// Package highlight renders syntax-highlighted file and diff HTML with chroma.
//
// Chroma runs in-process and is safe for concurrent use, so there is no
// worker pool.
package highlight

import (
	"bytes"
	"fmt"
	"html"
	"path/filepath"
	"strings"

	"github.com/alecthomas/chroma/v2"
	"github.com/alecthomas/chroma/v2/lexers"
	"github.com/gabriel-vasile/mimetype"

	"hearthforge/internal/util"
)

// BinaryDetectBytes is how much of a file is scanned for a NUL byte.
const BinaryDetectBytes = 8000

const (
	maxFileCache = 500
	maxDiffCache = 500
)

// classPrefix keeps chroma token classes out of the app's own class namespace.
const classPrefix = "ch-"

// Highlighter holds the bounded render caches. Create one at startup.
type Highlighter struct {
	inlineMaxBytes int64
	files          *util.Cache[string, FileView]
	diffs          *util.Cache[string, []RenderedDiffFile]
}

// New returns a Highlighter. inlineMaxBytes is config.InlineMaxBytes.
func New(inlineMaxBytes int64) *Highlighter {
	return &Highlighter{
		inlineMaxBytes: inlineMaxBytes,
		files:          util.NewCache[string, FileView](maxFileCache, 0),
		diffs:          util.NewCache[string, []RenderedDiffFile](maxDiffCache, 0),
	}
}

// HasBinaryContent reports whether the first BinaryDetectBytes contain a NUL.
// Git uses the same rule. Other control characters stay text.
func HasBinaryContent(content []byte) bool {
	if len(content) > BinaryDetectBytes {
		content = content[:BinaryDetectBytes]
	}
	return bytes.IndexByte(content, 0) >= 0
}

// DetectLang returns the chroma lexer name for a path, or "" when none matches.
// Chroma matches on filename globs, so Dockerfile and Makefile work too.
func DetectLang(path string) string {
	lexer := lexers.Match(filepath.Base(path))
	if lexer == nil {
		return ""
	}
	return lexer.Config().Name
}

// FileView describes how a blob should be shown.
// Type is one of "inline", "download", "binary" or "media".
type FileView struct {
	Type     string
	HTML     string
	Lines    int
	Size     int64
	MimeType string
}

// ServeFile classifies a blob and renders inline files as a line-numbered table.
// cacheKey may be empty to skip caching.
func (h *Highlighter) ServeFile(content []byte, filename, cacheKey string) FileView {
	if v, ok := h.files.Get(cacheKey); ok {
		return v
	}
	size := int64(len(content))

	mt := mimetype.Detect(content)
	switch strings.SplitN(mt.String(), "/", 2)[0] {
	case "image", "audio", "video":
		return FileView{Type: "media", MimeType: mt.String(), Size: size}
	}
	if HasBinaryContent(content) {
		return FileView{Type: "binary", Size: size}
	}
	if size > h.inlineMaxBytes {
		return FileView{Type: "download", Size: size}
	}

	text := string(content)
	view := FileView{
		Type:  "inline",
		HTML:  blobTable(text, DetectLang(filename)),
		Lines: strings.Count(text, "\n") + 1,
		Size:  size,
	}
	if cacheKey != "" {
		h.files.Set(cacheKey, view)
	}
	return view
}

// splitLines splits source text into display lines, dropping the trailing
// empty entry a final newline produces.
func splitLines(text string) []string {
	lines := strings.Split(text, "\n")
	if n := len(lines); n > 0 && lines[n-1] == "" {
		lines = lines[:n-1]
	}
	return lines
}

func blobTable(text, lang string) string {
	// One row per split entry, so a file ending in a newline gets a final
	// empty row. The previous highlighter did the same, so the line numbers
	// still match.
	// Chroma rewrites CRLF and a lone CR to LF before tokenising. Split the
	// same way, or a file with a bare CR gets shifted line numbers.
	src := strings.Split(strings.ReplaceAll(strings.ReplaceAll(text, "\r\n", "\n"), "\r", "\n"), "\n")
	rendered := highlightLines(text, lang, html.EscapeString)
	var b strings.Builder
	b.WriteString(`<table class="blob-table"><tbody>`)
	for i := range src {
		line := html.EscapeString(src[i])
		if i < len(rendered) {
			line = rendered[i]
		}
		n := i + 1
		fmt.Fprintf(&b, `<tr id="L%d"><td class="blob-ln"><a href="#L%d">%d</a></td><td class="blob-code">%s</td></tr>`, n, n, n, line)
	}
	b.WriteString("</tbody></table>")
	return b.String()
}

// tokenClass maps a token type to its CSS class, walking up to the parent type
// the way chroma's own HTML formatter does. It returns "" for unstyled tokens.
func tokenClass(t chroma.TokenType) string {
	for t != 0 {
		cls, ok := chroma.StandardTypes[t]
		if ok {
			if cls == "" {
				return ""
			}
			return classPrefix + cls
		}
		t = t.Parent()
	}
	return ""
}

// highlightLines tokenises code and returns one HTML fragment per source line.
// escape converts raw token text to HTML. It returns nil when lang is unknown
// or tokenising fails, and the caller falls back to plain escaped text.
func highlightLines(code, lang string, escape func(string) string) []string {
	if lang == "" {
		return nil
	}
	lexer := lexers.Get(lang)
	if lexer == nil {
		return nil
	}
	iter, err := chroma.Coalesce(lexer).Tokenise(nil, code)
	if err != nil {
		return nil
	}
	tokenLines := chroma.SplitTokensIntoLines(iter.Tokens())
	out := make([]string, len(tokenLines))
	for i, tokens := range tokenLines {
		var b strings.Builder
		for _, tok := range tokens {
			value := strings.TrimSuffix(tok.Value, "\n")
			if value == "" {
				continue
			}
			cls := tokenClass(tok.Type)
			if cls == "" {
				b.WriteString(escape(value))
				continue
			}
			fmt.Fprintf(&b, `<span class="%s">%s</span>`, cls, escape(value))
		}
		out[i] = b.String()
	}
	return out
}

// Code returns highlighted HTML for a code block body. lang is a chroma lexer
// name or alias, e.g. "js". Unknown languages come back HTML-escaped.
func Code(code, lang string) string {
	src := splitLines(code)
	rendered := highlightLines(code, lang, html.EscapeString)
	out := make([]string, len(src))
	for i := range src {
		out[i] = html.EscapeString(src[i])
		if i < len(rendered) {
			out[i] = rendered[i]
		}
	}
	result := strings.Join(out, "\n")
	if strings.HasSuffix(code, "\n") {
		result += "\n"
	}
	return result
}