Never delete empty files when deduplicating [patch]

FileDeduplicator.Test/DeduplicatorTests.cs

@@ -207,6 +207,34 @@ public void DeleteDuplicatesRemovesEveryCopyExceptTheKeeper()
 		}
 	}
 
+	/// <summary>
+	/// Empty files group together, but deleting one frees nothing and an empty <c>__init__.py</c>
+	/// or <c>.gitkeep</c> matters because it exists, so only the real duplicate is deleted.
+	/// </summary>
+	[TestMethod]
+	public void EmptyFilesAreNeverDeleted()
+	{
+		// Arrange
+		using TempTree tree = new();
+		AbsoluteFilePath initA = tree.Write("a/__init__.py", string.Empty);
+		AbsoluteFilePath initB = tree.Write("b/__init__.py", string.Empty);
+		AbsoluteFilePath copy = tree.Write("copy.txt", "shared");
+		AbsoluteFilePath keeper = tree.Write("k.txt", "shared");
+		IReadOnlyList<DuplicateGroup> duplicates = Duplicates(FileHasher.HashFiles([initA, initB, copy, keeper]));
+
+		// Act
+		DeduplicationResult result = Deduplicator.DeleteDuplicates(duplicates);
+
+		// Assert
+		Assert.HasCount(2, duplicates, "The empty files are still reported as a group.");
+		Assert.AreEqual(1, result.DeletedCount);
+		Assert.IsTrue(TempTree.Exists(initA), "An empty file must survive.");
+		Assert.IsTrue(TempTree.Exists(initB), "An empty file must survive.");
+		Assert.IsTrue(TempTree.Exists(keeper), "The keeper must survive.");
+		Assert.IsFalse(TempTree.Exists(copy), "The real duplicate should have been deleted.");
+		Assert.IsEmpty(result.SkippedFiles);
+	}
+
 	/// <summary>
 	/// The reclaimed byte count must reflect what was actually removed.
 	/// </summary>

FileDeduplicator.Test/DryRunEquivalenceTests.cs

@@ -31,7 +31,7 @@ private static HashSet<AbsoluteFilePath> PredictedDeletions(IReadOnlyList<Duplic
 	{
 		HashSet<AbsoluteFilePath> predicted = [];
 
-		foreach (DuplicateGroup group in duplicates)
+		foreach (DuplicateGroup group in duplicates.Where(Deduplicator.IsDeletable))
 		{
 			AbsoluteFilePath keeper = Deduplicator.SelectFileToKeep(group.Files);
 			foreach (AbsoluteFilePath file in group.Files.Where(f => f != keeper))

FileDeduplicator.Test/DuplicateReportTests.cs

@@ -43,6 +43,29 @@ public void PlanCountsEveryCopyExceptEachGroupsKeeper()
 		Assert.AreEqual((2 * 5L) + 4L, plan.BytesReclaimable);
 	}
 
+	/// <summary>
+	/// Empty files are never deleted, so the plan keeps all of them and counts none, matching what
+	/// Deduplicate then does.
+	/// </summary>
+	[TestMethod]
+	public void PlanKeepsEveryEmptyFile()
+	{
+		// Arrange
+		using TempTree tree = new();
+		_ = tree.Write("a/__init__.py", string.Empty);
+		_ = tree.Write("b/__init__.py", string.Empty);
+		_ = tree.Write("logs/.gitkeep", string.Empty);
+
+		// Act
+		DeletionPlan plan = DuplicateReport.PlanDeletions(DuplicatesIn(tree));
+
+		// Assert
+		Assert.AreEqual(0, plan.FileCount);
+		Assert.AreEqual(0, plan.BytesReclaimable);
+		Assert.DoesNotContain(line => line.Contains("DELETE:", StringComparison.Ordinal), plan.Listing, "No empty file may be listed for deletion.");
+		Assert.AreEqual(3, plan.Listing.Count(line => line.Contains("KEEP:", StringComparison.Ordinal)));
+	}
+
 	/// <summary>
 	/// The listing names one keeper and every other copy in each group, so the two sets together
 	/// account for every file the group holds.

FileDeduplicator/Deduplicator.cs

@@ -97,14 +97,26 @@ private static bool TryGetSize(AbsoluteFilePath file, out long size)
 	internal static AbsoluteFilePath SelectFileToKeep(List<AbsoluteFilePath> duplicates) =>
 		duplicates.OrderBy(f => f.FileName.WeakString.Length).ThenBy(f => f.WeakString, StringComparer.Ordinal).First();
 
+	/// <summary>
+	/// Reports whether deduplication may delete copies from a group.
+	/// </summary>
+	/// <remarks>
+	/// Every zero-length file hashes the same, so an empty <c>__init__.py</c>, a <c>.gitkeep</c> and a
+	/// <c>py.typed</c> marker land in one group. Such files matter because they exist, not for what
+	/// they hold, and deleting them frees nothing, so the group is still reported but never deleted.
+	/// </remarks>
+	/// <param name="group">The group to check.</param>
+	/// <returns><see langword="true"/> unless the group's files are empty.</returns>
+	internal static bool IsDeletable(DuplicateGroup group) => group.FileSize > 0;
+
 	internal static DeduplicationResult DeleteDuplicates(IReadOnlyList<DuplicateGroup> duplicateGroups)
 	{
 		int deletedCount = 0;
 		long bytesReclaimed = 0;
 		List<string> errors = [];
 		List<SkippedFile> skipped = [];
 
-		foreach (DuplicateGroup group in duplicateGroups)
+		foreach (DuplicateGroup group in duplicateGroups.Where(IsDeletable))
 		{
 			AbsoluteFilePath keeper = SelectFileToKeep(group.Files);
 

FileDeduplicator/DuplicateReport.cs

@@ -31,9 +31,21 @@ internal static DeletionPlan PlanDeletions(IReadOnlyList<DuplicateGroup> duplica
 
 		foreach (DuplicateGroup group in duplicates)
 		{
-			AbsoluteFilePath keeper = Deduplicator.SelectFileToKeep(group.Files);
-
 			listing.Add($"  Hash: {group.Hash[..12]}... ({FormatBytes(group.FileSize)}, {group.Files.Count} copies)");
+
+			if (!Deduplicator.IsDeletable(group))
+			{
+				listing.Add("    Empty files are never deleted:");
+				foreach (AbsoluteFilePath file in group.Files)
+				{
+					listing.Add($"    KEEP:   {file}");
+				}
+
+				listing.Add(string.Empty);
+				continue;
+			}
+
+			AbsoluteFilePath keeper = Deduplicator.SelectFileToKeep(group.Files);
 			listing.Add($"    KEEP:   {keeper}");
 
 			foreach (AbsoluteFilePath file in group.Files)

FileDeduplicator/Verbs/Scan.cs

@@ -68,14 +68,20 @@ internal override void Run(Scan options)
 		foreach (DuplicateGroup group in duplicates)
 		{
 			AbsoluteFilePath keeper = Deduplicator.SelectFileToKeep(group.Files);
+			bool deletable = Deduplicator.IsDeletable(group);
 			long wastedBytes = group.FileSize * (group.Files.Count - 1);
 			totalWastedBytes += wastedBytes;
 
 			Console.WriteLine($"  Hash: {group.Hash[..12]}... ({DuplicateReport.FormatBytes(group.FileSize)}, {group.Files.Count} copies)");
 
 			foreach (AbsoluteFilePath file in group.Files)
 			{
-				string marker = file == keeper ? " [KEEP]" : " [DELETE]";
+				string marker = (deletable, file == keeper) switch
+				{
+					(false, _) => " [KEEP, empty]",
+					(true, true) => " [KEEP]",
+					(true, false) => " [DELETE]",
+				};
 				Console.WriteLine($"    {file}{marker}");
 			}
 

README.md

@@ -58,6 +58,8 @@ FileDeduplicator Stats -p "C:\path\to\directory"
 
 When duplicates are found, the file with the **shortest filename** is kept. If two files have filenames of equal length, the one with the lexicographically first full path is kept.
 
+Empty files are reported as a group but never deleted. Deleting one frees nothing, and an empty file such as `__init__.py`, `.gitkeep` or `py.typed` usually matters because it exists.
+
 For example, given these duplicates:
 - `C:\photos\vacation\IMG_20240101_123456.jpg`
 - `C:\photos\beach.jpg`