Pad/internal/editor/line_index_fuzz_test.go

package editor

// Differential fuzzing for the incrementally-maintained LineIndex: every
// random edit is applied to the buffer AND replayed on the LineIndex with the
// exact production composition (UpdateLineIndexAfterDelete then
// UpdateLineIndexAfterInsert for a replace), and after every op the index must
// equal a full recomputation from the current content (lineStartOffsets
// oracle). LineIndex drift never corrupts the file (disk writes derive only
// from chunks), but it silently corrupts every line-based UI feature
// (scroll window, page keys, tap mapping, status bar), so it gets the same
// differential treatment as the buffer itself.

import (
	"bytes"
	"fmt"
	"math/rand"
	"strings"
	"testing"

	"pad/internal/io/pool/types"
)

// buildFuzzContent generates nLines pseudo-random lines (seeded) with a mix
// of ASCII, CJK and multi-byte content; noNewlines/crlf shape the line
// structure for the edge cases.
func buildFuzzContent(rng *rand.Rand, nLines int, noNewlines, crlf bool) string {
	var sb strings.Builder
	for i := 0; i < nLines; i++ {
		sb.WriteString(fuzzRuneString(rng, rng.Intn(40)+1))
		if i < nLines-1 || (crlf && i == nLines-1) {
			if !noNewlines {
				if crlf {
					sb.WriteString("\r\n")
				} else {
					sb.WriteByte('\n')
				}
			}
		}
	}
	if noNewlines {
		sb.WriteString(fuzzRuneString(rng, 20))
	}
	return sb.String()
}

func fuzzLineIndex(t *testing.T, label string, initial string, ops int) {
	t.Helper()
	chunkSize := 32
	rng := rand.New(rand.NewSource(int64(len(initial))*7919 + 13))
	cb := NewChunkedBuffer("/lineidx.txt", chunkSize, nil, "")
	var model []byte = []byte(initial)
	cb.SetContent(model)
	// Build the index the way the app does, then mutate it incrementally.
	cb.LineIndex = types.NewLineIndex(lineStartOffsets(model), 0, int64(len(model)))

	modelCap := 32 * 1024
	for i := 0; i < ops; i++ {
		if len(model) > modelCap {
			// Trim back (rune-aligned) to keep the oracle cheap.
			p := runeStartPositions(model)
			a, b := p[len(p)/4], p[len(p)/2]
			cb.Delete(a, b-a)
			cb.UpdateLineIndexAfterDelete(a, b)
			model = modelDelete(model, a, b)
		}

		// One random replace op (covers insert: b==a, and delete: text=="").
		p := runeStartPositions(model)
		a := p[rng.Intn(len(p))]
		b := p[rng.Intn(len(p))]
		if a > b {
			a, b = b, a
		}
		var text string
		switch rng.Intn(10) {
		case 0:
			text = "" // pure delete
		case 1:
			text = fuzzRuneString(rng, rng.Intn(30)+1) // no newlines
		default:
			// Text that adds structure: newlines / CRLF / CJK.
			text = fuzzRuneString(rng, rng.Intn(40)+1)
			switch rng.Intn(3) {
			case 0:
				text += "\n" + fuzzRuneString(rng, rng.Intn(10))
			case 1:
				text = "\n" + text
			case 2:
				text += "\r\n"
			}
		}

		// Production composition order (state.go HandleReplaceRange):
		if b > a {
			cb.Delete(a, b-a)
			model = modelDelete(model, a, b)
			cb.UpdateLineIndexAfterDelete(a, b)
		}
		if text != "" {
			cb.Insert(a, text)
			model = modelInsert(model, a, []byte(text))
			cb.UpdateLineIndexAfterInsert(a, text)
		}

		assertLineIndexMatchesOracle(t, cb, fmt.Sprintf("%s op %d", label, i))
	}
}

func TestLineIndex_Fuzz_RandomEdits(t *testing.T) {
	const ops = 3000
	cases := []struct {
		name    string
		content string
	}{
		{"mixedLines", buildFuzzContent(rand.New(rand.NewSource(1)), 80, false, false)},
		{"trailingNewline", buildFuzzContent(rand.New(rand.NewSource(2)), 60, false, false) + "\n"},
		{"noNewlines", fuzzRuneString(rand.New(rand.NewSource(3)), 800)},
		{"singleLine", "only one line, no newlines at all, fairly long"},
		{"crlfLines", buildFuzzContent(rand.New(rand.NewSource(4)), 60, false, true)},
		{"emptyStart", ""},
	}
	for _, c := range cases {
		t.Run(c.name, func(t *testing.T) {
			fuzzLineIndex(t, c.name, c.content, ops)
		})
	}
}

// TestLineIndex_Fuzz_TrailingEmptyLine pins the invariant that a file ending
// in '\n' has a trailing empty line in the index, across random edits that
// add/remove the final newline.
func TestLineIndex_Fuzz_TrailingEmptyLine(t *testing.T) {
	rng := rand.New(rand.NewSource(5))
	cb := NewChunkedBuffer("/trail.txt", 16, nil, "")
	model := []byte("a\nb\nc\n")
	cb.SetContent(model)
	cb.LineIndex = types.NewLineIndex(lineStartOffsets(model), 0, int64(len(model)))
	if cb.LineIndex.LineCount() != 4 {
		t.Fatalf("initial LineCount=%d, want 4 (trailing empty line)", cb.LineIndex.LineCount())
	}
	for i := 0; i < 1500; i++ {
		// Randomly add/remove the final newline and edit around the end.
		op := rng.Intn(4)
		switch op {
		case 0: // ensure ends with newline
			if len(model) > 0 && model[len(model)-1] != '\n' {
				pos := len(model)
				cb.Insert(pos, "\n")
				model = modelInsert(model, pos, []byte("\n"))
				cb.UpdateLineIndexAfterInsert(pos, "\n")
			}
		case 1: // strip trailing newlines
			oldLen := len(model)
			end := oldLen
			for end > 0 && model[end-1] == '\n' {
				end--
			}
			if end < oldLen {
				cb.Delete(end, oldLen-end)
				model = modelDelete(model, end, oldLen)
				cb.UpdateLineIndexAfterDelete(end, oldLen)
			}
		case 2: // append a line at the end
			pos := len(model)
			text := fuzzRuneString(rng, rng.Intn(10)+1) + "\n"
			cb.Insert(pos, text)
			model = modelInsert(model, pos, []byte(text))
			cb.UpdateLineIndexAfterInsert(pos, text)
		case 3: // delete the last line
			if last := bytesLastLineStart(model); last >= 0 {
				oldLen := len(model)
				cb.Delete(last, oldLen-last)
				model = modelDelete(model, last, oldLen)
				cb.UpdateLineIndexAfterDelete(last, oldLen)
			}
		}
		assertLineIndexMatchesOracle(t, cb, fmt.Sprintf("op %d", i))
		// Structural invariant: line count = newlines + 1, and a trailing '\n'
		// shows up as the index's last offset landing exactly at len(model)
		// (the trailing empty line).
		endsNL := len(model) > 0 && model[len(model)-1] == '\n'
		wantLines := 1 + utf8RuneCountNewlines(model)
		if cb.LineIndex.LineCount() != wantLines {
			t.Fatalf("op %d: LineCount=%d, want %d", i, cb.LineIndex.LineCount(), wantLines)
		}
		offs := cb.LineIndex.Offsets
		lastIsEnd := offs[len(offs)-1] == int32(len(model))
		// The final line starts at EOF iff the file is empty or ends in '\n'
		// (i.e. the final line is empty).
		wantLastIsEnd := len(model) == 0 || endsNL
		if lastIsEnd != wantLastIsEnd {
			t.Fatalf("op %d: last line start at %d, len %d, endsNL=%v", i, offs[len(offs)-1], len(model), endsNL)
		}
	}
}

// bytesLastLineStart returns the byte offset of the start of the last line in
// m (the position after the previous '\n', or 0), or -1 if m is empty.
func bytesLastLineStart(m []byte) int {
	if len(m) == 0 {
		return -1
	}
	if i := bytes.LastIndexByte(m, '\n'); i >= 0 {
		return i + 1
	}
	return 0
}

// utf8RuneCountNewlines counts '\n' bytes (byte scan; '\n' cannot appear
// inside a multi-byte UTF-8 sequence).
func utf8RuneCountNewlines(m []byte) int {
	n := 0
	for _, b := range m {
		if b == '\n' {
			n++
		}
	}
	return n
}