Report nullable ctor warnings for struct primary constructors #74844
Conversation
dotnet-issue-labeler
bot
added
Area-Compilers
untriaged
| if (method is SynthesizedPrimaryConstructor primaryCtor && method.ContainingType.IsStructType()) | ||
| { | ||
| body = BoundBlock.SynthesizedNoLocals(primaryCtor.GetSyntax()); | ||
| nullableInitialState = getInitializerState(body); | ||
| } | ||
| else if (method is SourceMemberMethodSymbol sourceMethod) |
There was a problem hiding this comment.
This path was causing us to miss nullable analyzing the struct primary constructor.
| @@ -639,8 +639,7 @@ void enforceMemberNotNull(SyntaxNode? syntaxOpt, LocalState state) | |||
| Debug.Assert(thisParameter is object); | |||
| thisSlot = GetOrCreateSlot(thisParameter); | |||
| } | |||
| // https://github.com/dotnet/roslyn/issues/46718: give diagnostics on return points, not constructor signature | |||
| var exitLocation = method.DeclaringSyntaxReferences.IsEmpty ? null : method.TryGetFirstLocation(); | |||
| var exitLocation = method is SynthesizedPrimaryConstructor || method.DeclaringSyntaxReferences.IsEmpty ? null : method.TryGetFirstLocation(); | |||
There was a problem hiding this comment.
If a type has a primary constructor, all other instance constructors will be required to chain to it (directly or indirectly), so it will be the only constructor which will get warnings for uninitialized non-nullable instance fields. That means the field/property declaration is really the best place to report diagnostics in such cases.
|
@dotnet/roslyn-compiler for review |
| @@ -1735,12 +1735,7 @@ private static void GetStateMachineSlotDebugInfo( | |||
|
|
|||
| initializersBody ??= GetSynthesizedEmptyBody(method); | |||
|
|
|||
| if (method is SynthesizedPrimaryConstructor primaryCtor && method.ContainingType.IsStructType()) | |||
| { | |||
| body = BoundBlock.SynthesizedNoLocals(primaryCtor.GetSyntax()); | |||
There was a problem hiding this comment.
What replicates this logic now? What body is created for the primary constructor?
There was a problem hiding this comment.
The same logic which was used for class primary constructors is now also used for struct primary constructors. You can see that where the assertions were adjusted.
I reviewed the code around this area and didn't spot anything which would be problematic for structs, but, it's quite possible I missed something.
Closes #74726