Implement direct delegates by kerams · Pull Request #19993 · dotnet/fsharp

kerams · 2026-06-24T18:26:47Z

Description

Implements fsharp/fslang-suggestions#1083, fixes #11898. RFC.

Test cases should cover everything mentioned in the language suggestion.

Checklist

Test cases added
Performance benchmarks added in case of performance changes
Release notes entry updated:

github-actions · 2026-06-24T18:27:28Z

❗ Release notes required

You can open this PR in browser to add release notes: open in github.dev

✅ Found changes and release notes in following paths:

Change path Release notes path Description

src/Compiler docs/release-notes/.FSharp.Compiler.Service/11.0.100.md

LanguageFeatures.fsi docs/release-notes/.Language/preview.md

github-actions · 2026-06-24T19:06:11Z

🔍 Tooling Safety Check — Affects-Compiler-Output
Affects-Compiler-Output: modifies IlxGen.fs IL emission for delegate expressions

Generated by PR Tooling Safety Check · opus46 4.6M · ◷

T-Gro · 2026-06-25T11:27:20Z

+
+module DirectDelegates =
+
+    let private coreOptions compilation =


Could you pls also add a call to the runtime's IL verifier, to make sure we produce valid IL ?

Example here, there are more helpers for it.

fsharp/tests/FSharp.Compiler.ComponentTests/EmittedIL/RealInternalSignature/ClassTypeInitialization.fs

Lines 46 to 49 in f8da10d

|> verifyPEFileWithSystemDlls

|> withOutputContainsAllInOrderWithWildcards [

"All Classes and Methods in*SimpleTypesInNamespace.exe Verified."

]

T-Gro · 2026-06-25T11:33:28Z

@@ -0,0 +1,417 @@
+


Unless there are valid , by-design (or by test-case) reasons for deviations across optimize/realsig switches, it is better to have a single .bsl file covering more runs.

That way the tests hold equal codegen for both versions together, and guarantee it cannot deviate.

(the test framework allows for that - if e.g. .RealInternalSignature{b}. part is missing, it takes the common name.

There are deviations depending on language version and optimizations. Therefore we have 4 snapshots of everything and can prove the feature is on under specific circumstances only. I suppose I can remove the signature switching though.

But honestly, the language feature clearly gates this, so if you think there is no point asserting the non-preview behavior, I'll happily remove half of the baselines.

T-Gro · 2026-06-25T11:42:21Z

+            not cenv.options.localOptimizationsEnabled
+            && tmvs |> List.exists (fun (v: Val) -> not v.IsCompilerGenerated)
+        then
+            // Eta-expanded delegate in an unoptimized build: the Invoke parameters are the user's own lambda
+            // variables, so keep the closure to preserve their names for debugging. Non-eta delegates (whose
+            // parameters are synthesized by BuildNewDelegateExpr) are always made direct; eta-expanded ones
+            // are made direct only in optimized builds, where debuggability is not a concern and the
+            // forwarding call survives to here.
+            None
+        else


If we are sure we are not losing any sequence points for any forms of lambdas, maybe the new stepping test helpers from @auduchinok could be used to guard that behavior.

Yes, None here causes a fallback to existing behavior, completely unchanged.

T-Gro · 2026-06-25T11:44:44Z

+                            hasWitnesses
+                            || newobj
+                            || isSuperInit
+                            || isSelfInit
+                            || isBaseCall
+                            || not (receiverShapeOk takesInstanceArg)
+                            || receiverNotBindable ()


I would like if this block of code has better explanation and maybe even encapsulation if possible (similar one is also for ILMethod ). To make sure future extensions of runtime/language know when to extend this block.

T-Gro · 2026-06-25T11:50:56Z

+
+open System
+
+// Section D, bullet 1: the optimizer inlines small function bodies before IlxGen runs. If a delegate


Another dimension: Inlining across F# assembly boundaries

T-Gro · 2026-06-25T11:51:23Z

+// 3. eta-expanded known target, tupled application, same compiled representation
+let case3_etaTupled () = Action<int, int>(fun a b -> handlerTupled (a, b))
+
+// 15. non-eta-expanded known target via partial application


I'll make sure the numbering is unified with the RFC cases

T-Gro · 2026-06-25T11:51:31Z

+
+open System
+
+// Section D, bullet 1: the optimizer inlines small function bodies before IlxGen runs. If a delegate


T-Gro · 2026-06-25T11:52:04Z

+
+open System
+
+// Section D, bullet 2: cases 15/16 (in DelegateKnownFunction.fs / DelegateStaticMethod.fs) capture a


Clean the development-time pointers/indices please.

T-Gro · 2026-06-25T11:55:01Z

+        |> coreOptions
+        |> withLangVersionPreview
+        |> withPreviewBaseline
+        |> compile


At least a representative for each positive case should be also running, not just compile.

T-Gro · 2026-06-25T11:56:33Z

+        |> coreOptions
+        |> compile
+        |> shouldSucceed
+        |> verifyILBaseline


One more thought - for the positive scenarios, can we be more specific and instead of .bsl do a pair of "contains" (function pointer load) and "not contains" (no closure allocation).

ldftn/ldvirtftn are present in both cases, along with putting arguments on the stack. The latter also requires a dup instruction. If your goal is to minimize potential future baseline churn, I'm not sure this would help.

Yep, the goal would to minimize the chance of an unrelated change affecting the .bsl.
But the code samples are well isolated, so its not a big deal.

…est case numbering

kerams · 2026-06-26T16:49:30Z

Ready for another pass, think I addressed everything. (But alternatives and unresolved questions in the RFC need answers still:))

smoothdeveloper · 2026-06-26T18:25:08Z

Add an opt-out compiler flag (default on) alongside the language feature.

I think this is not needed anymore, Tomas has added a msbuild way as well as compiler flag to disable specific language features:

--disableLanguageFeature:DirectDelegateConstruction

or

 <ItemGroup>
  <DisabledLanguageFeatures Include="DirectDelegateConstruction" />
</ItemGroup>

You may want to check it picks up as is to allow selectively disabling the feature.

T-Gro · 2026-06-29T09:38:52Z

    <Compile Include="EmittedIL\CompiledNameAttribute\CompiledNameAttribute.fs" />
    <Compile Include="EmittedIL\ComputationExpressions\ComputationExpressions.fs" />
    <Compile Include="EmittedIL\ComputedCollections\ComputedCollections.fs" />
+    <Compile Include="EmittedIL\DirectDelegates\DirectDelegates.fs" />


Since the use cases are known in the RFC and often framework-integrated, it would be great to have some smoke tests for this integration.

(ASP.NET routing/minimal APIs, dependency-injection containers, mocking libraries, serializers, and logging/diagnostics that print the bound method name. With F# they see Invoke on a closure rather than the intended method, which breaks or degrades these scenarios..)

Either going via "Regression tests" suite and referencing a separate repo and building it with the fresh compiler.
Or picking a few cases and doing an .fsx test with #load of a few chosen libraries?

T-Gro · 2026-07-03T09:28:29Z

+    && staticLeadingArgIsRefType g takesInstanceArg leadingArgs
+    && receiverNotByref g leadingArgs
+
+/// Confirm the target's IL signature is compatible with the delegate's Invoke. The type checker has already


Can you pls globally invoke the https://github.com/dotnet/fsharp/blob/main/.github/instructions/NoBloat.instructions.md instruction to shrink the number and verbosity of code comments added?

T-Gro · 2026-07-03T09:31:12Z

+            None
+        else
+            match classifyForwardingTarget (Optimizer.ExprHasEffect Optimizer.EffectContext.Emit) g tmvs body with
+            | DirectDelegateForwardingTargetCandidate.FSharpVal(vref, valUseFlags, tyargs, leadingArgs) ->


So we are now replicating the direct delegate in IlxGen as well as Optimizer, and that prevents the inline race? Or am I reading it wrong?

The recognizer (the piece of code that decides whether direct delegates can be used) runs in both places, but in the optimizer the only thing it does is suppress inlining.

kerams added 3 commits June 24, 2026 00:39

Add language feature and baselines tests

baf99e8

Implementation

87cff9a

Merge branch 'main' of https://github.com/dotnet/fsharp into del

6682524

kerams requested a review from a team as a code owner June 24, 2026 18:26

github-project-automation Bot added this to F# Compiler and Tooling Jun 24, 2026

github-project-automation Bot moved this to New in F# Compiler and Tooling Jun 24, 2026

kerams commented Jun 24, 2026

View reviewed changes

Comment thread src/Compiler/CodeGen/IlxGen.fs

github-actions Bot added the ⚠️ Affects-Compiler-Output Tooling check: PR touches IL emission or codegen label Jun 24, 2026

Release notes, custom delegate tests

894f6f0

T-Gro reviewed Jun 25, 2026

View reviewed changes

kerams added 4 commits June 25, 2026 18:58

Merge branch 'main' of https://github.com/dotnet/fsharp into del

67b1ca7

verifyPEFileWithSystemDlls, remove realsig keying, RFC <-> baseline t…

4270507

…est case numbering

Remove duplicate test cases

2d60fe4

Refactor

dc32fcd

Build fix

8017ff0

Build fix

cd6e3c6

T-Gro reviewed Jun 29, 2026

View reviewed changes

kerams added 15 commits June 29, 2026 14:34

Merge branch 'main' of https://github.com/dotnet/fsharp into del

adc9b61

Unit arg

be7517d

Extension methods

80a8668

Merge branch 'main' of https://github.com/dotnet/fsharp into del

f2f8999

Value types

9ddc0f1

Merge branch 'main' of https://github.com/dotnet/fsharp into del

d8f11c6

Reformat

1f6ea38

ASP.NET smoke test

fa4c43b

Merge branch 'main' of https://github.com/dotnet/fsharp into del

e96523d

Merge branch 'main' of https://github.com/dotnet/fsharp into del

8127868

Merge branch 'main' of https://github.com/dotnet/fsharp into del

49ba6f0

Merge branch 'main' of https://github.com/dotnet/fsharp into del

05a3970

Fix inline race

0e5d1ce

Merge branch 'main' of https://github.com/dotnet/fsharp into del

0410172

Post merge fixes

f485b25

T-Gro reviewed Jul 3, 2026

View reviewed changes

kerams added 2 commits July 3, 2026 18:49

Comments

2b9d108

Merge branch 'main' of https://github.com/dotnet/fsharp into del

69a1c61


		module DirectDelegates =

		let private coreOptions compilation =

	\|> verifyPEFileWithSystemDlls
	\|> withOutputContainsAllInOrderWithWildcards [
	"All Classes and Methods in*SimpleTypesInNamespace.exe Verified."
	]


		open System

		// Section D, bullet 1: the optimizer inlines small function bodies before IlxGen runs. If a delegate


		open System

		// Section D, bullet 2: cases 15/16 (in DelegateKnownFunction.fs / DelegateStaticMethod.fs) capture a

Uh oh!

Conversation

kerams commented Jun 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Checklist

Uh oh!

github-actions Bot commented Jun 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

❗ Release notes required

Uh oh!

Uh oh!

github-actions Bot commented Jun 24, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kerams Jun 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

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!

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!

kerams commented Jun 26, 2026

Uh oh!

smoothdeveloper commented Jun 26, 2026

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

kerams commented Jun 24, 2026 •

edited

Loading

github-actions Bot commented Jun 24, 2026 •

edited

Loading

kerams Jun 25, 2026 •

edited

Loading