Champion "Null-coalescing assignments" (16.3, Core 3)

[MadsTorgersen]

Introduces ??=.

Note: a minor change to ?? is included in this feature as implemented, so that it allows unconstrained type parameter types on the left.

• Proposal added: proposals/null-coalecing-assignment.md
• Discussed in LDM
• Decision in LDM
• Finalized (done, rejected, inactive)
• Spec'ed

LDM history:

• https://github.com/dotnet/csharplang/blob/master/meetings/2019/LDM-2019-04-15.md (result type)

[MgSam]

Glad to see this feature graduated from the "we will never do this!" bucket. :)

[Thaina]

Should we also support &&= and ||= operators?

Yes please

[devinvisible]

Very excited for this! Nice syntactic sugar similar to Ruby's ||=

[paulomorgado]

@Thaina, what are the use cases for &&= and ||=.

[Thaina]

@paulomorgado

var mustAllTrue = Condition0;
mustAllTrue &&= Condition1;
mustAllTrue &&= Condition2;

[CyrusNajmabadi]

@Thaina You can already do that today with:

var mustAllTrue = Condition0;
mustAllTrue &= Condition1;
mustAllTrue &= Condition2;

[yaakov-h]

@CyrusNajmabadi Wouldn't that still unconditionally evaluate the right-hand side? I'd expect in a situation such as:

var val = true;
val &&= BigExpensiveOperationReturningBool();

that BigExpensiveOperationReturningBool is not executed.

[CyrusNajmabadi]

@CyrusNajmabadi Wouldn't that still unconditionally evaluate the right-hand side?

Ah, point taken.

[paulomorgado]

if val is true, that means that it will end up with whatever BigExpensiveOperationReturningBool() returns, right?

Short circuiting would only happen for:

var v1 = false;
v1 &&= BigExpensiveOperationReturningBool();

var v2 = true;
v1 ||= BigExpensiveOperationReturningBool();

I'll agree that move statements to expressions can be more expressive.

[yaakov-h]

@paulomorgado true. 😅

[Spongman]

@MadsTorgersen very glad to see you finally changed your mind on this "contrived" scenario ;)

https://connect.microsoft.com/VisualStudio/Feedback/Details/585580

[MgSam]

@Spongman To be fair, Mads said in that thread he thought ??= might be worthwhile, but ||= and &&= were probably not given the finite budget for features. I agree with him on that.

[alrz]

dotnet/roslyn#13951 and dotnet/roslyn#1276 are also related here.

obj?.Property = value;
obj?.Event += handler;

[ashmind]

Note that I did a prototype for this one a while ago, at ashmind/roslyn@cf5ee99.

It was available on TryRoslyn for a while, but fell too far behind main roslyn branch at the moment -- I'll see if I can merge with latest.

[alrz]

This proposal would allow to null check an expression without assigning it (if permitted),

arg ??= throw new ArgumentNullException(nameof(arg));
// instead of
_ = arg ?? throw new ArgumentNullException(nameof(arg));

It could also be used with return, continue and break expressions.

[ghost]

@Thaina

Yes please

It is not an opinion... just text...

Yes it should be implemented but guys please consider real discussion provided here: #397

[Thaina]

@Shawn-Twinmaple We could do this

void DoStuff() => this.Data = this.Data ?? this.device?.ReadDataFromStream();

[jnm2]

@Thaina The shorthand for that will be ??=. Shawn is asking for a shorthand where this.Data is set to this.device?.ReadDataFromStream() if the latter is not null, regardless of whether this.Data was previously null.

[Thaina]

@jnm2

void DoStuff() => this.Data = this.device?.ReadDataFromStream() ?? this.Data;

?

[jnm2]

@Thaina Sort of, but not quite. In your example, you end up setting this.Data = this.Data if ReadDataFromStream returns null, but this can have side effects.

If you read Shawn's comment, he has the exact code you're looking for.

[Thaina]

@jnm2 I just think he just want to make it shorter to write. And so if it just a field I think it would have no side effect

If we need to avoid side effect for property too, well, maybe we need another feature request I have seen about allow putting expression in operator

void DoStuff() => this.Data = this.device?.ReadDataFromStream() ?? return;

[jnm2]

@Thaina Yes, that's exactly what he (and I) have just said. He wants it to be shorter to write.

I really want that ?? return feature (it just keeps coming up for me for other reasons) but I think it's already been turned down by the LDT (link). Also, it's not general enough to solve what Shawn is asking for. Imagine three of these statements in a row. You can't use ?? return more than once per method.

[Shawn-Twinmaple]

@Shawn-Twinmaple We could do this

void DoStuff() => this.Data = this.device?.ReadDataFromStream() ?? this.Data;

Yep exactly.

That become this:
void DoStuff() => this.Data ?!= this.device?.ReadDataFromStream();
In this way, the need to re-assign to itself is simply short-circuited away.

Since both ! and != are already implemented negation operators I'm not really sure which "operator symbology" would work best for this.

Symbology aside
??= is great for null coalescing assignments, but not-null assignments are also a problem too, for the same reason.

Ideally there would be a solution both for null and not-null assignments, as they are both very common place occurrences in code found everywhere.

The proposed solution must work with nullable valuetypes and nullable reference types.
It must also work with non-nullable value types and non-nullable reference types.

I have attached some test examples that I believe covers all 4 of those cases for both proposed operators...
160 lines of code, far too much to post here.
Comments, Problems and Solutions for each case are all in the file.
test_examples.zip

I think the two operators really go hand-in-hand just like == and != does today;
but as suggested, this might need to be in its own feature request...

[yaakov-h]

LDM Update: The result of the ??= expression will be the result of an implicit conversion from the RHS to the underlying type of the LHS, if one exists.

That means that #34 (comment) will happen (and hey, it looks like my exact example was used in the LDM 😄)

[jnm2]

@333fred and team, you're the best. Thanks for this!

[gafter]

See https://github.com/dotnet/csharplang/blob/master/meetings/2019/LDM-2019-04-15.md#result-type-of--expression

[PathogenDavid]

@gafter I think you hit the close+comment button by mistake.

[wanton7]

Does this feature mean I can finally use (list ??= new List<int>()).Add(7); instead of (list = list ?? new List<int>()).Add(7); in C# 8.0?

[yaakov-h]

Yes.

[333fred]

@jnm2 @yaakov-h dotnet/roslyn#36329 and #2591. We're hoping that this change will be in the final version of 16.2 or some preview before then.

[yaakov-h]

@333fred yay, thanks!

[ufcpp]

Is it possible to make LHS of the null-coalescing assignments by-ref?

using System;

class Program
{
    static void Main()
    {
        string[] values = new string[1];

        // OK
        ref var r1 = ref values[0];
        r1 ??= "abc";
        Console.WriteLine(values[0]);

        // NG
        ref var r2 = ref values[0] ??= "default string";
    }
}

[CoderVision]

I love the idea of the ??= feature! I have been periodically checking updated compilers to see if it is included and was happy to see that it has a preview status.

[333fred]

@ufcpp no compound assignment works that way, and it would conflict with the nullable-value-type change that was just implemented.

[jnm2]

@yaakov-h at #34 (comment):

LDM Update: The result of the ??= expression will be the result of an implicit conversion from the RHS to the underlying type of the LHS, if one exists.

That means that #34 (comment) will happen (and hey, it looks like my exact example was used in the LDM 😄)

@333fred This change made a real difference to me today. Thanks again!

private bool? foo;

public bool Foo => foo ??= Properties.GetDataSourceRowByKeyValue(EditValue) is null;