Fix #3850: recover ReadOnlySpan<T> array literals from the legacy lazy cache by sailro · Pull Request #3851 · icsharpcode/ILSpy

sailro · 2026-07-03T08:14:45Z

On target frameworks without RuntimeHelpers.CreateSpan (.NET Framework, or netstandard2.0 + System.Memory), Roslyn caches the backing array of a ReadOnlySpan<T> built from an array literal in a lazy <PrivateImplementationDetails> static field:

object obj = <PrivateImplementationDetails>.cache;
if (obj == null) {
    obj = new char[] { CR, LF };
    <PrivateImplementationDetails>.cache = (char[])obj;
}
... new ReadOnlySpan<char>((char[])obj) ...

The decompiled output referenced the compiler-synthesized <PrivateImplementationDetails> type, which is never declared and whose name is not expressible in C#, so the output failed to recompile (CS0400 in whole-project output).

Fix

A new CachedReadOnlySpanInitialization block transform collapses the lazy cache, mirroring CachedDelegateInitialization (which collapses the same lazy-static-field cache for anonymous-method delegates) and running right after it. Once the cache is collapsed, the existing array-initializer transforms recover the array literal, so the <PrivateImplementationDetails> reference disappears - the same result the modern RuntimeHelpers.CreateSpan path (net7+) already produces.

Before (does not recompile):

public static ReadOnlySpan<char> NewLine
{
    get
    {
        object obj = <PrivateImplementationDetails>.B7F5...7075_A1;
        if (obj == null)
        {
            obj = new char[2] { CR, LF };
            <PrivateImplementationDetails>.B7F5...7075_A1 = (char[])obj;
        }
        return new ReadOnlySpan<char>((char[])obj);
    }
}

After:

public static ReadOnlySpan<char> NewLine => new ReadOnlySpan<char>(new char[2] { CR, LF });

The transform only fires on a compiler-generated static cache field with this exact lazy-init shape, so it does not touch the existing <PrivateImplementationDetails> switch-on-string cache (covered by the existing CS1xSwitch_Debug test) or user code.

Test

ILPretty/CachedReadOnlySpanInitialization exercises the legacy lazy-cache IL and checks the recovered output.

Real-world occurrence: Markdig CodeInlineParser (net462).

… legacy lazy cache Roslyn caches a ReadOnlySpan<T> created from an array literal in a <PrivateImplementationDetails> field on target frameworks without RuntimeHelpers.CreateSpan (e.g. .NET Framework / netstandard2.0 + System.Memory): object obj = <PrivateImplementationDetails>.cache; if (obj == null) { obj = new char[] { '\r', '\n' }; <PrivateImplementationDetails>.cache = (char[])obj; } ... new ReadOnlySpan<char>((char[])obj) ... The decompiled output referenced the compiler-synthesized <PrivateImplementationDetails> type, whose escaped name is not expressible in C# and is never declared, so the output failed to recompile (CS0400). The modern RuntimeHelpers.CreateSpan form was already handled (TransformRuntimeHelpersCreateSpanInitialization); this adds the analogous handling for the legacy lazy-cache form, mirroring CachedDelegateInitialization (which collapses the same lazy-static-field cache for anonymous-method delegates). Once the cache is collapsed, the existing array-initializer transforms recover the array literal, so the <PrivateImplementationDetails> reference disappears. Test: ILPretty/CachedReadOnlySpanInitialization.

Copilot

Pull request overview

This PR addresses a decompilation correctness issue on frameworks without RuntimeHelpers.CreateSpan by collapsing Roslyn’s legacy lazy static cache pattern used for ReadOnlySpan<T> array literals, preventing unexpressible/undeclared <PrivateImplementationDetails> references in the generated C#.

Changes:

Add CachedReadOnlySpanInitialization block transform to collapse the legacy lazy-cache initialization pattern.
Register the new transform immediately after CachedDelegateInitialization in the IL transform pipeline.
Add an ILPretty regression test (CachedReadOnlySpanInitialization) with expected C# output.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
ICSharpCode.Decompiler/IL/Transforms/CachedReadOnlySpanInitialization.cs	New block transform that recognizes and collapses the legacy lazy cache pattern for `ReadOnlySpan<T>` array literals.
ICSharpCode.Decompiler/CSharp/CSharpDecompiler.cs	Inserts the new transform after `CachedDelegateInitialization` in the per-block transform pipeline.
ICSharpCode.Decompiler.Tests/TestCases/ILPretty/CachedReadOnlySpanInitialization.il	New ILPretty input reproducing the legacy cached-array pattern (using `<PrivateImplementationDetails>`).
ICSharpCode.Decompiler.Tests/TestCases/ILPretty/CachedReadOnlySpanInitialization.cs	Expected decompiled C# output fixture for the new ILPretty test.
ICSharpCode.Decompiler.Tests/ILPrettyTestRunner.cs	Adds a new NUnit test entry to run the new ILPretty case.
ICSharpCode.Decompiler.Tests/ICSharpCode.Decompiler.Tests.csproj	Includes the new `.il` test input in the project.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

siegfriedpammer · 2026-07-06T18:21:18Z

@@ -0,0 +1,126 @@
+// Copyright (c) 2024 ICSharpCode


Suggested change

// Copyright (c) 2024 ICSharpCode

// Copyright (c) 2026 Sebastien Lebreton

siegfriedpammer · 2026-07-06T18:32:21Z

+		static bool MatchIsNull(ILInstruction condition, ILVariable v)
+		{
+			// comp(ldloc V == ldnull)
+			if (condition.MatchCompEquals(out var left, out var right) && right.MatchLdNull() && left.MatchLdLoc(v))


the catch-all form to solve this would be a while (MatchLogicNot) loop that unpacks the negation. This is a proven pattern used throughout the codebase. After that use MatchCompEqualsNull to catch left == null and null == right at the same time.

siegfriedpammer · 2026-07-06T18:36:59Z

+		{
+			if (!context.Settings.ArrayInitializers)
+				return;
+			for (int i = context.IndexOfFirstAlreadyTransformedInstruction - 1; i >= 1; i--)


i >= 1 made me raise an eyebrow until I read the pattern further down... might be worth a comment.

siegfriedpammer · 2026-07-06T18:39:28Z

+		static bool DoTransform(Block block, int i, IfInstruction inst, BlockTransformContext context)
+		{
+			// The if must be a simple `if (...) { ... }` (no else) with a two-instruction body.
+			if (!inst.FalseInst.MatchNop() || inst.TrueInst is not Block trueBlock || trueBlock.Instructions.Count != 2)


nit: could use property patterns to make it shorter

siegfriedpammer · 2026-07-06T18:40:04Z

+				return false;
+			// storeBeforeIf: stloc V(ldobj(ldsflda cacheField)), cacheField a compiler-generated static field.
+			if (block.Instructions[i - 1] is not StLoc storeBeforeIf || storeBeforeIf.Value is not LdObj ldobj
+				|| !ldobj.Target.MatchLdsFlda(out var cacheField) || !cacheField.IsCompilerGeneratedOrIsInCompilerGeneratedClass())


should probably be split into multiple sequential if statements, like we do in other transforms, for readability

siegfriedpammer · 2026-07-06T18:45:20Z

+			}
+			// V is assigned exactly twice (before-if load + in-if init) and read exactly three times
+			// (null-check condition + cache write-back + one real downstream usage), with no address-of.
+			if (v.StoreCount != 2 || v.LoadCount != 3 || v.AddressCount != 0)


this check could be moved up, so we don't spend time on matching the full pattern, if this simple check fails at the end.

siegfriedpammer · 2026-07-06T18:46:34Z

 		}

+		[Test]
+		public async Task CachedReadOnlySpanInitialization()


is there a reason why this is not a normal C# Pretty test? ILPretty tests have the disadvantage that they don't detect changes in Roslyn's pattern detection, when we update or add a new compiler version.

siegfriedpammer · 2026-07-06T18:57:36Z

@@ -0,0 +1,55 @@
+.assembly extern mscorlib
+{


How did you generate this IL? It looks like it was generated by an LLM. Please use a C# program as starting point and "convert" it to IL using ildasm, ilspycmd or https://lab.razor.fyi. This way we ensure we actually test against the IL shape Roslyn emits (or emitted at a certain point in time) and the produced IL is generated by a decade old battle-tested tool instead of a "potentially hallucinating DWIM slot machine".

christophwille requested a review from Copilot July 3, 2026 08:58

Copilot started reviewing on behalf of christophwille July 3, 2026 08:58 View session

Copilot AI reviewed Jul 3, 2026

View reviewed changes

Comment thread ICSharpCode.Decompiler/IL/Transforms/CachedReadOnlySpanInitialization.cs Outdated

Comment thread ICSharpCode.Decompiler.Tests/ICSharpCode.Decompiler.Tests.csproj

sailro and others added 2 commits July 3, 2026 11:31

Apply suggestions from code review

778cccd

Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>

Regenerate file lists after transform change

5609214

siegfriedpammer reviewed Jul 6, 2026

View reviewed changes

siegfriedpammer requested changes Jul 6, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

Fix #3850: recover ReadOnlySpan<T> array literals from the legacy lazy cache#3851

Fix #3850: recover ReadOnlySpan<T> array literals from the legacy lazy cache#3851
sailro wants to merge 3 commits into
icsharpcode:masterfrom
sailro:fix-readonlyspan-privateimpldetails-cache

sailro commented Jul 3, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Uh oh!

siegfriedpammer Jul 6, 2026

Uh oh!

siegfriedpammer Jul 6, 2026

Uh oh!

siegfriedpammer Jul 6, 2026

Uh oh!

siegfriedpammer Jul 6, 2026

Uh oh!

siegfriedpammer Jul 6, 2026

Uh oh!

siegfriedpammer Jul 6, 2026

Uh oh!

siegfriedpammer Jul 6, 2026

Uh oh!

siegfriedpammer Jul 6, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

	// Copyright (c) 2024 ICSharpCode
	// Copyright (c) 2026 Sebastien Lebreton

Uh oh!

Uh oh!

Conversation

sailro commented Jul 3, 2026

Fix

Test

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants