue, nil
		}
		switch t := c.(type) {
		case *ast.Text:
			b.Write(t.Segment.Value(src))
			if t.SoftLineBreak() || t.HardLineBreak() {
				b.WriteByte(' ')
			}
		case *ast.String:
			b.Write(t.Value)
		case *ast.FencedCodeBlock, *ast.CodeBlock, *ast.HTMLBlock, *ast.RawHTML:
			return ast.WalkSkipChildren, nil
		case *ast.Paragraph, *ast.Heading, *ast.ListItem, *ast.TextBlock:
			b.WriteByte(' ')
		}
		return ast.WalkContinue, nil
	})
	return strings.Join(strings.Fields(b.String()), " ")
}

// sectionText is the plain text of the blocks after h up to the next
// heading of the same or a higher level, cut to max runes.
func sectionText(h *ast.Heading, src []byte, max int) string {
	var parts []string
	for n := h.NextSibling(); n != nil; n = n.NextSibling() {
		if next, ok := n.(*ast.Heading); ok && next.Level <= h.Level {
			break
		}
		if t := plainText(n, src); t != "" {
			parts = append(parts, t)
		}
	}
	return truncateRunes(strings.Join(parts, " "), max)
}

func truncateRunes(s string, max int) string {
	if utf8.RuneCountInString(s) <= max {
		return s
	}
	r := []rune(s)
	return strings.TrimSpace(string(r[:max]))
}

// fence is one fenced code block found by scanning Markdown lines.
type fence struct {
	open, close int // 0-based line indexes; close == -1 when unterminated
	indent      int
	char        byte
	count       int
	info        string
}

// scanFences finds fenced code blocks in Markdown lines, following the
// CommonMark opening and closing rules.
func scanFences(lines []string) []fence {
	var out []fence
	for i := 0; i < len(lines); i++ {
		f, ok := openFence(lines[i])
		if !ok {
			continue
		}
		f.open, f.close = i, -1
		for j := i + 1; j < len(lines); j++ {
			if closesFence(lines[j], f) {
				f.close = j
				break
			}
		}
		out = append(out, f)
		if f.close < 0 {
			break
		}
		i = f.close
	}
	return out
}

func openFence(line string) (fence, bool) {
	trimmed := strings.TrimLeft(line, " ")
	if len(trimmed) < 3 || (trimmed[0] != '`' && trimmed[0] != '~') {
		return fence{}, false
	}
	c := trimmed[0]
	n := 0
	for n < len(trimmed) && trimmed[n] == c {
		n++
	}
	if n < 3 {
		return fence{}, false
	}
	info := strings.TrimSpace(trimmed[n:])
	if c == '`' && strings.ContainsRune(info, '`') {
		return fence{}, false
	}
	return fence{indent: len(line) - len(trimmed), char: c, count: n, info: info}, true
}

func closesFence(line string, f fence) bool {
	trimmed := strings.TrimLeft(line, " ")
	n := 0
	for n < len(trimmed) && trimmed[n] == f.char {
		n++
	}
	return n >= f.count && strings.TrimSpace(trimmed[n:]) == ""
}

// rewriteMarkdown applies the raw-output transforms to a Markdown body:
// fence info strings are reduced to their language word (dropping src=),
// and link destinations outside code fences are replaced through links.
func rewriteMarkdown(body string, links map[string]string) string {
	lines := strings.Split(body, "\n")
	inFence := make([]bool, len(lines))
	for _, f := range scanFences(lines) {
		end := f.close
		if end < 0 {
			end = len(lines) - 1
		}
		for i := f.open; i <= end; i++ {
			inFence[i] = true
		}
	}
	origs := make([]string, 0, len(links))
	for o := range links {
		origs = append(origs, o)
	}
	slices.SortFunc(origs, func(a, b string) int { return cmp.Or(cmp.Compare(len(b), len(a)), strings.Compare(a, b)) })
	for _, f := range scanFences(lines) {
		if lang, _, _ := strings.Cut(f.info, " "); lang != f.info {
			lines[f.open] = strings.Repeat(" ", f.indent) + strings.Repeat(string(f.char), f.count) + lang
		}
	}
	for i, line := range lines {
		if inFence[i] || !strings.Contains(line, "](") {
			continue
		}
		for _, o := range origs {
			line = strings.ReplaceAll(line, "]("+o+")", "]("+links[o]+")")
			line = strings.ReplaceAll(line, "]("+o+" ", "]("+links[o]+" ")
		}
		lines[i] = line
	}
	return strings.Join(lines, "\n")
}