internal/diffp: add patience diff from main repo

This is the main repo's internal/diff, renamed because
there is a different internal/diff already in x/tools.

Change-Id: I7b6da718e46c9fa23931908b520b9f39c178206b
Reviewed-on: https://go-review.googlesource.com/c/tools/+/491915
Run-TryBot: Russ Cox <rsc@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
gopls-CI: kokoro <noreply+kokoro@google.com>
Reviewed-by: David Chase <drchase@google.com>

internal/diffp/diff.go

@@ -0,0 +1,264 @@
+// Copyright 2022 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package diffp implements a basic diff algorithm equivalent to patience diff.
+// It is a copy of internal/diff from the main Go repo, renamed to diffp to avoid
+// conflict with the existing golang.org/x/tools/internal/diff.
+package diffp
+
+import (
+	"bytes"
+	"fmt"
+	"sort"
+	"strings"
+)
+
+// A pair is a pair of values tracked for both the x and y side of a diff.
+// It is typically a pair of line indexes.
+type pair struct{ x, y int }
+
+// Diff returns an anchored diff of the two texts old and new
+// in the “unified diff” format. If old and new are identical,
+// Diff returns a nil slice (no output).
+//
+// Unix diff implementations typically look for a diff with
+// the smallest number of lines inserted and removed,
+// which can in the worst case take time quadratic in the
+// number of lines in the texts. As a result, many implementations
+// either can be made to run for a long time or cut off the search
+// after a predetermined amount of work.
+//
+// In contrast, this implementation looks for a diff with the
+// smallest number of “unique” lines inserted and removed,
+// where unique means a line that appears just once in both old and new.
+// We call this an “anchored diff” because the unique lines anchor
+// the chosen matching regions. An anchored diff is usually clearer
+// than a standard diff, because the algorithm does not try to
+// reuse unrelated blank lines or closing braces.
+// The algorithm also guarantees to run in O(n log n) time
+// instead of the standard O(n²) time.
+//
+// Some systems call this approach a “patience diff,” named for
+// the “patience sorting” algorithm, itself named for a solitaire card game.
+// We avoid that name for two reasons. First, the name has been used
+// for a few different variants of the algorithm, so it is imprecise.
+// Second, the name is frequently interpreted as meaning that you have
+// to wait longer (to be patient) for the diff, meaning that it is a slower algorithm,
+// when in fact the algorithm is faster than the standard one.
+func Diff(oldName string, old []byte, newName string, new []byte) []byte {
+	if bytes.Equal(old, new) {
+		return nil
+	}
+	x := lines(old)
+	y := lines(new)
+
+	// Print diff header.
+	var out bytes.Buffer
+	fmt.Fprintf(&out, "diff %s %s\n", oldName, newName)
+	fmt.Fprintf(&out, "--- %s\n", oldName)
+	fmt.Fprintf(&out, "+++ %s\n", newName)
+
+	// Loop over matches to consider,
+	// expanding each match to include surrounding lines,
+	// and then printing diff chunks.
+	// To avoid setup/teardown cases outside the loop,
+	// tgs returns a leading {0,0} and trailing {len(x), len(y)} pair
+	// in the sequence of matches.
+	var (
+		done  pair     // printed up to x[:done.x] and y[:done.y]
+		chunk pair     // start lines of current chunk
+		count pair     // number of lines from each side in current chunk
+		ctext []string // lines for current chunk
+	)
+	for _, m := range tgs(x, y) {
+		if m.x < done.x {
+			// Already handled scanning forward from earlier match.
+			continue
+		}
+
+		// Expand matching lines as far possible,
+		// establishing that x[start.x:end.x] == y[start.y:end.y].
+		// Note that on the first (or last) iteration we may (or definitey do)
+		// have an empty match: start.x==end.x and start.y==end.y.
+		start := m
+		for start.x > done.x && start.y > done.y && x[start.x-1] == y[start.y-1] {
+			start.x--
+			start.y--
+		}
+		end := m
+		for end.x < len(x) && end.y < len(y) && x[end.x] == y[end.y] {
+			end.x++
+			end.y++
+		}
+
+		// Emit the mismatched lines before start into this chunk.
+		// (No effect on first sentinel iteration, when start = {0,0}.)
+		for _, s := range x[done.x:start.x] {
+			ctext = append(ctext, "-"+s)
+			count.x++
+		}
+		for _, s := range y[done.y:start.y] {
+			ctext = append(ctext, "+"+s)
+			count.y++
+		}
+
+		// If we're not at EOF and have too few common lines,
+		// the chunk includes all the common lines and continues.
+		const C = 3 // number of context lines
+		if (end.x < len(x) || end.y < len(y)) &&
+			(end.x-start.x < C || (len(ctext) > 0 && end.x-start.x < 2*C)) {
+			for _, s := range x[start.x:end.x] {
+				ctext = append(ctext, " "+s)
+				count.x++
+				count.y++
+			}
+			done = end
+			continue
+		}
+
+		// End chunk with common lines for context.
+		if len(ctext) > 0 {
+			n := end.x - start.x
+			if n > C {
+				n = C
+			}
+			for _, s := range x[start.x : start.x+n] {
+				ctext = append(ctext, " "+s)
+				count.x++
+				count.y++
+			}
+			done = pair{start.x + n, start.y + n}
+
+			// Format and emit chunk.
+			// Convert line numbers to 1-indexed.
+			// Special case: empty file shows up as 0,0 not 1,0.
+			if count.x > 0 {
+				chunk.x++
+			}
+			if count.y > 0 {
+				chunk.y++
+			}
+			fmt.Fprintf(&out, "@@ -%d,%d +%d,%d @@\n", chunk.x, count.x, chunk.y, count.y)
+			for _, s := range ctext {
+				out.WriteString(s)
+			}
+			count.x = 0
+			count.y = 0
+			ctext = ctext[:0]
+		}
+
+		// If we reached EOF, we're done.
+		if end.x >= len(x) && end.y >= len(y) {
+			break
+		}
+
+		// Otherwise start a new chunk.
+		chunk = pair{end.x - C, end.y - C}
+		for _, s := range x[chunk.x:end.x] {
+			ctext = append(ctext, " "+s)
+			count.x++
+			count.y++
+		}
+		done = end
+	}
+
+	return out.Bytes()
+}
+
+// lines returns the lines in the file x, including newlines.
+// If the file does not end in a newline, one is supplied
+// along with a warning about the missing newline.
+func lines(x []byte) []string {
+	l := strings.SplitAfter(string(x), "\n")
+	if l[len(l)-1] == "" {
+		l = l[:len(l)-1]
+	} else {
+		// Treat last line as having a message about the missing newline attached,
+		// using the same text as BSD/GNU diff (including the leading backslash).
+		l[len(l)-1] += "\n\\ No newline at end of file\n"
+	}
+	return l
+}
+
+// tgs returns the pairs of indexes of the longest common subsequence
+// of unique lines in x and y, where a unique line is one that appears
+// once in x and once in y.
+//
+// The longest common subsequence algorithm is as described in
+// Thomas G. Szymanski, “A Special Case of the Maximal Common
+// Subsequence Problem,” Princeton TR #170 (January 1975),
+// available at https://research.swtch.com/tgs170.pdf.
+func tgs(x, y []string) []pair {
+	// Count the number of times each string appears in a and b.
+	// We only care about 0, 1, many, counted as 0, -1, -2
+	// for the x side and 0, -4, -8 for the y side.
+	// Using negative numbers now lets us distinguish positive line numbers later.
+	m := make(map[string]int)
+	for _, s := range x {
+		if c := m[s]; c > -2 {
+			m[s] = c - 1
+		}
+	}
+	for _, s := range y {
+		if c := m[s]; c > -8 {
+			m[s] = c - 4
+		}
+	}
+
+	// Now unique strings can be identified by m[s] = -1+-4.
+	//
+	// Gather the indexes of those strings in x and y, building:
+	//	xi[i] = increasing indexes of unique strings in x.
+	//	yi[i] = increasing indexes of unique strings in y.
+	//	inv[i] = index j such that x[xi[i]] = y[yi[j]].
+	var xi, yi, inv []int
+	for i, s := range y {
+		if m[s] == -1+-4 {
+			m[s] = len(yi)
+			yi = append(yi, i)
+		}
+	}
+	for i, s := range x {
+		if j, ok := m[s]; ok && j >= 0 {
+			xi = append(xi, i)
+			inv = append(inv, j)
+		}
+	}
+
+	// Apply Algorithm A from Szymanski's paper.
+	// In those terms, A = J = inv and B = [0, n).
+	// We add sentinel pairs {0,0}, and {len(x),len(y)}
+	// to the returned sequence, to help the processing loop.
+	J := inv
+	n := len(xi)
+	T := make([]int, n)
+	L := make([]int, n)
+	for i := range T {
+		T[i] = n + 1
+	}
+	for i := 0; i < n; i++ {
+		k := sort.Search(n, func(k int) bool {
+			return T[k] >= J[i]
+		})
+		T[k] = J[i]
+		L[i] = k + 1
+	}
+	k := 0
+	for _, v := range L {
+		if k < v {
+			k = v
+		}
+	}
+	seq := make([]pair, 2+k)
+	seq[1+k] = pair{len(x), len(y)} // sentinel at end
+	lastj := n
+	for i := n - 1; i >= 0; i-- {
+		if L[i] == k && J[i] < lastj {
+			seq[k] = pair{xi[i], yi[J[i]]}
+			k--
+		}
+	}
+	seq[0] = pair{0, 0} // sentinel at start
+	return seq
+}

internal/diffp/diff_test.go

@@ -0,0 +1,44 @@
+// Copyright 2022 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package diffp
+
+import (
+	"bytes"
+	"path/filepath"
+	"testing"
+
+	"golang.org/x/tools/txtar"
+)
+
+func clean(text []byte) []byte {
+	text = bytes.ReplaceAll(text, []byte("$\n"), []byte("\n"))
+	text = bytes.TrimSuffix(text, []byte("^D\n"))
+	return text
+}
+
+func Test(t *testing.T) {
+	files, _ := filepath.Glob("testdata/*.txt")
+	if len(files) == 0 {
+		t.Fatalf("no testdata")
+	}
+
+	for _, file := range files {
+		t.Run(filepath.Base(file), func(t *testing.T) {
+			a, err := txtar.ParseFile(file)
+			if err != nil {
+				t.Fatal(err)
+			}
+			if len(a.Files) != 3 || a.Files[2].Name != "diff" {
+				t.Fatalf("%s: want three files, third named \"diff\"", file)
+			}
+			diffs := Diff(a.Files[0].Name, clean(a.Files[0].Data), a.Files[1].Name, clean(a.Files[1].Data))
+			want := clean(a.Files[2].Data)
+			if !bytes.Equal(diffs, want) {
+				t.Fatalf("%s: have:\n%s\nwant:\n%s\n%s", file,
+					diffs, want, Diff("have", diffs, "want", want))
+			}
+		})
+	}
+}

internal/diffp/testdata/allnew.txt

@@ -0,0 +1,13 @@
+-- old --
+-- new --
+a
+b
+c
+-- diff --
+diff old new
+--- old
++++ new
+@@ -0,0 +1,3 @@
++a
++b
++c

internal/diffp/testdata/allold.txt

@@ -0,0 +1,13 @@
+-- old --
+a
+b
+c
+-- new --
+-- diff --
+diff old new
+--- old
++++ new
+@@ -1,3 +0,0 @@
+-a
+-b
+-c

internal/diffp/testdata/basic.txt

@@ -0,0 +1,35 @@
+Example from Hunt and McIlroy, “An Algorithm for Differential File Comparison.”
+https://www.cs.dartmouth.edu/~doug/diff.pdf
+
+-- old --
+a
+b
+c
+d
+e
+f
+g
+-- new --
+w
+a
+b
+x
+y
+z
+e
+-- diff --
+diff old new
+--- old
++++ new
+@@ -1,7 +1,7 @@
++w
+ a
+ b
+-c
+-d
++x
++y
++z
+ e
+-f
+-g

internal/diffp/testdata/dups.txt

@@ -0,0 +1,40 @@
+-- old --
+a
+
+b
+
+c
+
+d
+
+e
+
+f
+-- new --
+a
+
+B
+
+C
+
+d
+
+e
+
+f
+-- diff --
+diff old new
+--- old
++++ new
+@@ -1,8 +1,8 @@
+ a
+ $
+-b
+-
+-c
++B
++
++C
+ $
+ d
+ $

internal/diffp/testdata/end.txt

@@ -0,0 +1,38 @@
+-- old --
+1
+2
+3
+4
+5
+6
+7
+eight
+nine
+ten
+eleven
+-- new --
+1
+2
+3
+4
+5
+6
+7
+8
+9
+10
+-- diff --
+diff old new
+--- old
++++ new
+@@ -5,7 +5,6 @@
+ 5
+ 6
+ 7
+-eight
+-nine
+-ten
+-eleven
++8
++9
++10

internal/diffp/testdata/eof.txt

@@ -0,0 +1,9 @@
+-- old --
+a
+b
+c^D
+-- new --
+a
+b
+c^D
+-- diff --

internal/diffp/testdata/eof1.txt

@@ -0,0 +1,18 @@
+-- old --
+a
+b
+c
+-- new --
+a
+b
+c^D
+-- diff --
+diff old new
+--- old
++++ new
+@@ -1,3 +1,3 @@
+ a
+ b
+-c
++c
+\ No newline at end of file

internal/diffp/testdata/eof2.txt

@@ -0,0 +1,18 @@
+-- old --
+a
+b
+c^D
+-- new --
+a
+b
+c
+-- diff --
+diff old new
+--- old
++++ new
+@@ -1,3 +1,3 @@
+ a
+ b
+-c
+\ No newline at end of file
++c

internal/diffp/testdata/long.txt

@@ -0,0 +1,62 @@
+-- old --
+1
+2
+3
+4
+5
+6
+7
+8
+9
+10
+11
+12
+13
+14
+14½
+15
+16
+17
+18
+19
+20
+-- new --
+1
+2
+3
+4
+5
+6
+8
+9
+10
+11
+12
+13
+14
+17
+18
+19
+20
+-- diff --
+diff old new
+--- old
++++ new
+@@ -4,7 +4,6 @@
+ 4
+ 5
+ 6
+-7
+ 8
+ 9
+ 10
+@@ -12,9 +11,6 @@
+ 12
+ 13
+ 14
+-14½
+-15
+-16
+ 17
+ 18
+ 19

internal/diffp/testdata/same.txt

@@ -0,0 +1,5 @@
+-- old --
+hello world
+-- new --
+hello world
+-- diff --

internal/diffp/testdata/start.txt

@@ -0,0 +1,34 @@
+-- old --
+e
+pi
+4
+5
+6
+7
+8
+9
+10
+-- new --
+1
+2
+3
+4
+5
+6
+7
+8
+9
+10
+-- diff --
+diff old new
+--- old
++++ new
+@@ -1,5 +1,6 @@
+-e
+-pi
++1
++2
++3
+ 4
+ 5
+ 6

internal/diffp/testdata/triv.txt

@@ -0,0 +1,40 @@
+Another example from Hunt and McIlroy,
+“An Algorithm for Differential File Comparison.”
+https://www.cs.dartmouth.edu/~doug/diff.pdf
+
+Anchored diff gives up on finding anything,
+since there are no unique lines.
+
+-- old --
+a
+b
+c
+a
+b
+b
+a
+-- new --
+c
+a
+b
+a
+b
+c
+-- diff --
+diff old new
+--- old
++++ new
+@@ -1,7 +1,6 @@
+-a
+-b
+-c
+-a
+-b
+-b
+-a
++c
++a
++b
++a
++b
++c