Ensure fully deterministic zip output (#45)

* Ensure fully deterministic ZIP output across net48 and net10.0

  - Replace framework ZipArchive writing with raw ZipStorer to guarantee
    compression method 0 (Stored) on all frameworks (net48 ignores
    CompressionLevel.NoCompression)
  - Write raw zlib stored blocks in PngNormalizer to avoid DEFLATE
    differences between framework ZLibStream implementations
  - Sort ZIP entries by name using ordinal comparison
  - Sort [Content_Types].xml elements deterministically (ContentTypesPatcher)
  - Renumber all relationship IDs in .rels files to DeterministicId{n} and
    remap corresponding r:id references in content XML
  - Add verification tests for stored entries, deterministic relationship
    IDs, and sorted content types

* .

* .

* .

Author: SimonCropp

claude.md

@@ -43,12 +43,37 @@ The codebase uses a patcher pattern for normalizing OOXML content:
 - Each patcher implements `IPatcher` interface
 - `IsMatch(Entry entry)` - determines which files the patcher applies to (e.g., "word/document.xml")
 - `PatchXml(XDocument xml)` - modifies the XML in-place
-- Register patchers in `DeterministicPackage.cs` patchers list
-- **Order matters** - patchers run in sequence
+- Register patchers via `CreatePatchers()` factory in `DeterministicPackage.cs` (fresh instance per conversion)
+- **Order matters** - relationship patchers must run before their content patchers (e.g., `WorkbookRelationshipPatcher` before `WorkbookPatcher`)
+
+#### Paired Patchers
+
+Some patchers work in pairs: a relationship patcher renumbers IDs in `.rels` files and stores the mapping, then a content patcher remaps `r:id` references in the corresponding XML:
+
+- `WorkbookRelationshipPatcher` → `WorkbookPatcher` (xl/_rels/workbook.xml.rels → xl/workbook.xml)
+- `DocumentRelationshipPatcher` → `DocumentPatcher` (word/_rels/document.xml.rels → word/document.xml)
+
+The content patcher receives the relationship patcher via constructor injection.
+
+#### Relationship ID Renumbering
+
+`RelationshipRenumber` (in IPatcher.cs) provides shared helpers:
+- `RenumberAndSort(XDocument)` — sorts relationships by Type+Target, renumbers to `DeterministicId{n}`, returns old→new mapping
+- `RemapIds(XDocument, mapping)` — replaces `r:id` attribute values in content XML using the mapping
+
+#### Content Types Sorting
+
+`ContentTypesPatcher` sorts `[Content_Types].xml` elements by local name, then Extension, then PartName to ensure deterministic order across frameworks.
+
+### ZIP Output
+
+- `ZipStorer` rewrites ZIP archives with all entries using compression method 0 (Stored), bypassing net48's `ZipArchive` which ignores `CompressionLevel.NoCompression`
+- Entries are sorted by `FullName` using `StringComparer.Ordinal`
+- `PngNormalizer` writes raw zlib stored blocks (CMF+FLG + DEFLATE stored blocks + Adler-32) instead of using `ZLibStream`, which produces different output on net48 vs net10.0
 
 Example patcher structure:
 ```csharp
-class DocumentPatcher : IPatcher
+class DocumentPatcher(DocumentRelationshipPatcher relsPatcher) : IPatcher
 {
     static XNamespace wp = "http://schemas.openxmlformats.org/drawingml/2006/wordprocessingDrawing";
     static XNamespace pic = "http://schemas.openxmlformats.org/drawingml/2006/picture";
@@ -58,9 +83,12 @@ class DocumentPatcher : IPatcher
 
     public void PatchXml(XDocument xml)
     {
-        var root = xml.Root!;
-        var elements = root.Descendants(wp + "docPr").ToList();
-        // Normalize IDs...
+        // Normalize drawing IDs...
+        // Then remap relationship IDs
+        if (relsPatcher.IdMapping.Count > 0)
+        {
+            RelationshipRenumber.RemapIds(xml, relsPatcher.IdMapping);
+        }
     }
 }
 ```

readme.md

@@ -44,7 +44,7 @@ Example file formats that leverage System.IO.Packaging
 using var sourceStream = File.OpenRead(packagePath);
 await DeterministicPackage.ConvertAsync(sourceStream, targetStream);
 ```
-<sup><a href='/src/Tests/Tests.cs#L174-L179' title='Snippet source file'>snippet source</a> | <a href='#snippet-ConvertAsync' title='Start of snippet'>anchor</a></sup>
+<sup><a href='/src/Tests/Tests.cs#L226-L231' title='Snippet source file'>snippet source</a> | <a href='#snippet-ConvertAsync' title='Start of snippet'>anchor</a></sup>
 <!-- endSnippet -->
 
 
@@ -56,7 +56,7 @@ await DeterministicPackage.ConvertAsync(sourceStream, targetStream);
 using var sourceStream = File.OpenRead(packagePath);
 await DeterministicPackage.ConvertAsync(sourceStream, targetStream);
 ```
-<sup><a href='/src/Tests/Tests.cs#L174-L179' title='Snippet source file'>snippet source</a> | <a href='#snippet-ConvertAsync' title='Start of snippet'>anchor</a></sup>
+<sup><a href='/src/Tests/Tests.cs#L226-L231' title='Snippet source file'>snippet source</a> | <a href='#snippet-ConvertAsync' title='Start of snippet'>anchor</a></sup>
 <!-- endSnippet -->
 
 

src/DeterministicIoPackaging/DeterministicPackage.cs

@@ -5,18 +5,24 @@ public static partial class DeterministicPackage
     public static DateTime StableDate { get; } = new(2020, 1, 1, 0, 0, 0, DateTimeKind.Utc);
     public static DateTimeOffset StableDateOffset { get; } = new(StableDate);
 
-    static IReadOnlyList<IPatcher> patchers =
-    [
-        new RelationshipPatcher(),
-        new SheetPatcher(),
-        new WorkbookPatcher(),
-        new WorkbookRelationshipPatcher(),
-        new CorePatcher(),
-        new SheetRelationshipPatcher(),
-        new DocumentRelationshipPatcher(),
-        new DocumentPatcher(),
-        new NumberingPatcher()
-    ];
+    static IReadOnlyList<IPatcher> CreatePatchers()
+    {
+        var workbookRelsPatcher = new WorkbookRelationshipPatcher();
+        var documentRelsPatcher = new DocumentRelationshipPatcher();
+        return
+        [
+            new ContentTypesPatcher(),
+            new RelationshipPatcher(),
+            new SheetPatcher(),
+            workbookRelsPatcher,
+            new WorkbookPatcher(workbookRelsPatcher),
+            new CorePatcher(),
+            new SheetRelationshipPatcher(),
+            documentRelsPatcher,
+            new DocumentPatcher(documentRelsPatcher),
+            new NumberingPatcher()
+        ];
+    }
 
     static Archive CreateArchive(Stream target) => new(target, ZipArchiveMode.Create, leaveOpen: true);
 
@@ -30,7 +36,7 @@ static Archive ReadArchive(Stream source)
         return new(source, ZipArchiveMode.Read, leaveOpen: true);
     }
 
-    static void DuplicateEntry(Entry sourceEntry, Archive targetArchive)
+    static void DuplicateEntry(Entry sourceEntry, Archive targetArchive, IReadOnlyList<IPatcher> currentPatchers)
     {
         if (IsPsmdcp(sourceEntry))
         {
@@ -41,7 +47,7 @@ static void DuplicateEntry(Entry sourceEntry, Archive targetArchive)
         var targetEntry = CreateEntry(sourceEntry, targetArchive);
         using var targetStream = targetEntry.Open();
 
-        foreach (var patcher in patchers)
+        foreach (var patcher in currentPatchers)
         {
             if (!patcher.IsMatch(sourceEntry))
             {
@@ -63,7 +69,7 @@ static void DuplicateEntry(Entry sourceEntry, Archive targetArchive)
         sourceStream.CopyTo(targetStream);
     }
 
-    static async Task DuplicateEntryAsync(Entry sourceEntry, Archive targetArchive, Cancel cancel)
+    static async Task DuplicateEntryAsync(Entry sourceEntry, Archive targetArchive, IReadOnlyList<IPatcher> currentPatchers, Cancel cancel)
     {
         if (IsPsmdcp(sourceEntry))
         {
@@ -73,7 +79,7 @@ static async Task DuplicateEntryAsync(Entry sourceEntry, Archive targetArchive,
         using var sourceStream = await sourceEntry.OpenAsync(cancel);
         var targetEntry = CreateEntry(sourceEntry, targetArchive);
         using var targetStream = await targetEntry.OpenAsync(cancel);
-        foreach (var patcher in patchers)
+        foreach (var patcher in currentPatchers)
         {
             if (!patcher.IsMatch(sourceEntry))
             {

src/DeterministicIoPackaging/DeterministicPackage_Convert.cs

@@ -20,13 +20,14 @@ public static async Task<MemoryStream> ConvertAsync(Stream source)
 
     public static void Convert(Stream source, Stream target)
     {
+        var patchers = CreatePatchers();
         var intermediate = new MemoryStream();
         using (var sourceArchive = ReadArchive(source))
         using (var targetArchive = CreateArchive(intermediate))
         {
             foreach (var sourceEntry in sourceArchive.OrderedEntries())
             {
-                DuplicateEntry(sourceEntry, targetArchive);
+                DuplicateEntry(sourceEntry, targetArchive, patchers);
             }
         }
 
@@ -35,13 +36,14 @@ public static void Convert(Stream source, Stream target)
 
     public static async Task ConvertAsync(Stream source, Stream target, Cancel token = default)
     {
+        var patchers = CreatePatchers();
         var intermediate = new MemoryStream();
         using (var sourceArchive = ReadArchive(source))
         using (var targetArchive = CreateArchive(intermediate))
         {
             foreach (var sourceEntry in OrderedEntries(sourceArchive))
             {
-                await DuplicateEntryAsync(sourceEntry, targetArchive, token);
+                await DuplicateEntryAsync(sourceEntry, targetArchive, patchers, token);
             }
         }
 

src/DeterministicIoPackaging/GlobalUsings.cs

@@ -1,3 +1,3 @@
-global using System.Globalization;
 global using System.IO.Compression;
-global using System.Xml.Linq;
+global using System.IO.Hashing;
+global using System.Xml.Linq;

src/DeterministicIoPackaging/Patching/ContentTypesPatcher.cs

@@ -0,0 +1,17 @@
+class ContentTypesPatcher : IPatcher
+{
+    public bool IsMatch(Entry entry) =>
+        entry.FullName is "[Content_Types].xml";
+
+    public void PatchXml(XDocument xml)
+    {
+        var root = xml.Root!;
+        var elements = root.Elements()
+            .OrderBy(_ => _.Name.LocalName)
+            .ThenBy(_ => (string?)_.Attribute("Extension") ?? "")
+            .ThenBy(_ => (string?)_.Attribute("PartName") ?? "")
+            .ToList();
+
+        root.ReplaceAll(elements);
+    }
+}

src/DeterministicIoPackaging/Patching/DocumentPatcher.cs

@@ -1,4 +1,4 @@
-class DocumentPatcher : IPatcher
+class DocumentPatcher(DocumentRelationshipPatcher relsPatcher) : IPatcher
 {
     static XNamespace wp = "http://schemas.openxmlformats.org/drawingml/2006/wordprocessingDrawing";
     static XNamespace pic = "http://schemas.openxmlformats.org/drawingml/2006/picture";
@@ -23,5 +23,10 @@ public void PatchXml(XDocument xml)
             // Use index + 1 for 1-based numbering (common in Office Open XML)
             elementsWithIds[i].Attribute("id")!.Value = (i + 1).ToString();
         }
+
+        if (relsPatcher.IdMapping.Count > 0)
+        {
+            RelationshipRenumber.RemapIds(xml, relsPatcher.IdMapping);
+        }
     }
 }

src/DeterministicIoPackaging/Patching/DocumentRelationshipPatcher.cs

@@ -1,15 +1,10 @@
 class DocumentRelationshipPatcher : IPatcher
 {
+    internal Dictionary<string, string> IdMapping { get; private set; } = [];
+
     public bool IsMatch(Entry entry) =>
         entry.FullName is "word/_rels/document.xml.rels";
 
-    public void PatchXml(XDocument xml)
-    {
-        var root = xml.Root!;
-        var relationships = root.Elements()
-            .OrderBy(_ => _.Attribute("Type")!.Value)
-            .ThenBy(_ => _.Attribute("Target")!.Value)
-            .ToList();
-        root.ReplaceAll(relationships);
-    }
+    public void PatchXml(XDocument xml) =>
+        IdMapping = RelationshipRenumber.RenumberAndSort(xml);
 }

src/DeterministicIoPackaging/Patching/IPatcher.cs

@@ -1,5 +1,5 @@
-interface IPatcher
+interface IPatcher
 {
     public void PatchXml(XDocument xml);
     public bool IsMatch(Entry entry);
-}
+}

src/DeterministicIoPackaging/Patching/RelationshipPatcher.cs

@@ -6,28 +6,18 @@ public bool IsMatch(Entry entry) =>
     public void PatchXml(XDocument xml)
     {
         var root = xml.Root!;
-        var relationships = root.Elements()
-            .OrderBy(_ => _.Attribute("Type")!.Value)
-            .ThenBy(_ => _.Attribute("Target")!.Value)
-            .ToList();
 
-        foreach (var element in relationships.Where(IsPsmdcpElement).ToList())
+        foreach (var element in root.Elements().Where(IsPsmdcpElement).ToList())
         {
-            relationships.Remove(element);
+            element.Remove();
         }
 
-        for (var index = 0; index < relationships.Count; index++)
-        {
-            var relationship = relationships[index];
-            relationship.Attribute("Id")!.SetValue($"DeterministicId{index + 1}");
-        }
-
-        root.ReplaceAll(relationships);
+        RelationshipRenumber.RenumberAndSort(xml);
     }
 
     static bool IsPsmdcpElement(XElement element)
     {
         var target = element.Attribute("Target")!;
         return target.Value.EndsWith(".psmdcp");
     }
-}
+}