dotnet/csharplang

The official repo for the design of the C# programming language.

Richard Gibson

@Richiban

Which channel do people use on the C# discord to do this kind of csharplang chat? There are LOADS of them but none seem to fit, with the possible exception of Roslyn? https://discord.com/channels/143867839282020352/598678594750775301

HaloFour

@HaloFour

Which Discord are you checking? Should be #csharp on DotNetEvolution Discord server: https://aka.ms/dotnet-discord

Antony Male

@canton7

Yeah, I haven't found an equivalent of this channel

HaloFour

@HaloFour

There are two Discord servers, you need to be on the DotNetEvolution one, that has the language design channels and a couple of LDG members frequent. Sometimes it's busier than here, sometimes not.

Antony Male

@canton7

Aah, gotcha. Thanks!

Richard Gibson

@Richiban

Thanks!

masonwheeler

@masonwheeler

How does the compiler get this confused?

Antony Male

@canton7

@masonwheeler JToken defines an explicit cast to string?, see here. Since invoking that explicit operator on a JToken might result in a null string, the compiler's complaining that you're casting it to a non-nullable string

masonwheeler

@masonwheeler

Bleh, of course it's the cast. Thanks, good catch.

Eyal Alon

@eyalalonn

Does anyone know whether there is any meta site for developercommunity to report problems? because when the issue is edited it displays it correctly but when saved not only it doesn't save the format but the information added to the OP, can I post the issue on Roslyn? :|

CyrusNajmabadi

@CyrusNajmabadi

what's the issue @eyalalonn ?

Eyal Alon

@eyalalonn

Well it comes down to my test project builds fine but the tests wouldn’t run not with VS 2022 for Mac (Preview) which might be reasonable but it doesn't run from the CLI as well whereas with Rider everything works fine.

CyrusNajmabadi

@CyrusNajmabadi

doesn't seem like a roslyn issue. :-/

Eyal Alon

@eyalalonn

So what can I do? the feedback site doesn't work.

CyrusNajmabadi

@CyrusNajmabadi

can you give me an example? do you have a feedback item already open?

if so, link me to it.

Eyal Alon

@eyalalonn

yes

https://developercommunity.visualstudio.com/t/Unit-test--could-not-be-loaded/10061422?space=41&scope=follow

Now I want to add more information to it but it wouldn't let me :(

CyrusNajmabadi

@CyrusNajmabadi

log out/in?

then try to add more info. if it fails, send me a screenshot of what failed.

Eyal Alon

@eyalalonn

okay, thanks. ;)

Yes it worked! thanks! ;)

CyrusNajmabadi

@CyrusNajmabadi

cool :)

sorry about that

Eyal Alon

@eyalalonn

I've added C# as a tag which worked but it wouldn't remove C++ or setup from the tag list and I did sign-out and in again. :D

Might takes time for it to update dunno.

Eyal Alon

@eyalalonn

Is there any reason the compiler can infer the type of TRequest in the following case:

public static async Task AddRow<TRequest>(int tableId, TRequest request)

When calling the method like AddRow(..., new MyRequestObject())

But not partially when there is more than one type parameter like in this this?

public static async Task<TResponse> AddRow<TResponse,TRequest>(int tableId, TRequest request)

The first method does not exist so there is no ambiguity so why can't I just do AddRow<MyResponseObject>(..., new MyRequestObject())?

masonwheeler

@masonwheeler

Because the ambiguity can exist, so type parameters are all-or-nothing. It doesn't look at this and say "here's a reference to something with two type parameters and there's only one filled in, so let's see if we can infer the other one." It says "here's a reference with type parameters. Let's see what that matches. Only one match, it has two type parameters, the reference only has one. Error!"

Eyal Alon

@eyalalonn

@masonwheeler It comes down to this feature isn't implemented but the question is whether there is something blocking them from adding something like that?

masonwheeler

@masonwheeler

That's actually not the way the language designers think about it. It's not a "feature that isn't implemented;" it's a thing that is not a feature of the language. The question to ask is not "why isn't this there?" but "should this be there at all?"

Is there a compelling reason why they should devote limited resources to making that, specifically, into a feature rather than all the other things they could be working on instead?

Eyal Alon

@eyalalonn

@masonwheeler

It's not a "feature that isn't implemented;" it's a thing that is not a feature of the language.

What are you trying to say here?

The question to ask is not "why isn't this there?" but "should this be there at all?"

Asking why it can do this and not the other isn't quite as asking why it isn't there. :) and yes I think that more type inference especially when it comes down to generics is useful.

Is there a compelling reason why they should devote limited resources to making that,
specifically, into a feature rather than all the other things they could be working on instead?

Yes because this is exactly how things work. I agreed to do something so I absolutely must prioritize everything I agree on to the next release over existing features but dunno that's something the LDT needs to decide I only asked a simple question, I didn't propose anything.

masonwheeler

@masonwheeler

What are you trying to say here?

Exactly what I said. An unimplemented feature is something the team has decided to do but hasn't implemented yet. This is something that has not been decided. It is not a feature, unimplemented or otherwise.

Asking why it can do this and not the other isn't quite as asking why it isn't there.

The answer to "why it isn't there" is always the same: because nobody has built it. That may sound glib but it's quite literally true. (See above, re: limited resources.) There has to be a very good reason to spend time and effort designing, speccing, coding, testing, and documenting a new feature. Minus 100 points is a good article for understanding the team's perspective on it; I'd recommend familiarizing yourself with the principle it explains if you want to ask for new language features.

Eyal Alon

@eyalalonn

@masonwheeler I'm familiar with the concept. I've been here for many, many years but again I'm not complaining or proposing them to add it as a feature; otherwise, I'd posted a proposal.

HaloFour

@HaloFour

There are a few out there. I get the sense that the language team might be more interested in existential/associated types rather than expanding generic inference in those cases. Same thing, basically, but the other generic parameters aren't really part of the signature.

Eyal Alon

@eyalalonn

@HaloFour Thanks.

masonwheeler

@masonwheeler

Anyone know why the latest version of Visual Studio has stopped showing me grayed out usings that aren't used in the file, but only for some projects and not others?

CyrusNajmabadi

@CyrusNajmabadi

File issue. Sounds like a regression.

masonwheeler

@masonwheeler

Is it possible to put an Attribute marked as [AttributeUsage(AttributeTargets.Constructor)] on the autogenerated constructor from a record?

Joe4evr

@Joe4evr

@masonwheeler doesn't look like it from a little rudimentary experimentation, see if there's an open discussion

HaloFour

@HaloFour

@masonwheeler Where does Delphi have inherited constructors?

masonwheeler

@masonwheeler

@HaloFour The entire language. If you define a constructor on a base class, you can use it on any of its descendants. The place it gets really useful, though, is when you start working with virtual constructors. You can call a virtual constructor with a metaclass reference, pass in the parameters, and get an instance of the class in question.

As I mentioned on the Github discussion, this was largely put in place to support form deserialization. The DFM (form data) files have class names in them, and the RTTI system has something functionally equivalent to a string->metaclass dictionary for all of the Component types used in your program. So you take your class name, get a metaclass from there, invoke the virtual constructor, and you have a Component-typed reference whose actual type is the class you wanted. Use RTTI to fill in the properties, then repeat recursively on all the child components until you have your whole form.

HaloFour

@HaloFour

So that's specifically related to constructors marked as virtual, otherwise it works like other OOP languages where the derived class can call the base constructor but you can't create the derived class by calling the base constructor.

masonwheeler

@masonwheeler

Nope. :point_up:
If you define a constructor on a base class, you can use it on any of its descendants.

HaloFour

@HaloFour

Even if it defines its own constructors that require additional parameters?

masonwheeler

@masonwheeler

type BaseClass = class
   constructor Create(value: integer);
end;

DerivedClass = class(BaseClass)
   constructor Create(value1: integer; value2: string);
end;

...

MyObj := DerivedClass.Create(5); //this is perfectly valid

HaloFour

@HaloFour

Wow, that feels so wrong

masonwheeler

@masonwheeler

It's not what you're used to.

Where communities thrive

People

Repo info

Activity