* ZPAQ: real journaling/streaming format, zpaq 7.15 reads what we write

.github/workflows/ci.yml

@@ -174,6 +174,9 @@ jobs:
           # reiser4progs owns Reiser4: fsck.reiser4 and debugfs.reiser4 judge every volume the
           # Reiser4 writer creates or edits, in the gate below.
           sudo apt-get install -y reiser4progs
+          # zpaq (7.15) owns ZPAQ: it must list and extract what the ZPAQ writer produces, and the
+          # reader must decode every method zpaq writes, in the gate below.
+          sudo apt-get install -y zpaq
           sudo apt-get install -y "linux-modules-extra-$(uname -r)" || true
           sudo modprobe gfs2 || true
 
@@ -235,6 +238,19 @@ jobs:
           --filter "FullyQualifiedName~Compression.Tests.Reiser4.Reiser4ProfileExternalTests|FullyQualifiedName~Compression.Tests.Reiser4.Reiser4MutationExternalTests|FullyQualifiedName~Compression.Tests.Reiser4.Reiser4NativeTreeExternalTests"
           --logger "trx;LogFileName=reiser4-oracle-tests.trx"
 
+      # Hard gate, both directions: zpaq must list and extract byte for byte what the ZPAQ writer
+      # produces (every method), the reader must decode what zpaq writes (every method, versions,
+      # deduplication, streaming), and our compressed data blocks must equal zpaq 7.15's.
+      - name: ZPAQ oracle
+        if: matrix.os == 'ubuntu-latest'
+        env:
+          CWB_REQUIRE_ZPAQ: "1"
+        run: >
+          dotnet test Compression.Tests
+          --configuration Release --no-build
+          --filter "FullyQualifiedName~Compression.Tests.Zpaq.ZpaqExternalOracleTests"
+          --logger "trx;LogFileName=zpaq-oracle-tests.trx"
+
       # bcachefs-tools' own checker judges every volume this package writes or
       # edits: the allocation information, the usage totals and the namespace are
       # cross-examined against each other, and bucket generations against the

Compression.Tests/BuildingBlocks/ZpaqBuildingBlockTests.cs

@@ -33,7 +33,7 @@ public void RepeatedSentence_RoundTripsAndCompresses() {
 
   [Test, Category("HappyPath"), Category("RoundTrip")]
   public void HighlyRepetitive_RoundTripsAndCompresses() {
-    var data = new byte[256];
+    var data = new byte[4096];  // large enough that the ~120-byte ZPAQ block header does not dominate
     Array.Fill(data, (byte)0x61);
     var compressed = Bb.Compress(data);
     var round = Bb.Decompress(compressed);

Compression.Tests/Compression.Tests.csproj

@@ -91,6 +91,10 @@
     <EmbeddedResource Include="FreeArc\ReferenceVectors\*.arc">
       <LogicalName>FreeArcVectors.%(Filename)%(Extension)</LogicalName>
     </EmbeddedResource>
+    <!-- Archives the reference zpaq 7.15 wrote, with a manifest of their contents; see Zpaq/ReferenceVectors/README.md. -->
+    <EmbeddedResource Include="Zpaq\ReferenceVectors\*.zpaq;Zpaq\ReferenceVectors\manifest.tsv">
+      <LogicalName>ZpaqVectors.%(Filename)%(Extension)</LogicalName>
+    </EmbeddedResource>
     <!-- CVF cluster payloads the dmsdos compressors wrote; see Cvf/ReferenceVectors/README.md. -->
     <EmbeddedResource Include="Cvf\ReferenceVectors\*.bin">
       <LogicalName>CvfVectors.%(Filename)%(Extension)</LogicalName>

Compression.Tests/Zpaq/ReferenceVectors/README.md

@@ -0,0 +1,19 @@
+# zpaq 7.15 reference archives
+
+Every `.zpaq` file here was written by the reference archiver **zpaq v7.15** (Matt Mahoney,
+public domain), the Windows x64 build `zpaq64.exe` shipped in
+<http://mattmahoney.net/dc/zpaq715.zip> ("compiled Aug 17 2016"), on 2026-10-19.
+
+`mkfixtures.py` (documentation, not a build input) generated the input files deterministically
+and ran `zpaq64.exe a <archive> <dir> [-method M]` in the work directory, so the stored names are
+relative (`set/text.txt`). Only the archives and `manifest.tsv` are checked in: the manifest gives,
+per archive and version, every name with its size and SHA-256 (size `-1` marks a directory).
+
+| File | Layout | How it was made | What it exercises |
+|------|--------|-----------------|-------------------|
+| `set.m0.zpaq` .. `set.m5.zpaq` | journaling | `a set.mN.zpaq set -method N` | every predefined method: 0 (dedup only), 1/2 (LZ77, packed codes, hash and suffix-array match finders), 3 (BWT or LZ77+CM), 4/5 (CM with ICM/ISSE chains, match model, mixers, SSE); E8E9 on `x86.bin`; `dup.txt` deduplicated against `text.txt`; an empty file; a subdirectory |
+| `versions.zpaq` | journaling | three `a versions.zpaq v` runs: add `notes.txt`+`a.bin`; change `a.bin`, add `b.txt`; delete `notes.txt` | three transactions, an update, a deletion record, fragments shared across versions |
+| `stream.zpaq` | streaming | `a stream.zpaq st -method s4.3ci1` | streaming layout (one block per file, filename and date in the segment header), BWT + order-0 ICM + order-1 ISSE |
+
+`zpaq64.exe l versions.zpaq -all` reports `3 versions, 6 files, 4 fragments`; `zpaq64.exe l
+stream.zpaq` reports two files of unknown size (the streaming layout records none).

Compression.Tests/Zpaq/ReferenceVectors/manifest.tsv

@@ -0,0 +1,52 @@
+# archive	version	name	size	sha256 (size -1 = directory)
+set.m0.zpaq	1	set/dup.txt	20000	6d7dcfff654e4f0b73c9c6748e20101a654d91340e808028d9cce43314e176b6
+set.m0.zpaq	1	set/empty.dat	0	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
+set.m0.zpaq	1	set/random.bin	4096	3a370c921e1d689daae1f95c1bc8fd70c3a2294d9a602603866c88f1678133ed
+set.m0.zpaq	1	set/sub/	-1	-
+set.m0.zpaq	1	set/sub/table.bin	16384	c30bfd6c26ff4d6994184a48cfe912d634d84741a506a379a756a88d03ce17c8
+set.m0.zpaq	1	set/text.txt	20000	6d7dcfff654e4f0b73c9c6748e20101a654d91340e808028d9cce43314e176b6
+set.m0.zpaq	1	set/x86.bin	16384	f80aea167fd0f16c0e8c59c394b8316b5626494ce468e4ef358e9b32cd3c05d4
+set.m1.zpaq	1	set/dup.txt	20000	6d7dcfff654e4f0b73c9c6748e20101a654d91340e808028d9cce43314e176b6
+set.m1.zpaq	1	set/empty.dat	0	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
+set.m1.zpaq	1	set/random.bin	4096	3a370c921e1d689daae1f95c1bc8fd70c3a2294d9a602603866c88f1678133ed
+set.m1.zpaq	1	set/sub/	-1	-
+set.m1.zpaq	1	set/sub/table.bin	16384	c30bfd6c26ff4d6994184a48cfe912d634d84741a506a379a756a88d03ce17c8
+set.m1.zpaq	1	set/text.txt	20000	6d7dcfff654e4f0b73c9c6748e20101a654d91340e808028d9cce43314e176b6
+set.m1.zpaq	1	set/x86.bin	16384	f80aea167fd0f16c0e8c59c394b8316b5626494ce468e4ef358e9b32cd3c05d4
+set.m2.zpaq	1	set/dup.txt	20000	6d7dcfff654e4f0b73c9c6748e20101a654d91340e808028d9cce43314e176b6
+set.m2.zpaq	1	set/empty.dat	0	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
+set.m2.zpaq	1	set/random.bin	4096	3a370c921e1d689daae1f95c1bc8fd70c3a2294d9a602603866c88f1678133ed
+set.m2.zpaq	1	set/sub/	-1	-
+set.m2.zpaq	1	set/sub/table.bin	16384	c30bfd6c26ff4d6994184a48cfe912d634d84741a506a379a756a88d03ce17c8
+set.m2.zpaq	1	set/text.txt	20000	6d7dcfff654e4f0b73c9c6748e20101a654d91340e808028d9cce43314e176b6
+set.m2.zpaq	1	set/x86.bin	16384	f80aea167fd0f16c0e8c59c394b8316b5626494ce468e4ef358e9b32cd3c05d4
+set.m3.zpaq	1	set/dup.txt	20000	6d7dcfff654e4f0b73c9c6748e20101a654d91340e808028d9cce43314e176b6
+set.m3.zpaq	1	set/empty.dat	0	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
+set.m3.zpaq	1	set/random.bin	4096	3a370c921e1d689daae1f95c1bc8fd70c3a2294d9a602603866c88f1678133ed
+set.m3.zpaq	1	set/sub/	-1	-
+set.m3.zpaq	1	set/sub/table.bin	16384	c30bfd6c26ff4d6994184a48cfe912d634d84741a506a379a756a88d03ce17c8
+set.m3.zpaq	1	set/text.txt	20000	6d7dcfff654e4f0b73c9c6748e20101a654d91340e808028d9cce43314e176b6
+set.m3.zpaq	1	set/x86.bin	16384	f80aea167fd0f16c0e8c59c394b8316b5626494ce468e4ef358e9b32cd3c05d4
+set.m4.zpaq	1	set/dup.txt	20000	6d7dcfff654e4f0b73c9c6748e20101a654d91340e808028d9cce43314e176b6
+set.m4.zpaq	1	set/empty.dat	0	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
+set.m4.zpaq	1	set/random.bin	4096	3a370c921e1d689daae1f95c1bc8fd70c3a2294d9a602603866c88f1678133ed
+set.m4.zpaq	1	set/sub/	-1	-
+set.m4.zpaq	1	set/sub/table.bin	16384	c30bfd6c26ff4d6994184a48cfe912d634d84741a506a379a756a88d03ce17c8
+set.m4.zpaq	1	set/text.txt	20000	6d7dcfff654e4f0b73c9c6748e20101a654d91340e808028d9cce43314e176b6
+set.m4.zpaq	1	set/x86.bin	16384	f80aea167fd0f16c0e8c59c394b8316b5626494ce468e4ef358e9b32cd3c05d4
+set.m5.zpaq	1	set/dup.txt	20000	6d7dcfff654e4f0b73c9c6748e20101a654d91340e808028d9cce43314e176b6
+set.m5.zpaq	1	set/empty.dat	0	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
+set.m5.zpaq	1	set/random.bin	4096	3a370c921e1d689daae1f95c1bc8fd70c3a2294d9a602603866c88f1678133ed
+set.m5.zpaq	1	set/sub/	-1	-
+set.m5.zpaq	1	set/sub/table.bin	16384	c30bfd6c26ff4d6994184a48cfe912d634d84741a506a379a756a88d03ce17c8
+set.m5.zpaq	1	set/text.txt	20000	6d7dcfff654e4f0b73c9c6748e20101a654d91340e808028d9cce43314e176b6
+set.m5.zpaq	1	set/x86.bin	16384	f80aea167fd0f16c0e8c59c394b8316b5626494ce468e4ef358e9b32cd3c05d4
+versions.zpaq	1	v/a.bin	3000	e17a27da77a519cad73adb40b725139379c057af5023170700519facc6e378e7
+versions.zpaq	1	v/notes.txt	6000	76c3dcfd3f6bd471dcbd9d27638a1663c5d1d0342dc1d3733d30e475922f7eff
+versions.zpaq	2	v/a.bin	3000	eaf76ffcb88a0e9add0d967f5c7c7f1918ac1c087fbf90c0c73030328d11684b
+versions.zpaq	2	v/b.txt	5000	09baec6ca8b2af5f64ce5cd913b5172d83008bc0d418dafe44b9f0f14264a121
+versions.zpaq	2	v/notes.txt	6000	76c3dcfd3f6bd471dcbd9d27638a1663c5d1d0342dc1d3733d30e475922f7eff
+versions.zpaq	3	v/a.bin	3000	eaf76ffcb88a0e9add0d967f5c7c7f1918ac1c087fbf90c0c73030328d11684b
+versions.zpaq	3	v/b.txt	5000	09baec6ca8b2af5f64ce5cd913b5172d83008bc0d418dafe44b9f0f14264a121
+stream.zpaq	1	st/one.txt	9000	fe0aff6997ba887be6c6d568be054f5600765c45a0ff3762a7c39f2d34a9f56a
+stream.zpaq	1	st/two.bin	7000	46f2d7bf41d57c5c516fad8b79c3630e264e86d8a43f22dae0dac6db793e87ba

Compression.Tests/Zpaq/ReferenceVectors/mkfixtures.py

@@ -0,0 +1,113 @@
+"""Builds the zpaq 7.15 reference archives for Compression.Tests/Zpaq/ReferenceVectors.
+
+Usage: python -I mkfixtures.py <zpaq.exe> <workdir> <outdir>
+The inputs are generated deterministically here; only the archives and a manifest
+(name, size, SHA-256 per version) are checked in.
+"""
+import hashlib, os, shutil, subprocess, sys
+
+zpaq, work, out = sys.argv[1], sys.argv[2], sys.argv[3]
+
+class Lcg:
+    def __init__(self, seed): self.s = seed & 0xffffffff
+    def next(self):
+        self.s = (self.s * 1103515245 + 12345) & 0xffffffff
+        return self.s >> 16
+
+WORDS = ("the of and to in is that for it as with was on be by at this which from or an are "
+         "archive journal fragment block version compress model context mixer predictor stream "
+         "zpaq level method index header data file directory size date hash checksum").split()
+
+def text(n, seed):
+    r = Lcg(seed); b = bytearray()
+    while len(b) < n:
+        w = WORDS[r.next() % len(WORDS)]
+        b += w.encode()
+        k = r.next() % 16
+        b += b'.\n' if k == 0 else (b', ' if k == 1 else b' ')
+    return bytes(b[:n])
+
+def rand(n, seed):
+    r = Lcg(seed); return bytes(r.next() & 255 for _ in range(n))
+
+def x86(n, seed):
+    r = Lcg(seed); b = bytearray()
+    while len(b) < n:
+        k = r.next() % 8
+        if k == 0: b += bytes([0xE8, r.next() & 255, r.next() & 15, 0, 0])
+        elif k == 1: b += bytes([0x8B, 0x45 + (r.next() & 7) * 8, r.next() & 0x7C])
+        elif k == 2: b += bytes([0xE9, r.next() & 255, 0xFF, 0xFF, 0xFF])
+        else: b += bytes([0x89, 0x44, 0x24, (r.next() & 15) * 4])
+    return bytes(b[:n])
+
+def table(n):
+    return bytes(((i * 7) ^ (i >> 4)) & 255 if i % 4 else (i >> 2) & 255 for i in range(n))
+
+def write(root, name, data):
+    p = os.path.join(root, name)
+    os.makedirs(os.path.dirname(p), exist_ok=True)
+    with open(p, 'wb') as f: f.write(data)
+
+def run(*args, cwd):
+    r = subprocess.run([zpaq, *args], cwd=cwd, capture_output=True, text=True)
+    if r.returncode != 0:
+        raise SystemExit(f"zpaq {' '.join(args)} failed:\n{r.stdout}\n{r.stderr}")
+
+def manifest_of(root):
+    rows = []
+    for dp, dns, fns in os.walk(root):
+        for d in dns:
+            rel = os.path.relpath(os.path.join(dp, d), os.path.dirname(root)).replace('\\', '/') + '/'
+            rows.append((rel, -1, '-'))
+        for f in fns:
+            p = os.path.join(dp, f)
+            rel = os.path.relpath(p, os.path.dirname(root)).replace('\\', '/')
+            data = open(p, 'rb').read()
+            rows.append((rel, len(data), hashlib.sha256(data).hexdigest()))
+    return sorted(rows)
+
+shutil.rmtree(work, ignore_errors=True); os.makedirs(work); os.makedirs(out, exist_ok=True)
+lines = ["# archive\tversion\tname\tsize\tsha256 (size -1 = directory)"]
+
+# One file set, every predefined method.
+s = os.path.join(work, 'set')
+t = text(20000, 1)
+write(s, 'text.txt', t)
+write(s, 'dup.txt', t)                 # deduplicated against text.txt
+write(s, 'x86.bin', x86(16384, 2))
+write(s, 'random.bin', rand(4096, 3))
+write(s, 'empty.dat', b'')
+write(s, 'sub/table.bin', table(16384))
+setrows = manifest_of(s)
+for m in range(6):
+    name = f'set.m{m}.zpaq'
+    run('a', name, 'set', '-method', str(m), cwd=work)
+    shutil.copy(os.path.join(work, name), out)
+    lines += [f"{name}\t1\t{n}\t{z}\t{h}" for n, z, h in setrows]
+
+# Three versions: add, modify + add, delete.
+v = os.path.join(work, 'v')
+write(v, 'notes.txt', text(6000, 4))
+write(v, 'a.bin', rand(3000, 5))
+lines += [f"versions.zpaq\t1\t{n}\t{z}\t{h}" for n, z, h in manifest_of(v)]
+run('a', 'versions.zpaq', 'v', cwd=work)
+write(v, 'a.bin', rand(3000, 6))
+write(v, 'b.txt', text(5000, 7))
+lines += [f"versions.zpaq\t2\t{n}\t{z}\t{h}" for n, z, h in manifest_of(v)]
+run('a', 'versions.zpaq', 'v', cwd=work)
+os.remove(os.path.join(v, 'notes.txt'))
+lines += [f"versions.zpaq\t3\t{n}\t{z}\t{h}" for n, z, h in manifest_of(v)]
+run('a', 'versions.zpaq', 'v', cwd=work)
+shutil.copy(os.path.join(work, 'versions.zpaq'), out)
+
+# Streaming layout (method s...): BWT + order-0/1 CM, one block per file.
+st = os.path.join(work, 'st')
+write(st, 'one.txt', text(9000, 8))
+write(st, 'two.bin', x86(7000, 9))
+run('a', 'stream.zpaq', 'st', '-method', 's4.3ci1', cwd=work)
+shutil.copy(os.path.join(work, 'stream.zpaq'), out)
+lines += [f"stream.zpaq\t1\t{n}\t{z}\t{h}" for n, z, h in manifest_of(st) if z >= 0]
+
+with open(os.path.join(out, 'manifest.tsv'), 'w', newline='\n') as f:
+    f.write('\n'.join(lines) + '\n')
+print('ok')

Compression.Tests/Zpaq/ZpaqEngineTests.cs

@@ -0,0 +1,128 @@
+using System.Text;
+using Compression.Tests.Csc;
+using FileFormat.Zpaq;
+
+namespace Compression.Tests.Zpaq;
+
+/// <summary>
+/// The libzpaq conversion below the archive layer: suffix arrays, the ZPAQL compiler and
+/// interpreter, method expansion, and single-block streams.
+/// </summary>
+[TestFixture]
+public class ZpaqEngineTests {
+
+  // ── suffix array ─────────────────────────────────────────────────────────
+
+  private static int[] NaiveSuffixArray(byte[] s) =>
+    Enumerable.Range(0, s.Length).OrderBy(i => i, Comparer<int>.Create((a, b) => s.AsSpan(a).SequenceCompareTo(s.AsSpan(b)))).ToArray();
+
+  [Test, Category("HappyPath")]
+  public void SuffixArray_MatchesNaiveSort([Values(0, 1, 2, 3, 17, 100, 1000)] int length, [Values(2, 4, 256)] int alphabet) {
+    var rng = new Random(length * 31 + alphabet);
+    var s = new byte[length];
+    for (var i = 0; i < s.Length; ++i)
+      s[i] = (byte)rng.Next(alphabet);
+    Assert.That(ZpaqSuffixArray.Build(s), Is.EqualTo(NaiveSuffixArray(s)));
+  }
+
+  [Test, Category("BoundaryCase")]
+  public void SuffixArray_PeriodicAndConstantInputs() {
+    foreach (var s in new[] { Enumerable.Repeat((byte)7, 500).ToArray(), Encoding.ASCII.GetBytes(string.Concat(Enumerable.Repeat("abcab", 97))), Encoding.ASCII.GetBytes("mississippi") })
+      Assert.That(ZpaqSuffixArray.Build(s), Is.EqualTo(NaiveSuffixArray(s)));
+  }
+
+  // ── squash / stretch (specification section 7, checked by libzpaq's own self test) ──
+
+  [Test, Category("HappyPath")]
+  public void SquashAndStretch_MatchLibzpaqChecksums() {
+    uint stsum = 0, sqsum = 0;
+    for (var i = 32767; i >= 0; --i)
+      stsum = stsum * 3 + (uint)ZpaqPredictor.Stretch(i);
+    for (var i = 4095; i >= 0; --i)
+      sqsum = sqsum * 3 + (uint)ZpaqPredictor.Squash(i - 2048);
+    Assert.Multiple(() => {
+      Assert.That(stsum, Is.EqualTo(3887533746u));
+      Assert.That(sqsum, Is.EqualTo(2278286169u));
+    });
+  }
+
+  // ── compiler ─────────────────────────────────────────────────────────────
+
+  [Test, Category("HappyPath")]
+  public void Compiler_ProducesLibzpaqMinCfgBytecode() {
+    // min.cfg from the libzpaq documentation; the expected bytes are libzpaq's built-in "Model 1".
+    const string minCfg = "comp 1 2 0 0 2 0 icm 16 1 isse 19 0 hcomp *b=a a=0 d=0 hash b-- hash *d=a d++ b-- hash b-- hash *d=a halt post 0 end";
+    var hz = new ZpaqlMachine();
+    var pz = new ZpaqlMachine();
+    _ = new ZpaqCompiler(minCfg, null, hz, pz);
+    var ms = new ZpaqBuffer();
+    hz.Write(ms, false);
+    byte[] model1 = [26, 0, 1, 2, 0, 0, 2, 3, 16, 8, 19, 0, 0, 96, 4, 28, 59, 10, 59, 112, 25, 10, 59, 10, 59, 112, 56, 0];
+    Assert.That(ms.ToArray(), Is.EqualTo(model1));
+  }
+
+  [Test, Category("Exception")]
+  public void Compiler_RejectsBadConfigs([Values("comp 0 0 0 0 1 0 bogus hcomp halt end", "comp 0 0 0 0 0 hcomp endif halt end", "comp 0 0 0 0 0 hcomp halt")] string config) =>
+    Assert.That(() => new ZpaqCompiler(config, null, new(), new()), Throws.InstanceOf<InvalidDataException>());
+
+  [Test, Category("HappyPath")]
+  public void MakeConfig_ExpandsArguments() {
+    var args = new int[9];
+    var config = ZpaqMethods.MakeConfig("x6.3ci1", args);
+    Assert.Multiple(() => {
+      Assert.That(args[..2], Is.EqualTo(new[] { 6, 3 }));
+      Assert.That(config, Does.StartWith("comp 9 16 $1+20 $1+20 2\n0 icm"));
+      Assert.That(config, Does.Contain("pcomp bwtrle c ;"));
+      Assert.That(ZpaqMethods.MakeConfig("0", args), Is.EqualTo("comp 0 0 0 0 0 hcomp end\n"));
+    });
+  }
+
+  // ── single-block streams ─────────────────────────────────────────────────
+
+  private static byte[] RoundTrip(byte[] data, string method) {
+    var output = new ZpaqBuffer();
+    ZpaqMethods.CompressBlock(new ZpaqBuffer(data), output, method, "name", null, storeSha1: true);
+    return ZpaqStream.Decompress(output.Span);
+  }
+
+  [Test, Category("RoundTrip")]
+  public void Block_EveryTransformAndModel_RoundTrips([Values("0", "x0.0c0.0.255", "x0.1.4.0.3.20", "x0.2.4.0.3.20c0.0.511", "x0.3ci1", "x0.4", "x0.5.4.0.3.20", "x0.6.4.0.3.20", "x0.7ci1",
+      "x0.1.4.0.5.21", "x0.0c256.0.255.1511.255i1aw1m16s8.32.255", "x0.0c0.1010.255t", "x0.0ci1,1,1,1,2aw2mm16tst")] string method) {
+    byte[] data = [.. CscSamples.Text(9000, 31), .. CscSamples.X86(7000, 32), .. CscSamples.RandomBytes(2000, 33)];
+    Assert.That(RoundTrip(data, method), Is.EqualTo(data));
+  }
+
+  [Test, Category("RoundTrip")]
+  public void Block_LevelsWithHints_RoundTrip([Values("1", "2", "3", "4", "5")] string level, [Values(",0,0", ",200,1", ",120,2", ",250,3")] string hints) {
+    var data = CscSamples.Mixed();
+    Assert.That(RoundTrip(data, level + "0" + hints), Is.EqualTo(data));
+  }
+
+  [Test, Category("BoundaryCase")]
+  public void Block_EmptyAndOneByte_RoundTrip([Values("0", "x0.1.4.0.3.20", "x0.2.4.0.3.20", "x0.3ci1", "x0.0ci1")] string method) {
+    Assert.Multiple(() => {
+      Assert.That(RoundTrip([], method), Is.Empty);
+      Assert.That(RoundTrip([42], method), Is.EqualTo(new byte[] { 42 }));
+    });
+  }
+
+  [Test, Category("Exception")]
+  public void Block_FlippedChecksum_IsDetected() {
+    var output = new ZpaqBuffer();
+    ZpaqMethods.CompressBlock(new ZpaqBuffer(CscSamples.Text(5000, 34)), output, "x0.0ci1", null, null, storeSha1: true);
+    var bytes = output.ToArray();
+    bytes[^3] ^= 1;  // inside the SHA-1 trailer, before the end-of-block byte
+    Assert.That(() => ZpaqStream.Decompress(bytes), Throws.InstanceOf<InvalidDataException>());
+  }
+
+  [Test, Category("HappyPath")]
+  public void E8E9_IsUndoneByThePostprocessor() {
+    var data = CscSamples.X86(30_000, 35);
+    var copy = (byte[])data.Clone();
+    ZpaqMethods.E8E9(copy);
+    Assert.Multiple(() => {
+      Assert.That(copy, Is.Not.EqualTo(data), "the transform changes call targets");
+      Assert.That(RoundTrip(data, "x0.4"), Is.EqualTo(data));
+    });
+  }
+}