patchwork

package doc

import (
	"crypto/sha256"
	"encoding/hex"
	"strings"
)

// Block kinds.
const (
	KindParagraph = "paragraph"
	KindHeading   = "heading"
	KindCode      = "code"
	KindListItem  = "list_item"
	KindQuote     = "quote"
	KindHTML      = "html"
	KindTable     = "table"
	KindBreak     = "break"
)

// Block is the unit of the block-level diff: a paragraph, a heading, one
// list item with whatever is nested under it, a whole code block, quote or
// table.
type Block struct {
	Kind   string `json:"kind"`
	Source string `json:"source"`
	Text   string `json:"text"`
	Hash   string `json:"-"`
}

// SplitBlocks cuts Markdown into blocks with a line scanner instead of a full
// CommonMark parse. A diff needs stable block boundaries, not a render tree,
// and a parser's tree loses the exact source span of fences and thematic
// breaks. The rules follow CommonMark where it matters for boundaries.
func SplitBlocks(body string) []Block {
	lines := strings.Split(body, "\n")
	n := len(lines)
	var blocks []Block
	emit := func(kind string, start, end int) {
		for end > start && isBlank(lines[end-1]) {
			end--
		}
		src := strings.Join(lines[start:end], "\n")
		blocks = append(blocks, Block{Kind: kind, Source: src, Text: plainText(kind, src), Hash: blockHash(src)})
	}

	for i := 0; i < n; {
		l := lines[i]
		if isBlank(l) {
			i++
			continue
		}
		start := i
		switch {
		case fenceMarker(l) != "":
			f := fenceMarker(l)
			for i++; i < n && !closesFence(lines[i], f); i++ {
			}
			if i < n {
				i++
			}
			emit(KindCode, start, i)
		case isATXHeading(l):
			i++
			emit(KindHeading, start, i)
		case isThematicBreak(l):
			i++
			emit(KindBreak, start, i)
		case isQuote(l):
			for i++; i < n && !isBlank(lines[i]) && (isQuote(lines[i]) || !interrupts(lines[i])); i++ {
			}
			emit(KindQuote, start, i)
		case isListItem(l):
			markerIndent, contentIndent, _ := listMarker(l)
			i++
			for i < n {
				if isBlank(lines[i]) {
					j := i
					for j < n && isBlank(lines[j]) {
						j++
					}
					if j < n && indentOf(lines[j]) >= contentIndent {
						i = j
						continue
					}
					break
				}
				if mi, _, ok := listMarker(lines[i]); ok && mi <= markerIndent && !isThematicBreak(lines[i]) {
					break
				}
				if indentOf(lines[i]) < contentIndent && interruptsList(lines[i]) {
					break
				}
				i++
			}
			emit(KindListItem, start, i)
		case isHTMLStart(l):
			for i++; i < n && !isBlank(lines[i]); i++ {
			}
			emit(KindHTML, start, i)
		case i+1 < n && strings.Contains(l, "|") && isDelimiterRow(lines[i+1]):
			for i += 2; i < n && !isBlank(lines[i]) && strings.Contains(lines[i], "|"); i++ {
			}
			emit(KindTable, start, i)
		case indentOf(l) >= 4:
			for i++; i < n; {
				if isBlank(lines[i]) {
					j := i
					for j < n && isBlank(lines[j]) {
						j++
					}
					if j < n && indentOf(lines[j]) >= 4 {
						i = j
						continue
					}
					break
				}
				if indentOf(lines[i]) < 4 {
					break
				}
				i++
			}
			emit(KindCode, start, i)
		default:
			kind := KindParagraph
			for i++; i < n && !isBlank(lines[i]); i++ {
				if isSetextUnderline(lines[i]) {
					i++
					kind = KindHeading
					break
				}
				if interrupts(lines[i]) {
					break
				}
			}
			emit(kind, start, i)
		}
	}
	return blocks
}

func blockHash(src string) string {
	sum := sha256.Sum256([]byte(strings.Join(strings.Fields(src), " ")))
	return hex.EncodeToString(sum[:8])
}

func isBlank(l string) bool { return strings.TrimSpace(l) == "" }

// indentOf counts leading columns, with tabs advancing to the next stop of 4.
func indentOf(l string) int {
	col := 0
	for _, c := range l {
		switch c {
		case ' ':
			col++
		case '\t':
			col += 4 - col%4
		default:
			return col
		}
	}
	return col
}

func fenceMarker(l string) string {
	if indentOf(l) > 3 {
		return ""
	}
	t := strings.TrimLeft(l, " \t")
	if len(t) < 3 || (t[0] != '`' && t[0] != '~') {
		return ""
	}
	n := 0
	for n < len(t) && t[n] == t[0] {
		n++
	}
	if n < 3 || (t[0] == '`' && strings.Contains(t[n:], "`")) {
		return ""
	}
	return t[:n]
}

func closesFence(l, open string) bool {
	if indentOf(l) > 3 {
		return false
	}
	t := strings.TrimLeft(l, " \t")
	n := 0
	for n < len(t) && t[n] == open[0] {
		n++
	}
	return n >= len(open) && strings.TrimSpace(t[n:]) == ""
}

func isATXHeading(l string) bool {
	if indentOf(l) > 3 {
		return false
	}
	t := strings.TrimLeft(l, " \t")
	n := 0
	for n < len(t) && t[n] == '#' {
		n++
	}
	return n >= 1 && n <= 6 && (n == len(t) || t[n] == ' ' || t[n] == '\t')
}

func isThematicBreak(l string) bool {
	if indentOf(l) > 3 {
		return false
	}
	t := strings.TrimSpace(l)
	if t == "" || !strings.ContainsRune("-*_", rune(t[0])) {
		return false
	}
	count := 0
	for _, c := range t {
		switch {
		case c == rune(t[0]):
			count++
		case c == ' ' || c == '\t':
		default:
			return false
		}
	}
	return count >= 3
}

func isSetextUnderline(l string) bool {
	if indentOf(l) > 3 {
		return false
	}
	t := strings.TrimSpace(l)
	return t != "" && (strings.Trim(t, "=") == "" || strings.Trim(t, "-") == "")
}

func isQuote(l string) bool {
	return indentOf(l) <= 3 && strings.HasPrefix(strings.TrimLeft(l, " \t"), ">")
}

func isHTMLStart(l string) bool {
	if indentOf(l) > 3 {
		return false
	}
	t := strings.TrimLeft(l, " \t")
	if len(t) < 2 || t[0] != '<' {
		return false
	}
	c := t[1]
	return c == '/' || c == '!' || c == '?' || (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z')
}

func isDelimiterRow(l string) bool {
	t := strings.TrimSpace(l)
	t = strings.TrimPrefix(t, "|")
	t = strings.TrimSuffix(t, "|")
	if t == "" {
		return false
	}
	for _, cell := range strings.Split(t, "|") {
		c := strings.TrimSpace(cell)
		c = strings.TrimPrefix(c, ":")
		c = strings.TrimSuffix(c, ":")
		if c == "" || strings.Trim(c, "-") != "" {
			return false
		}
	}
	return true
}

// listMarker reports a list item start: "- ", "* ", "+ ", "1. " or "1) ".
// contentIndent is the column where the item's text begins, which nested
// content must reach to stay inside the item.
func listMarker(l string) (markerIndent, contentIndent int, ok bool) {
	markerIndent = indentOf(l)
	if markerIndent > 3 {
		return 0, 0, false
	}
	t := strings.TrimLeft(l, " \t")
	m := 0
	switch {
	case t != "" && strings.ContainsRune("-*+", rune(t[0])):
		m = 1
	default:
		for m < len(t) && m < 9 && t[m] >= '0' && t[m] <= '9' {
			m++
		}
		if m == 0 || m >= len(t) || (t[m] != '.' && t[m] != ')') {
			return 0, 0, false
		}
		m++
	}
	if m < len(t) && t[m] != ' ' && t[m] != '\t' {
		return 0, 0, false
	}
	spaces := 0
	for m+spaces < len(t) && t[m+spaces] == ' ' {
		spaces++
	}
	if spaces == 0 || spaces > 4 || m+spaces == len(t) {
		spaces = 1
	}
	return markerIndent, markerIndent + m + spaces, true
}

func isListItem(l string) bool {
	_, _, ok := listMarker(l)
	return ok && !isThematicBreak(l)
}

// interrupts reports whether a line ends a paragraph without a blank line.
// An ordered item interrupts only when it starts at 1, as in CommonMark, so
// "2026. was a long year" stays inside its paragraph.
func interrupts(l string) bool {
	if isATXHeading(l) || fenceMarker(l) != "" || isThematicBreak(l) || isQuote(l) || isHTMLStart(l) {
		return true
	}
	if !isListItem(l) {
		return false
	}
	t := strings.TrimLeft(l, " \t")
	if strings.ContainsRune("-*+", rune(t[0])) {
		return true
	}
	return strings.HasPrefix(t, "1.") || strings.HasPrefix(t, "1)")
}

func interruptsList(l string) bool {
	return isATXHeading(l) || fenceMarker(l) != "" || isThematicBreak(l) || isQuote(l) || isHTMLStart(l)
}