llmgw/internal/web/modelcard.go

package web

import (
	"bytes"
	"html/template"
	"net/url"
	"strings"

	"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/renderer/html"
	"github.com/yuin/goldmark/text"
	"github.com/yuin/goldmark/util"
)

// renderCard turns a model card into HTML for the mirror page.
//
// The card is a stranger's text, so what it may do is narrow:
//   - raw HTML is dropped, which goldmark does unless told otherwise, and
//     links to javascript: and the like are emptied, which it also does;
//   - images are not loaded but shown as links, because an admin page that
//     fetches whatever a README points at tells a stranger who is looking;
//   - relative links, to the repository's own LICENSE and files, are made to
//     point at the repository rather than at this page.
func renderCard(card, repoBase string) template.HTML {
	return safeMarkdown(card, repoBase)
}

// safeMarkdown renders a stranger's markdown under the rules above. base, when
// set, is what relative links resolve against.
func safeMarkdown(src, base string) template.HTML {
	if strings.TrimSpace(src) == "" {
		return ""
	}
	md := goldmark.New(
		goldmark.WithExtensions(extension.GFM),
		goldmark.WithParserOptions(parser.WithASTTransformers(
			util.Prioritized(cardLinks{base: base}, 100))),
		goldmark.WithRendererOptions(
			renderer.WithNodeRenderers(util.Prioritized(imageAsLink{}, 100))),
	)
	var out bytes.Buffer
	if err := md.Convert([]byte(src), &out); err != nil {
		return ""
	}
	// Safe by construction: goldmark in its default, safe mode, with the two
	// changes above. Not something a template may escape again.
	return template.HTML(out.String()) //nolint:gosec
}

// cardLinks resolves a card's relative links against its repository.
type cardLinks struct{ base string }

func (c cardLinks) Transform(doc *ast.Document, _ text.Reader, _ parser.Context) {
	base, err := url.Parse(c.base)
	if err != nil || c.base == "" {
		return
	}
	_ = ast.Walk(doc, func(n ast.Node, entering bool) (ast.WalkStatus, error) {
		if !entering {
			return ast.WalkContinue, nil
		}
		var dest *[]byte
		switch l := n.(type) {
		case *ast.Link:
			dest = &l.Destination
		case *ast.Image:
			dest = &l.Destination
		default:
			return ast.WalkContinue, nil
		}
		u, err := url.Parse(string(*dest))
		if err != nil || u.IsAbs() || strings.HasPrefix(string(*dest), "#") {
			return ast.WalkContinue, nil
		}
		*dest = []byte(base.ResolveReference(u).String())
		return ast.WalkContinue, nil
	})
}

// imageAsLink renders an image as a link to it, named by its alt text.
type imageAsLink struct{}

func (imageAsLink) RegisterFuncs(reg renderer.NodeRendererFuncRegisterer) {
	reg.Register(ast.KindImage, func(w util.BufWriter, src []byte, n ast.Node, entering bool) (ast.WalkStatus, error) {
		if !entering {
			return ast.WalkContinue, nil
		}
		img := n.(*ast.Image)
		alt := strings.TrimSpace(string(nodeText(img, src)))
		if alt == "" {
			alt = "image"
		}
		dest := img.Destination
		if html.IsDangerousURL(dest) {
			_, _ = w.WriteString(`<span class="img">[` + template.HTMLEscapeString(alt) + `]</span>`)
			return ast.WalkSkipChildren, nil
		}
		_, _ = w.WriteString(`<a class="img" rel="noopener noreferrer" href="` +
			template.HTMLEscapeString(string(util.URLEscape(dest, true))) + `">[` +
			template.HTMLEscapeString(alt) + `]</a>`)
		return ast.WalkSkipChildren, nil
	})
}

// nodeText is the plain text beneath a node.
func nodeText(n ast.Node, src []byte) []byte {
	var b bytes.Buffer
	for c := n.FirstChild(); c != nil; c = c.NextSibling() {
		if t, ok := c.(*ast.Text); ok {
			b.Write(t.Segment.Value(src))
			continue
		}
		b.Write(nodeText(c, src))
	}
	return b.Bytes()
}