Extensions: ref safety analysis #77967

jcouv · 2025-04-02T11:08:19Z

Analyze the receiver for an extension as an argument.
Properties are already analyzed as invocations.

Relates to test plan #76130

jcouv · 2025-04-04T06:35:23Z

@AlekseyTs @jjonescz for review. Thanks

src/Compilers/CSharp/Portable/Binder/Binder.ValueChecks.cs

jjonescz · 2025-04-07T09:09:56Z

src/Compilers/CSharp/Portable/Binder/Binder.ValueChecks.cs

+
+            if (!argRefKindsOpt.IsDefault)
+            {
+                argRefKindsOpt = [RefKind.None, .. argRefKindsOpt];


Why RefKind.None and not extensionParameter.RefKind? #Resolved

Because the argument is passed without an explicit ref kind, just like the receiver of a classic extension method.

AlekseyTs · 2025-04-07T16:25:24Z

src/Compilers/CSharp/Portable/Binder/Binder.ValueChecks.cs

@@ -1971,9 +2000,12 @@ bool isRefEscape
            Debug.Assert(AllParametersConsideredInEscapeAnalysisHaveArguments(argsOpt, parameters, argsToParamsOpt));
 #endif

+            MethodInfo localMethodInfo = methodInfo;


methodInfo

Is there a specific reason why we keep the original value accessible in the rest of the method through this parameter? This might be bug prone in the long run. #Resolved

Resolved offline

AlekseyTs · 2025-04-07T20:50:27Z

src/Compilers/CSharp/Portable/Binder/Binder.ValueChecks.cs

@@ -2128,9 +2204,12 @@ bool isRefEscape
            Debug.Assert(AllParametersConsideredInEscapeAnalysisHaveArguments(argsOpt, parameters, argsToParamsOpt));
 #endif

+            MethodInfo localMethodInfo = methodInfo;


methodInfo

Similar comment as in GetInvocationEscapeScope #Resolved

AlekseyTs · 2025-04-07T20:51:47Z

src/Compilers/CSharp/Portable/Binder/Binder.ValueChecks.cs

@@ -2253,9 +2332,9 @@ private bool CheckInvocationEscapeWithUpdatedRules(
                        // For consistency with C#10 implementation, we don't report an additional error
                        // for the receiver. (In both implementations, the call to Check*Escape() above
                        // will have reported a specific escape error for the receiver though.)
-                        if ((object)((argument as BoundCapturedReceiverPlaceholder)?.Receiver ?? argument) != receiver && symbol is not SignatureOnlyMethodSymbol)
+                        if ((object)((argument as BoundCapturedReceiverPlaceholder)?.Receiver ?? argument) != receiver && (Symbol?)methodInfo.Symbol is not SignatureOnlyMethodSymbol)


(Symbol?)

Is this cast necessary? #Closed

AlekseyTs · 2025-04-07T20:52:40Z

src/Compilers/CSharp/Portable/Binder/Binder.ValueChecks.cs

@@ -2778,17 +2857,20 @@ private bool CheckInvocationArgMixing(
            SafeContext localScopeDepth,
            BindingDiagnosticBag diagnostics)
        {
+            MethodInfo localMethodInfo = methodInfo;


methodInfo

Similar comment as in GetInvocationEscapeScope #Resolved

AlekseyTs

LGTM (commit 7)

AlekseyTs · 2025-04-07T21:44:55Z

src/Compilers/CSharp/Portable/Binder/Binder.ValueChecks.cs

@@ -1971,77 +2000,84 @@ bool isRefEscape
            Debug.Assert(AllParametersConsideredInEscapeAnalysisHaveArguments(argsOpt, parameters, argsToParamsOpt));
 #endif

+            MethodInfo localMethodInfo = ReplaceWithExtensionImplementationIfNeeded(methodInfo, ref parameters, ref receiver, ref argsOpt, ref argRefKindsOpt, ref argsToParamsOpt);
+
            if (methodInfo.UseUpdatedEscapeRules)


if (methodInfo.UseUpdatedEscapeRules)

Should this check be moved into the local function?

I prefer to leave this structure/pattern (if (newRules) handleNewRules() else handleOldRules())

I prefer to leave this structure/pattern (if (newRules) handleNewRules() else handleOldRules())

I wasn't suggesting to change that, and I do not think moving the if into the local function would change that.

* Addressing feedback * Addressing PR feedback * Addressing PR comments * Remove workaround for dotnet/msbuild#10306 The VS SDK was fixed, so the problem shouldn't occur anymore. * move to a workspace factory * No need for solution * Simplify handlers * Add helper * move more over * move more over * Add remove calls * Move handlers over * Move remote service over * async lock * async lock * async lock * More work * Fixup * Correct * Move to immutable dictionary * in progress * in progress * in progress * in progress * in progress * in progress * Simplify * Serialize * simplify * simplify * simplify * in progress * in progress * Reset code * Clelar * renames * Simplify * rename * No more null * Simplify * Exception throwing * Extensions: update feature status (#77923) * Add ignored directives feature status (#77965) * Extensions: misc checks on receiver parameter and extension members (#77937) * Add porject back in * CHange how locking works * Simplify * Docs * Add new message for getting registered method names * IN progres * in progress * Simplify * Simplify * Handle exceptions * Simplify * Docs * move method * Simplify and regions * Tweaks * Docs * Break into files * Simplify exception handling * Simplify * Simplify * Move * Docs * Simplify * Simplify * Share code * Split into local and remote calls * Add throw * Add throw * Simplify * Simplify * Simplify * Simplify * Simplify * Simplify * Simplify * Simplify * Simplify * Simplify * Simplify * Simplify * Avoid potential disposal exception during MEF shutdown (#77990) Previously, if VS was shutdown before Workspace.EnsureEventListeners was invoked, we would cause an exception during MEF disposal. This would happen because our workspace disposal would attempt to get a service that hadn't already been cached, and thus would ask MEF to compose the item during MEF disposal (which doesn't work). Instead, just cache the IWorkspaceEventListenerService for use during the dispose. * Remove unused elements form PublishData.json (#77994) Removing some elements that seem to be left over from how the publish tooling used to work * Reduce main thread switches in VisualStudioProjectFactory.CreateAndAddToWorkspaceAsync (#77920) * Reduce main thread switches in VisualStudioProjectFactory.CreateAndAddToWorkspaceAsync This method is called once for each project in the solution during solution open, and each call was essentially switching to the main thread twice (once explicitly, once via call to _visualStudioWorkspaceImpl.RefreshProjectExistsUIContextForLanguageAsync). For my testing of opening Roslyn.sln, this removed over 600 main thread switches. Instead: 1) Move the first main thread switch to just occur once in an initialization task. This task is joined when CreateAndAddToWorkspaceAsync is invoked to ensure that that initialization is completed. 2) Move the second main thread switch to instead move inside the callback from an ABWQ. The code being called needs the UI thread, but doesn't need to occur synchronously during the call to CreateAndAddToWorkspaceAsync. * Explicitly releasE * Cancellation is fine * Reorder * remove net * docs * docs * remove exception handling * Add public api * lint * Reduce main thread switches in VisualStudioProjectFactory.CreateAndAddToWorkspaceAsync (#77920) (#77998) * Reduce main thread switches in VisualStudioProjectFactory.CreateAndAddToWorkspaceAsync This method is called once for each project in the solution during solution open, and each call was essentially switching to the main thread twice (once explicitly, once via call to _visualStudioWorkspaceImpl.RefreshProjectExistsUIContextForLanguageAsync). For my testing of opening Roslyn.sln, this removed over 600 main thread switches. Instead: 1) Move the first main thread switch to just occur once in an initialization task. This task is joined when CreateAndAddToWorkspaceAsync is invoked to ensure that that initialization is completed. 2) Move the second main thread switch to instead move inside the callback from an ABWQ. The code being called needs the UI thread, but doesn't need to occur synchronously during the call to CreateAndAddToWorkspaceAsync. * make nested * Simplify * Simplify * Simplify * Simplify * Simplify * restore * Update dependencies from https://github.com/dotnet/source-build-reference-packages build 20250402.2 (#77982) Microsoft.SourceBuild.Intermediate.source-build-reference-packages From Version 10.0.618101 -> To Version 10.0.620202 Co-authored-by: dotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com> * Allow null * Simplify * No VoidResult * Make everything non-capturing * 'nameof(T<>)' not supported yet. * Update references * Revert "Another attempt to move MiscellaneousFilesWorkspace to a bg thread (#77983) This reverts commit 2e8fcbe. * fix * Exception work * Update Roslyn.sln * More exception work * Add remote end * Complete the exception handling work * Add docs * docs * Add docs * Rename type * Rename type * update publishdata * Change SolutionState.ProjectStates from an ImmutableDictionary<ProjectId, ProjectState> to an ImmutableArray<ProjectState> (#77971) * *** WIP. Definitely needs supporting speedometer numbers before being considered *** Change SolutionState to keep/expose and ImmutableArray<ProjectState> instead of an IReadOnlyList<ProjectId> By exposing the project states, there is a bunch of code that can be changed to enumerate over that collection rather than over the values of an ImmutableDictionary<ProjectId, ProjectState>. That enumeration is actually *very* expensive. For example, during roslyn load, local etl traces indicates the following CPU usage enumerating this dictionary: Under ComputeDocumentIdsWithFilePath: 2.6 seconds Under GetFirstRelatedDocumentId: 1.1 seconds There is a bit of a tradeoff associated with this change. The IReadOnlyList<ProjectId> could be reused very often, whereas there is much less opportunity for sharing ImmutableArray<ProjectState> when forking the solution state. I expect there to be more memory allocated in the speedometer runs, so I'd like to get a feel on how much worse that is before elevating this PR out of draft status. Additionally, SolutionState had a CWT over a sorted version of this IReadOnlyList<ProjectId>. As mentioned earlier, the sharing of this data is quite a bit less common now that it's an ImmutableArray<ProjectState>. I locally measured the stopwatch time spent always sorting this data in ComputeChecksumsAsync and it was a total of 4 ms (averaged over two separate VS runs) when loading the roslyn sln. The ImmutableDictionary is still present in SolutionState, but it was renamed to indicate that it's primarily useful as a lookup tool, and all references treating it as an enumeration mechanism were changed to instead use the ImmutableArray<ProjectState>. * Update dependencies from https://github.com/dotnet/source-build-reference-packages build 20250404.2 (#78008) Microsoft.SourceBuild.Intermediate.source-build-reference-packages From Version 10.0.620202 -> To Version 10.0.620402 Co-authored-by: dotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com> * Fix VSTypeScript lsp tests * localize servicehub service name * Simplify workspace initialization in the LSP server Right now we have the logic for which analyzers get added to a workspace in Program.cs, which calls a special method that specifically initializes the main host workspace. This refactors it so that could be used to initialize any workspace, and removes any tricky ordering problems that might happen by using cleaner MEF imports when we have them. * Update test * Update dependencies from https://github.com/dotnet/arcade build 20250404.5 Microsoft.SourceBuild.Intermediate.arcade , Microsoft.DotNet.Arcade.Sdk , Microsoft.DotNet.Helix.Sdk , Microsoft.DotNet.XliffTasks From Version 9.0.0-beta.25164.2 -> To Version 9.0.0-beta.25204.5 * Update maintenance-packages versions (#78005) * Update maintenance-packages versions * Update dependencies from https://github.com/dotnet/source-build-reference-packages build 20250404.2 Microsoft.SourceBuild.Intermediate.source-build-reference-packages From Version 10.0.620202 -> To Version 10.0.620402 * Address feedback --------- Co-authored-by: dotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com> * Fix EA layering for Razor.ExternalAccess (#77927) Moves things to /Shared to ship in both MS.CA.EA.Razor and MS.CA.EA.Razor.Features At some point we can rename MS.CA.EA.Razor to MS.CA.EA.Razor.EditorFeatures but I didn't bother with that in this PR Shipping in both to not require dual insertion. We can fix this after a consuming change in Razor is inserted * Delete * Update PublishData.json after VS snap to rel/d17.14 * Handle nameof(indexer) in ref analysis (#78015) * Make a couple of features non-experimental * Ensure external access extensions package gets codebase * Feedback * Add extension message handler service teest * Add extension message handler service test * FIx * Add testS * In progress * Test work * Tests * Add test * More tests * Add tests * Add tests * Add tests * Add test * Add test * Add test * Add test * Add test * Update comment * Disable TransitiveVersioningPinning for RoslynAnalyzers. * Remove dependency on EditorFeatures from lsp tests * Remove dependency on EditorFeatures from codelens layer * Update dependencies from https://github.com/dotnet/source-build-reference-packages build 20250407.2 Microsoft.SourceBuild.Intermediate.source-build-reference-packages From Version 10.0.620402 -> To Version 10.0.620702 * Remove dependency on EditorFeatures from build tools * Update dependency versions * Extensions: ref safety analysis (#77967) * Update PublishData.json after VS span to main * In progress * Fix crash in backing field nullability cycle scenario (#77993) * More work * Pass along when an extension was unloaded * Add teset * Add tests * Docs * Simplify * Using alias * Fix * Add tests * Add docs * Update gladstone/roslyn api * Simplify channels in SemanticSearch * Semantic Search: Add support for async queries and FAR tool (#77906) * Simplify further * Simplify further * Localize exception messages * Docs * Docs * Docs * named args * named args * Docs * Move type * Docs * Fix name * Compile just for NET * Unseal LSP types (#78041) * Simplify analyzer api * Update RoslynAnalyzer package projects with dependencies This will fix issues with transitive pinning during source builds. * Ensure LSP uses actual signature help trigger characters * Improve detection of code whose updates may not have effect (#78009) * Do not return metadata names for document symbols * Fix nullability warnings * Review feedback * Remove EditorFeatures from OOP (#78069) * Add back EA.Razor for servicing branches publishdata is pulled from main always. this needs to be present even though the package no longer exists for servicing branches * Remove unused ISemanticSearchWorkspaceHost (#78083) * Split query execution into compile and execute calls (#78081) * Update resourceManagement.yml (#77948) * Expression trees: support optional arguments and named arguments in parameter order (#77972) * Update ignored directives public API (#77968) * Remove experiment * Add Content token to IgnoredDirectiveTriviaSyntax * Add Content token to ShebangDirectiveTriviaSyntax * Fixup public API * Fixup semantic search API list * Fix CLI flag in error message * Remove Update API * Update SemanticSearch API list --------- Co-authored-by: Matteo Prosperi <[email protected]> Co-authored-by: Jason Malinowski <[email protected]> Co-authored-by: Cyrus Najmabadi <[email protected]> Co-authored-by: Cyrus Najmabadi <[email protected]> Co-authored-by: Julien Couvreur <[email protected]> Co-authored-by: Jan Jones <[email protected]> Co-authored-by: Matteo Prosperi <[email protected]> Co-authored-by: Todd Grunke <[email protected]> Co-authored-by: Jared Parsons <[email protected]> Co-authored-by: dotnet-maestro[bot] <42748379+dotnet-maestro[bot]@users.noreply.github.com> Co-authored-by: dotnet-maestro[bot] <dotnet-maestro[bot]@users.noreply.github.com> Co-authored-by: David Barbet <[email protected]> Co-authored-by: Carlos Sánchez López <[email protected]> Co-authored-by: Andrew Hall <[email protected]> Co-authored-by: Charles Stoner <[email protected]> Co-authored-by: Tomas Matousek <[email protected]> Co-authored-by: Joey Robichaud <[email protected]> Co-authored-by: Rikki Gibson <[email protected]> Co-authored-by: Tomáš Matoušek <[email protected]>

jcouv added Area-Compilers Feature - Extension Everything The extension everything feature labels Apr 2, 2025

jcouv self-assigned this Apr 2, 2025

ghost added the untriaged Issues and PRs which have not yet been triaged by a lead label Apr 2, 2025

jcouv force-pushed the extensions-escape branch from 9867564 to da69bef Compare April 2, 2025 13:14

Extensions: ref safety analysis

2b0180d

jcouv force-pushed the extensions-escape branch from da69bef to 2b0180d Compare April 3, 2025 11:06

jcouv marked this pull request as ready for review April 3, 2025 12:56

jcouv requested a review from a team as a code owner April 3, 2025 12:56

jcouv requested review from jjonescz and AlekseyTs April 3, 2025 12:57

jcouv marked this pull request as draft April 3, 2025 14:30

jcouv added 3 commits April 3, 2025 19:35

Revert code change

4e6ab4a

Alternate approach

eb391b9

Comment

7cdc808

jcouv marked this pull request as ready for review April 4, 2025 06:34

jjonescz reviewed Apr 4, 2025

View reviewed changes

Address feedback

87979cc

jcouv marked this pull request as draft April 7, 2025 07:08

jcouv added 2 commits April 7, 2025 09:28

Skip if static. Use new parameters for clarity

a49e052

Re-use parameters

56934ed

jcouv marked this pull request as ready for review April 7, 2025 08:46

jjonescz reviewed Apr 7, 2025

View reviewed changes

jjonescz approved these changes Apr 7, 2025

View reviewed changes

AlekseyTs reviewed Apr 7, 2025

View reviewed changes

AlekseyTs approved these changes Apr 7, 2025

View reviewed changes

Address feedback

e40d3f8

AlekseyTs reviewed Apr 7, 2025

View reviewed changes

jcouv merged commit 256273c into dotnet:main Apr 8, 2025
24 checks passed

jcouv deleted the extensions-escape branch April 8, 2025 07:06

dotnet-policy-service bot added this to the Next milestone Apr 8, 2025

This was referenced Apr 19, 2025

[Automated] PRs inserted in VS build main-10518.29 #78209

Closed

[Automated] PRs inserted in VS build feature.debugger.main-10521.160 #78252

Closed

dotnet-bot mentioned this pull request May 31, 2025

[Automated] PRs inserted in VS build feature.telemetry.d18OpenTelemetry-10630.96 #78782

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Extensions: ref safety analysis #77967

Extensions: ref safety analysis #77967

Uh oh!

jcouv commented Apr 2, 2025

Uh oh!

jcouv commented Apr 4, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jjonescz Apr 7, 2025 •

edited by jcouv

Loading

Uh oh!

jcouv Apr 7, 2025

Uh oh!

AlekseyTs Apr 7, 2025 •

edited by jcouv

Loading

Uh oh!

jcouv Apr 7, 2025

Uh oh!

AlekseyTs Apr 7, 2025 •

edited by jcouv

Loading

Uh oh!

AlekseyTs Apr 7, 2025 •

edited

Loading

Uh oh!

AlekseyTs Apr 7, 2025 •

edited by jcouv

Loading

Uh oh!

AlekseyTs left a comment

Uh oh!

AlekseyTs Apr 7, 2025 •

edited

Loading

Uh oh!

jcouv Apr 8, 2025

Uh oh!

AlekseyTs Apr 8, 2025

Uh oh!

Uh oh!

Uh oh!

Extensions: ref safety analysis #77967

Extensions: ref safety analysis #77967

Uh oh!

Conversation

jcouv commented Apr 2, 2025

Uh oh!

jcouv commented Apr 4, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jjonescz Apr 7, 2025 • edited by jcouv Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jcouv Apr 7, 2025

Choose a reason for hiding this comment

Uh oh!

AlekseyTs Apr 7, 2025 • edited by jcouv Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jcouv Apr 7, 2025

Choose a reason for hiding this comment

Uh oh!

AlekseyTs Apr 7, 2025 • edited by jcouv Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

AlekseyTs Apr 7, 2025 • 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!

AlekseyTs Apr 7, 2025 • edited by jcouv Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

AlekseyTs left a comment

Choose a reason for hiding this comment

Uh oh!

AlekseyTs Apr 7, 2025 • 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!

jcouv Apr 8, 2025

Choose a reason for hiding this comment

Uh oh!

AlekseyTs Apr 8, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

jjonescz Apr 7, 2025 •

edited by jcouv

Loading

AlekseyTs Apr 7, 2025 •

edited by jcouv

Loading

AlekseyTs Apr 7, 2025 •

edited by jcouv

Loading

AlekseyTs Apr 7, 2025 •

edited

Loading

AlekseyTs Apr 7, 2025 •

edited by jcouv

Loading

AlekseyTs Apr 7, 2025 •

edited

Loading