Skip to content

Conversation

@jcouv
Copy link
Member

@jcouv jcouv commented May 29, 2025

From discussion with WG, relaxing this restriction allows for greater portability of classic extension methods. But we're keeping the check for members that can't be called with explicit type arguments.
To be confirmed with LDM next week.

New rule: https://github.com/dotnet/csharplang/blob/main/meetings/2025/LDM-2025-06-04.md#extension-declaration-validation

Closes #78472

Relates to test plan #76130

@jcouv jcouv self-assigned this May 29, 2025
@jcouv jcouv added Area-Compilers Feature - Extension Everything The extension everything feature labels May 29, 2025
@jcouv jcouv force-pushed the extensions-inferrability branch from 550cc8e to 8de7718 Compare May 29, 2025 21:26
@jcouv jcouv force-pushed the extensions-inferrability branch from 8de7718 to ace0477 Compare May 29, 2025 21:47
return parameter;
}

static void checkUnderspecifiedGenericExtension(ParameterSymbol parameter, ImmutableArray<TypeParameterSymbol> typeParameters, BindingDiagnosticBag diagnostics)
Copy link
Member Author

@jcouv jcouv May 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

📝 moved this logic to SourceNamedTypeSymbol.AfterMembersCompletedChecks as the added GetMembers call was causing a circularity issue (out-of-date after new LDM decision)

@jcouv jcouv marked this pull request as ready for review May 30, 2025 22:20
@jcouv jcouv requested a review from a team as a code owner May 30, 2025 22:20

void checkUnderspecifiedGenericExtension(ParameterSymbol parameter, ImmutableArray<TypeParameterSymbol> typeParameters, BindingDiagnosticBag diagnostics)
{
if (GetMembers().All((m, a) => m is MethodSymbol { MethodKind: MethodKind.Ordinary }, arg: (object?)null))
Copy link
Contributor

@AlekseyTs AlekseyTs Jun 2, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

All

Is there an overload that doesn't need unused additional argument? Perhaps we have Any instead? #Closed


if (TryGetOrCreateExtensionMarker() is { Parameters: [var extensionParameter] })
{
checkUnderspecifiedGenericExtension(extensionParameter, TypeParameters, diagnostics);
Copy link
Contributor

@AlekseyTs AlekseyTs Jun 2, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

checkUnderspecifiedGenericExtension

I think that long term it would be better to perform this check per member and report an error per member. For example, once we enable indexers, I can imagine that some indexer declarations (referencing remaining type arguments in their parameters) will be allowed, and others won't be allowed. #Closed


void checkUnderspecifiedGenericExtension(ParameterSymbol parameter, ImmutableArray<TypeParameterSymbol> typeParameters, BindingDiagnosticBag diagnostics)
{
if (GetMembers().All((m, a) => m is MethodSymbol { MethodKind: MethodKind.Ordinary }, arg: (object?)null))
Copy link
Contributor

@AlekseyTs AlekseyTs Jun 2, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

MethodKind.Ordinary

I think operators should be allowed too. At least, consider leaving a comment to follow up, the one that starts with a link to the test plan. Operators are being worked on as we speak. BTW, similar to indexers, I think that some operator declarations would be allowed and other operator declarations wouldn't be allowed. This is even stronger reason to moving to per member check/reporting. #Closed

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The criteria we last discussed in WG is whether it's possible to use the member with explicit type arguments.
So I don't think we'll allow either indexers or operators in non-inferrable extension blocks.

FWIW, here's the update I captured in the spec (to be reviewed in LDM this week):

__Inferrability:__ All the type parameters of an extension block must be used in the receiver type when the extension block
contains a non-method member. 
This makes it always possible to infer the type arguments when applied to a receiver of the given receiver type and
the member doesn't allow explicit type arguments.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I do see that in the summary, but, to be honest, I do not remember us having a serious discussion about the criteria to the point that I must have forgotten us deciding on having the restriction. At the same time, from our "Tensors and extension operators" meeting I vaguely remember us expressing a desire to support the following scenario from #78472 (comment):

extension<TTensor, TScalar>(TTensor)
    where TTensor : IReadOnlyTensor<TTensor, TScalar>
    where TScalar : IAdditionOperators<TScalar, TScalar, TScalar>
{
    public static TTensor operator +(TTensor left, TScalar right);
}

Copy link
Contributor

@AlekseyTs AlekseyTs Jun 3, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Here is a link to the specific summary message from Tanner specifically mentioning the following shape as planned for .NET 10:

extension<TTensor, TScalar>(TTensor)
    where TTensor : IReadOnlyTensor<TTensor, TScalar>
    where TScalar : IAdditionOperators<TScalar, TScalar, TScalar>
{
    public static TTensor operator +(TTensor left, TScalar right);
}


var comp = CreateCompilation(src);
comp.VerifyEmitDiagnostics(
// (1,9): error CS9286: 'object' does not contain a definition for 'M' and no accessible extension member 'M' for receiver of type 'object' could be found (are you missing a using directive or an assembly reference?)
Copy link
Contributor

@AlekseyTs AlekseyTs Jun 2, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

// (1,9): error CS9286: 'object' does not contain a definition for 'M' and no accessible extension member 'M' for receiver of type 'object' could be found (are you missing a using directive or an assembly reference?)

Are we tracking an improvement for diagnostics in this scenarios? #Closed

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, we have a general tracking comment in ResolveExtensions: "// Tracked by #76130 : we'll want to describe what went wrong in a useful way (see OverloadResolutionResult.ReportDiagnostics)"
I'll also add one for this specific test

Diagnostic(ErrorCode.ERR_ExtensionResolutionFailed, "object.M").WithArguments("object", "M").WithLocation(4, 19));

src = """
var x = I.M; // binds to I1.M (method)
Copy link
Contributor

@AlekseyTs AlekseyTs Jun 2, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

// binds to I1.M (method)

It would be good to observe the fact. This suggestion is applicable to the other three similar comments as well. #Closed

@AlekseyTs
Copy link
Contributor

AlekseyTs commented Jun 2, 2025

Done with review pass (commit 1) #Closed

Julien Couvreur added 2 commits June 2, 2025 17:15
AlekseyTs
AlekseyTs previously approved these changes Jun 3, 2025
Copy link
Contributor

@AlekseyTs AlekseyTs left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM (commit 3). However, I personally think that the implemented rule is unnecessarily restrictive. Therefore, I suggest delaying the merge until after June 4th LDM, where I hope we finalize the design.

@jcouv jcouv marked this pull request as draft June 4, 2025 21:13
@jcouv jcouv changed the title Extensions: relax inferrability rule for methods Extensions: relax inferrability rule Jun 5, 2025
@jcouv jcouv requested a review from AlekseyTs June 5, 2025 01:16
@jcouv jcouv dismissed AlekseyTs’s stale review June 5, 2025 01:17

Revised based on LDM decision today

@jcouv jcouv marked this pull request as ready for review June 5, 2025 01:20
parameter.Type.VisitType(collectTypeParameters, arg: usedTypeParameters);
}

foreach (var typeParameter in extension.TypeParameters)
Copy link
Contributor

@AlekseyTs AlekseyTs Jun 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

extension.TypeParameters

Consider adding an early return if we have no type parameters #Closed

extensionParameter.Type.VisitType(collectTypeParameters, arg: usedTypeParameters);
foreach (var parameter in parameters)
{
parameter.Type.VisitType(collectTypeParameters, arg: usedTypeParameters);
Copy link
Contributor

@AlekseyTs AlekseyTs Jun 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

parameter.Type.VisitType(collectTypeParameters, arg: usedTypeParameters);

Consider breaking out early if we already recorded extension.TypeParameters.Length type parameters. #Closed

}

var usedTypeParameters = PooledHashSet<TypeParameterSymbol>.GetInstance();
extensionParameter.Type.VisitType(collectTypeParameters, arg: usedTypeParameters);
Copy link
Contributor

@AlekseyTs AlekseyTs Jun 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

extensionParameter.Type.VisitType(collectTypeParameters, arg: usedTypeParameters);

Consider returning early if we already recorded extension.TypeParameters.Length type parameters after this. #Closed

}

var usedTypeParameters = PooledHashSet<TypeParameterSymbol>.GetInstance();
extensionParameter.Type.VisitType(collectTypeParameters, arg: usedTypeParameters);
Copy link
Contributor

@AlekseyTs AlekseyTs Jun 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

extensionParameter.Type.VisitType(collectTypeParameters, arg: usedTypeParameters);

It feels unfortunate that we have to repeat this for every member. #Closed

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Agreed. I could leave an follow-up comment in "optimization/MQ" bucket, but I think I'm fine to leave as-is

@AlekseyTs
Copy link
Contributor

AlekseyTs commented Jun 5, 2025

Done with review pass (commit 5) #Closed

@jaredpar
Copy link
Member

jaredpar commented Jun 5, 2025

@333fred second reviewer


/// <summary>
/// For the type representing an extension declaration, returns the receiver parameter symbol.
/// Note: this may be null even if <see cref="IsExtension"/> is true, in error cases.
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

📝 I'd already added that comment to the corresponding public API, but it seems useful here too

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Think you should add some remarks to indicate that it will not be null in the case of extension(SomeType), just an unnamed parameter.

parameter.Type.VisitType(collectTypeParameters, arg: usedTypeParameters);
}

if (usedTypeParameters.Count == extensionArity && extensionMemberArity == 0)
Copy link
Contributor

@AlekseyTs AlekseyTs Jun 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

if (usedTypeParameters.Count == extensionArity && extensionMemberArity == 0)

I was actually suggesting to do this inside the loop. #Closed

Copy link
Contributor

@AlekseyTs AlekseyTs left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM (commit 6), with an optimization suggestion #78758 (review)

@jcouv
Copy link
Member Author

jcouv commented Jun 5, 2025

@333fred for second review. Thanks

return;
}

var usedTypeParameters = PooledHashSet<TypeParameterSymbol>.GetInstance();
Copy link
Member

@333fred 333fred Jun 5, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does this need some kind of comparison flags? Do we have tests where the type parameter differs by nullability? #Resolved

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does this need some kind of comparison flags? Do we have tests where the type parameter differs by nullability?

We are dealing with declarations, therefore, we don't need to worry about any substitution/alpha renaming going on.


/// <summary>
/// For the type representing an extension declaration, returns the receiver parameter symbol.
/// Note: this may be null even if <see cref="IsExtension"/> is true, in error cases.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Think you should add some remarks to indicate that it will not be null in the case of extension(SomeType), just an unnamed parameter.

</data>
<data name="ERR_UnderspecifiedExtension" xml:space="preserve">
<value>The extended type '{0}' must reference all the type parameters declared by the extension, but type parameter '{1}' is not referenced.</value>
<value>Every type parameter from the extension block must be referenced by the extension parameter or a parameter of this member. But type parameter `{0}` is not referenced.</value>
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this is far too wordy.

Suggested change
<value>Every type parameter from the extension block must be referenced by the extension parameter or a parameter of this member. But type parameter `{0}` is not referenced.</value>
<value>The type parameter `{0}` is not referenced by either the extension parameter or a parameter of this member.</value>

@jcouv jcouv requested a review from 333fred June 6, 2025 14:02
@jcouv jcouv enabled auto-merge (squash) June 6, 2025 17:59
@jcouv jcouv merged commit df87e95 into dotnet:main Jun 6, 2025
23 of 24 checks passed
@dotnet-policy-service dotnet-policy-service bot added this to the Next milestone Jun 6, 2025
@RikkiGibson RikkiGibson modified the milestones: Next, 18.0 P1 Aug 19, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Area-Compilers Feature - Extension Everything The extension everything feature

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Fail to compile C# 14.0 extension with interface constraint using implicit additional type parameters

5 participants