C# should support indexed property

[gafter]

@ygc369 commented on Tue Apr 21 2015

for example:
class A
{
int[] a, b;
public int pa[int index]
{
get { return a[index]; };
set { a[index]=value; };
};
public int pb[int index]
{
get { return b[index]; };
set { b[index]=value; };
};
};

---

@ygc369 commented on Thu Oct 15 2015

nobody is interested in this?

---

@jrmoreno1 commented on Sat Mar 05 2016

Well, I like the idea, but it's been out there for a while and rejected because of "ambiguous syntax" (not sure what is ambiguous about it myself). I like the idea, because semantically an indexed property is not an item in a collection. In fact I found this while checking to make sure that it hadn't been added since the last time I thought it would make things easier. In this case I was hoping to build a facade around Session, where most of the items are not just per Session but per window/tab. Index properties would make that trivial: Session.ExtensionMethodReturnsFacade().Property[page].

---

@alrz commented on Sat Mar 05 2016

If the the type has an indexer you would get that syntax,

class Foo { public int this[int index] { ... } }
class Bar { public Foo Foo { get; } }

var bar = new Bar();
int foo = bar.Foo[5];

And if you define Foo property as an "indexed property" it'll become ambiguous. I don't see why this is useful.

---

@jrmoreno1 commented on Sat Mar 05 2016

@alrz and you see value in writing and instantiating a class in order to return an indexed value? The class Foo is never going to be used for anything other than the property Foo in Bar. As for the ambiguity -- I still don't see it, at least as far as the compiler is concerned. For people looking at the line in isolation, it wouldn't be clear whether it was an indexed property or a property for an indexer or Foo is just an array or dictionary, but I don't see how that matters. Any way you look at it, it is returning a value from something else. Looking at the definition will tell you what it is.

---

@HaloFour commented on Sat Mar 05 2016

From http://roslyn.codeplex.com/discussions/542484

Please explain why creating nested classes with separate indexers does not satisfy your requirement.

I don't believe this is a feature you're using very often, so extra typing isn't a very good reason why we should add a whole new feature.

If you are using it a lot, I'd like an actual example, because I believe multiple indexers is rarely appropriate.

And that nested type can be a struct, no allocations necessary.

---

@alrz commented on Sat Mar 05 2016

As for the ambiguity -- I still don't see it

See,

class Foo { public Bar this[int index] { ... } }
class Bar { public Foo Foo[int index] { ... } }

var bar = new Bar();
var foo = bar.Foo[5]; // which indexer I'm calling?

For people looking at the line in isolation, it wouldn't be clear whether it was an indexed property or a property for an indexer or Foo is just an array or dictionary, but I don't see how that matters.

If it doesn't matter why don't you use a method instead? You are basically suggesting a method that has [] instead of (),

class Bar { public Foo Foo[int index] { ... } }
class Bar { public Foo Foo(int index) { ... } }

Why is this important then?

---

@alrz commented on Sat Mar 05 2016

As for assignment, ref returns make it possible, but for more complicated scenarios you should seriously consider a method.

---

@jrmoreno1 commented on Sun Mar 06 2016

@HaloFour I gave an example of creating a facade around Sesssion where a page id is used to avoid collissions between tabs. Most of the properties should be indexed. C#, unlike VB, requires that an indexed property be the default property of a class. C# is conflating two entirely separate concepts, and in doing so being unclear.

@alrz in VB, where that works, it calls the indexer in Bar (which returns a Foo), not the default property of a Foo (which returns a Bar). I don't see how it could be otherwise. As for why it matters, you hit upon it in your second message: assignment. A property handles both retrieving and setting a value. Also a method that returns a value is a calculation, while a property is state.

As for ref return, I know the CLR supports it, but c# doesn't, and even if it did, that's not quite the same thing as a property.

---

@HaloFour commented on Sun Mar 06 2016

So your facade exposes a bunch of "named indexable" properties each taking the same index and each returning a different value?

var x = obj.X[id];
var y = obj.Y[id];

Why would that be better than having one normal indexer which returns an object containing all of the properties for that page?

var x = obj[id].X;
var y = obj[id].Y;

Then you can pass all of the properties around without each consumer having to know/care about how they're indexed.

Your use case sounds like the very design that the language should be discouraging.

---

@jrmoreno1 commented on Fri Mar 18 2016

First let me say to that I understand that everything starts at -100, and this isn't a pony let alone a unicorn. I'm not trying to say "do it" but rather, "it shouldn't be rejected". Anyway...

Sorta, not all of the properties will be indexed. Maybe half, pssoibly less.

And doing it as you suggests works, but it creates an artificial distinction between properties on the wrapper and the indexed properties (class page with some properties). Session.w().D seems quite different from Session.w()[id].Y, as opposed to Session.w().D and Session.w().Y[id].

And I don't see why Session.w()[id].Y should be considered better than Session.w().Y[id](other than it works). The later seems more natural to me. The first seems to be saying "get the page of wrapper with the value id and then get the Y property" and the second "get the property Y of wrapper and then get the id value". But page doesn't really exist, it is an additional abstraction that is necessary only in order to make the access actually work.

And since page doesn't model any real object and the properties aren't related in any way, the abstraction doesn't help you reason about the program.

Again, I'm not saying this is a must-have or the best way (let alone the only) way to solve any particular problem. Just that it makes sense and would work, and that the "it's ambiguous" argument is, uhm, weak at best. It provides feature parity with VB. Don't reject the idea of someone doing it.

---

@Corey-M commented on Thu Apr 07 2016

Feature parity aside, I believe that this is still a valid extension to C#. The fact that it is possible in VB.NET shows that it is possible and CLR-supported, so that's not an impediment. The issues raised here seem to revolve around the way that it is used in a particular example... which seems to be a frivolous objection, since there are certainly other use cases which don't necessarily suffer from the same perceived problems.

For instance, I have been working on an extendable wrapper for executing scripts and commands in an embedded PowerShell Runspace. Part of my code looks something like this:

public class Indexer<TKey, TValue> 
{
    private readonly Func<TKey, TValue> _getter;
    private readonly Action<TKey, TValue> _setter;
    public Indexer(Func<TKey, TValue> getter, Action<TKey, TValue> setter)
    {
        _getter = getter;
        _setter = setter;
    }
    public TValue this[TKey key] 
    {
        get { return _getter(key); }
        set { _setter(key, value); }
    }
}

public class PSRunner : IPSRunner
{
    private readonly Runspace _runspace;
    private readonly Indexer<string, object> _variables;
    public Indexer<string, object> Variables { get { return _variables; } }

    public PSRunner(Runspace runspace)
    {
        _runspace = runspace;
        if (_runspace.RunspaceStateInfo.State != RunspaceState.Opened)
            _runspace.Open();
        _variables = new Indexer<string, object>(
            name => _runspace.SessionStateProxy.GetVariable(name),
            (name, value) => _runspace.SessionStateProxy.SetVariable(name, value)
        );
    }
}

This allows me to completely abstract the session state of the PowerShell Runspace and treat the PowerShell variables collection almost as if it were a dictionary without allowing full dictionary semantics that would not work in this situation. All I need this to do (at the moment anyhow) is get and set variables by name, which is easily handled by the above code.

I have other collections that I would also like to be able to access this way, and will end up coding as above, but would much prefer to do this:

public class PSRunner : IPSRunner
{
    private readonly Runspace _runspace;
    public object Variables[string name]
    {
        get { return _runspace.GetVariable(name); }
        set { _runspace.SetVariable(name, value); }
    }

    public PSRunner(Runspace runspace)
    {
        _runspace = runspace;
        if (_runspace.RunspaceStateInfo.State != RunspaceState.Opened)
            _runspace.Open();
    }
}

Shorter code, much more self-explanatory, fewer issues with object initialization (no null assignment to the _variables field in a rogue constructor), etc.

Another issue with the first code block above is that lambda expressions do funny things to garbage collection of the objects they capture. I've had to implement the Indexer as an IDisposable to work around this and ensure that my object get disposed of correctly. I could use weak referencing to resolve the issue, but that way leads to madness and endless custom indexer classes.

And finally, there is not way to prevent or track external references to Indexer instances outside of the class. This could lead to a situation where the Variables index property could be referenced externally and either prevent the correct disposal or garbage collection of my PSRunner instance. A proper indexer property implementation should not allow this to happen.

---

@alrz commented on Thu Apr 07 2016

@Corey-M I believe #6136 can address your use case, something like

extension Runspace {
  public object this[string key] {
    get { return GetVariable(key); }
    set { SetVariable(key, value); }
  }
}

---

@Corey-M commented on Thu Apr 07 2016

@alrz Nice thought, but in order to use an extension on an object I would need to expose that object, which I do not wish to do in this case. All interactions with the Runspace instance need to be mediated by my wrapper class(es). In order to attain proper isolation using extension properties I would have to create, instantiate and expose yet another wrapper object... at which point I am back where I started with an Indexer class with all of the same issues.

I suppoort #6136 in principle - I think it'd be a nice feature to have - but don't see a simple implementation path. Indexed properties (named indexers) already have been implemented in VB so there are no obvious impediments to C# implementation. This is a comparatively simple feature to add.

---

@alrz commented on Thu Apr 07 2016

To hide implementation you'd better use interfaces (#8127).

---

@Corey-M commented on Thu Apr 07 2016

@alrz I'm sorry but your extensions do not fit my needs, and interfaces do not prevent any of the issues I have raised. A reference to an interface can still be cached external to the class instance, the interface still needs to be assigned a concrete object whose lifecycle requires management, code is rendered less readable and so on. All of these points (and more) would be resolved by indexed properties.

---

@HaloFour commented on Fri Apr 08 2016

I could create a delegate to your indexed property and hold onto it forever, thereby holding a root reference to your Runspace class and preventing it from ever being collected. There's nothing you can ever do to prevent anything like that.

Yes, it would be easy to add. It always has been. Yet the team intentionally didn't add it, and they continue to question their utility and discourage their pattern today. They never need a reason to not do something.

---

@Corey-M commented on Sat Apr 09 2016

@HaloFour In the current incarnation yes, you could hold all sorts of references, including a lot of unintentional ones. That's the problem.

But how would that work for an indexed property? What would it look like to do so? Referencing the property itself without an index should fail at compile time and with an index it is not going to be returning a delegate (unless that's what is stored in the backing collection). How exactly would you get that delegate reference? How would you get a delegate to a standard property get/set method for that matter? Reflection?

If bypassing via Reflection is reason not to have a feature then you might as well rip out private and protected modifiers since reflection makes them worthless.

Yes, it would be easy to add. It always has been. Yet the team intentionally didn't add it, and they continue to question their utility and discourage their pattern today. They never need a reason to not do something.

So what you're saying is I should never bother asking or showing support for features that might have come up before because The Team is all-wise, their decisions should never be questioned and no reasons need be given?

Wow. Just wow.

I thought that maybe if we could explain the utility of the feature, explain how we intend to use it and what benefits we would get from doing so, showed support for the feature and so on that maybe - just maybe - they might decide that the feature was actually worth taking another look at, and maybe one day I might get an improvement that will help me with this mess.

Is that not the point of forums like this?

---

@HaloFour commented on Sat Apr 09 2016

But how would that work for an indexed property?

You can create a delegate to any method, including property accessors. Simplest C# syntax would be through a closure:

Func<string, object> d = index => runner.Variables[index];

So what you're saying is I should never bother asking or showing support for features that might have come up before because The Team is all-wise, their decisions should never be questioned and no reasons need be given?

Repeating the same questions will generally just result in the same answers. I think it's fine to question those answers, just as I think it's fine to question the questioning of those answers. The Team ultimately owns their language, though.

---

@Corey-M commented on Sat Apr 09 2016

@HaloFour Your lambda doesn't extract a delegate for the property get method, it closes around the parent object (runner in the example). This is no different to holding any form of reference to the parent object, and you're right that there's nothing I can do to stop it.

What I can do though is track those references down in the the code, or in the IL if necessary. Adding my IIndexer interface and the Indexer class to the search increases the problem space and reduces the likelihood that I'll find the errant reference. Indexed properties would not do so.

The Team ultimately owns their language, though.

Have I implied otherwise? If so I apologize. But should I consider their dislike of the idea the final word and not try to show how it could be beneficial?

---

@alrz commented on Sat Apr 09 2016

This is already backlogged so considering all the things that can be done for C# it is probably up for C# 11.

---

@HaloFour commented on Sat Apr 09 2016

@Corey-M You're right, C# offers no direct syntax to obtain a delegate for a property. However, there are several proposals to add that syntax, not to say that any of them will be implemented. But property accessor methods are just normal methods so some languages could offer that syntax, or you could use reflection to obtain a delegate directly:

PSRunner runner = new PSRunner();
PropertyInfo variablesProperty = typeof(PSRunner).getProperty("Variables");
MethodInfo getterMethod = variablesProperty.GetGetMethod();
Func<string, object> d = (Func<string, object>)Delegate.CreateDelegate(typeof(Func<string, object>), runner, getterMethod);

Have I implied otherwise? If so I apologize. But should I consider their dislike of the idea the final word and not try to show how it could be beneficial?

No, but you'd need a team member to champion the feature in order to get it implemented, even if you (or someone else) were to actually do the work. If they don't think that it's worth the cost in terms of effort and permanent support then it won't happen. In terms of showing how it could be beneficial, I'm not seeing anything new here.

[jjvanzon]

Example with proposed indexer syntax:

// Implementation

internal class SoundCalculator : ISoundCalculator
{
    public double Values[int listIndex]
    {
        get; set;
    }

    public double Values[string name]
    {
        get; set;
    }

    public double Values[string name, int listIndex]
    {
        get; set;
    }

    public double Values[DimensionEnum dimensionEnum]
    {
        get; set;
    }

    public double Values[DimensionEnum dimensionEnum, int listIndex]
    {
        get; set;
    }

    public void Calculate(float[] buffer, int frameCount, double t0)
    {
        // ...
    }
}

// Interface

internal interface ISoundCalculator
{
    double Values[int listIndex];
    double Values[string name];
    double Values[string name, int listIndex];
    double Values[DimensionEnum dimensionEnum];
    double Values[DimensionEnum dimensionEnum, int listIndex];
    void Calculate(float[] buffer, int frameCount, double t0);
}

Example with current indexer syntax:

// Implementations

internal class SoundCalculator : ISoundCalculator
{
    public ISoundCalculatorIndexers Values { get; } = new SoundCalculatorIndexers();

    public void Calculate(float[] buffer, int frameCount, double t0)
    {
        // ...
    }
}

internal class SoundCalculatorIndexers : ISoundCalculatorIndexers
{
    public double this[int listIndex]
    {
        get; set;
    }

    public double this[string name]
    {
        get; set;
    }

    public double this[string name, int listIndex]
    {
        get; set;
    }

    public double this[DimensionEnum dimensionEnum]
    {
        get; set;
    }

    public double this[DimensionEnum dimensionEnum, int listIndex]
    {
        get; set;
    }
}

// Interfaces

internal interface ISoundCalculator
{
    ISoundCalculatorIndexers Values { get; }
    void Calculate(float[] buffer, int frameCount, double t0);
}

internal interface ISoundCalculatorIndexers
{
    double this[int listIndex] { get; set; }
    double this[string name] { get; set; }
    double this[string name, int listIndex] { get; set; }
    double this[DimensionEnum dimensionEnum] { get; set; }
    double this[DimensionEnum dimensionEnum, int listIndex] { get; set; }
}

Caller code (same for current indexer syntax and proposed indexer syntax):

// Caller

internal class CallerOfSoundCalculator
{
    public void DoSomething()
    {
        ISoundCalculator calculator = new SoundCalculator();

        const int noteListIndex = 1;

        double frequency = calculator.Values[DimensionEnum.Frequency, noteListIndex];
        double volume = calculator.Values[DimensionEnum.Volume];

        frequency *= 2;
        volume *= 0.6;

        calculator.Values[DimensionEnum.Frequency, noteListIndex] = frequency;
        calculator.Values[DimensionEnum.Volume] = volume;
    }
}

That makes the original indexers much like the proposed new indexers. I still think it's weird that it needs its own interface and class. I hope the examples help.

[miloush]

I don't understand the ambiguity argument, isn't the answer to do the same thing VB does?

Also all the arguments seem to be targeting you as the authors of the code. Doesn't really help if you need to consume 3rd party or interop that use this feature.

[jjvanzon]

I removed my post with example code that was only complicated, because I did it wrong.

[Thaina]

I always wish we could have this feature

And if we have this feature it would be natural to introduce extension property syntax

public static Ext
{
    public static string SafeString[this object obj]
    {
         get => obj?.ToString();
    }

    public static string SafeStringIndex[this object[] array,int i]
    {
         get => array?.Length > i ? array[i].ToString() : null;
    }
}

object[] array;
var name = array[0].SafeString;
var name0 = array.SafeStringIndex[0];

[TobiasKnauss]

I have started my project with C++/CLI (due to including C++ DLLs and other reasons) and currently convert most of the code to C#. NOTICE: Like VB.net, C++/CLI also is a .net language
C++/CLI does have indexed/parameterized properties, and it works really well:

String^ Source::get (int i_ixSource)
{
  if (i_ixSource < 0 || i_ixSource >= Size)
    return nullptr;

  return m_asSource[i_ixSource];
}

void Source::set (int i_ixSource, String^ i_sFile)
{
  if (i_ixSource < 0 || i_ixSource >= Size)
    return;

  m_asSource[i_ixSource] = i_sFile;
}

It's a big pain in the ass to work around that now and find a suitable implementation in C#.
I want this feature in C# ASAP.

[sonicbhoc]

I'm in the same boat as @TobiasKnauss; I am working on a C++/CLI project in which I'm exposing few properties through indexed properties. I still don't get why C++/CLI, VB.NET, AND F# all support this feature, but C# doesn't. It seems pretty convenient, and getting the equivalent outcome in C#shown in the post by @jjvanzon is needlessly obtuse.

[scalablecory]

My assumption is that C# designers wanted to reduce the risk that something looks like a property returning a collection. I'm OK with that -- being able to scan through code is useful, and depending on Intellisense would not be an adequate replacement for that.

The only reason I can see for doing this is to improve performance via increased locality and reduced GC pressure, which may be a good reason to revisit this. But again, C# has always placed productivity over performance.

[TobiasKnauss]

@scalablecory: if this feature is not about productivity, what else? Indexed properties make programming much easier in certain situations: you don't have to get a copy of a collection, but you can get directly that one element that you're looking for. There's no chance for misinterpretation, compiler and Intellisense will give you all useful info, and if you also choose a good name for it, then nothing can go wrong.

[scalablecory]

@TobiasKnauss I'm just repeating myself at this point, but again Intellisense wouldn't be good enough. It is good to be able to read code with your eyeballs and not need to hover over everything to understand what is happening.

I also have a hard time thinking of the usefulness for this. I can't think of any instance where I would have used it, and none of the examples given look like good code to me. Perhaps if you can provide some broader real-world examples that would benefit from this -- ideally it'd demonstrate enough "bang" that for the "buck" of reduced readability.

[jjvanzon]

Excuse the lengthy post.

When I realized the proposed indexer syntax does not differ greatly from the current indexer syntax, I rested my case, because I am OK with it in its current form.

But given the recent continuation of the discussion, I feel moved to add a few more arguments to the mix.

The fact that I had some difficulty wrapping my head around how to implement these multiple indexers, says something. I am a pretty experienced programmer, so this should be easy for me. Yet it was not. You could argue I must be simply be too dumb to find this easy, but I somehow don't think that's it. It's more like... you're thinking 'I want an indexer here. I just want [] to work' and it takes you a while to realize you are going to need to instantiate a totally different object. Your intent and what you have to program have some sort of gap in between. That might be something the C# language designers find important (expressiveness).

I think this may actually prevent less experienced programmers from using these things, which would be kind of a shame.

Also, if you want these indexers to be part of an interface, that you expect other programmers to implement, it prevents you from saying: 'Here you have an interface. Implement it, and you're good.' Implementors instead now have to implement two interfaces and kind of awkwardly tie them together. Implementors might bump into the same kind of mental leap you have to take to simply 'make [] work here'. What was that principle called? Principle of least astonishment? It kind of applies here.

The alternative? Just not use indexers. But that kind of misses the point. This should be about how we would like to use indexers. Not how we would like to... not use indexers.

The example I wrote is pretty much a real world example. It is part of a sound synthesizer program that I made, that hooks your MIDI keyboard to a user-configured sound calculation.

As for the 'none of it looks like good code' argument. It makes you sound judgy. I actually benefitted from the multiple ways to pass musical parameters to my sound calculation. Sometimes my calculation has one pitch, sometimes it has more, sometimes I need to set all the pitches at once, sometimes a very specific one. Actually the user kinda needs this flexibility. You want to be able to specify a musical parameter name yourself, but the system also needs a standard vocabulary of parameters to know which is which, hence the enum. And to get everything tied together nicely between user, keyboard, sound output and the calculation, my code needs to be able to talk with the calculation in the most easy to read form that I can come up with. I am well aware of the code smells it has, but it turned out to work quite handy this way. (I think that is a pretty compelling argument in software design.)

[Corey-M]

There are several arguments that seem to support this feature, and so far the objections seem to be primarily based on "I don't want that in my code" or "it'd be too hard to read" or similar sentiments.

For instance:

I also have a hard time thinking of the usefulness for this. I can't think of any instance where I would have used it, and none of the examples given look like good code to me.

I have a hard time thinking of a valid use for attributes on generic type parameters, but they exist. So being unable to think of a good use for a feature seems to have zero impact on the discussion.

Some of the points raised so far in favour:

• Feature parity with other .NET languages: C++/CLI, VB, F#.
• Requires no changes to IL to achieve.
• No more indexer objects whose entire purpose is to emulate this feature.
• Less code required to implement indexer objects and manage their lifecycle.
• Avoids problems with internal references blocking GC.
• Prevent external references, captures, etc. to the indexer objects.
• Improves interop with libraries from other languages that use or expect this feature.

And against (with some notes):

• The syntax is ambiguous.
  Syntactic ambiguity was never really explained, so I'm happy to be further educated on it.
• It can be implemented using methods.
  Yes, we can implement with explicit get_X(index)/set_X(index, value) methods. Just like we can (for the most part) with properties. Syntactic sugar is used in some cases to provide features that we could easily achieve with methods, so this seems like a pointless objection.
• Properties that look like they return a collection but don't are confusing.
  They look exactly like the current this[index] indexer does on a class. While that is commonly used in collection types it pops up all over the place. For better or worse the genie is already out of this particular bottle, let's put it to work.
• Proposal #nnn would let you do this a different way.
  That's nice. Let me know when that becomes an actual feature of the language.

[CyrusNajmabadi]

I'm not sure i understand any of these points:

No more indexer objects whose entire purpose is to emulate this feature.
Less code required to implement indexer objects and manage their lifecycle.
Avoids problems with internal references blocking GC.
Prevent external references, captures, etc. to the indexer objects.

If i really felt like i needed this in C# i would simply return a struct that exposed such an indexer on it. i.e. something like:

class Whatever {
  MyIndexer SomeProp => new MyIndexer { _this = this };

  private int GetValue(string a, string b) => ...;

  public struct MyIndexer {
    internal Whatever _this;

    public int this[string a, string b] => _this.GetValue(a, b);
  }
}

Now you have parameterized indexers with no overhead. Indeed, in practice the runtime should be smart enough to elide effectively all of this and know that when you do whatever.SomeProp(a, b) that that's a direct call to whatever.GetValue(a, b).

GC and capture simply do not play into this at all.

[CyrusNajmabadi]

I still don't get why C++/CLI, VB.NET, AND F# all support this feature, but C# doesn't

This is the reverse of how we generally think about language features. We don't start with 'x, y and z have it, so C# should'. Instead we have to have a strong reason why such a feature is really needed for the language, and why existing workarounds aren't satisfactory.

For example, C# can already consume parameterized properties. So if you have one of these properties in C++/CLI or VB or F#, C# can already use them.

So the only question is about production of them. And, right not, it has never been super compelling to do provide language support for this. There isn't high demand in it (i.e. we haven't really had any requests about this from huge APIs like teh BCL for going on 15+ years), and (as shown in my previous post) there are simple ways to accomplish the same goal with existing constructs.

Yes, it would make thigns simpler from an authoring perspective to just have this in the language. But only marginally so. From the consumption side it would be the same. Give that htis only really helps the API author, and only a little, and there aren't that many API authors that need it, that's why it's always been a feature that has never really had anyone really wanting to invest their time into it.

[CyrusNajmabadi]

Are there not already several 'features' of C# that are purely syntactic sugar whose only purpose is to simplify written code

Of course. However, we always weigh that simplification against the existing mechanism available.

In this case, there seems to be little need to make the authoring of these properties easier. And there are already solutions that the owner can use that makes the consumption side nice. The latter is what we believe matters most here, and it is already good enough today.

[StellarWolfEntertainment]

@CyrusNajmabadi The existence of workarounds should not weigh against the decision for a feature it should weigh in favor of, it shows that someone (or some group of people) needed a feature badly enough that they took the time to devise a method by which they could emulate the desired behavior, a codebase full of workarounds because those workarounds are why things were never implemented is sloppy code that will break very easily, but if those features do get implemented the code can become more stable relying on a defined system as opposed to the interactions between multiple systems any of which could change even slightly causing the expected interaction behavior to be entirely different.

[CyrusNajmabadi]

The existence of workarounds should not weigh against the decision for a feature it should weigh in favor o

It weighs against it as is greatly diminishes the need to come up with a new solution.

are why things were never implemented is sloppy code that will break very easily

What would break easily with teh workaround here?

[StellarWolfEntertainment]

@CyrusNajmabadi my comment regarding the 'breaking' doesn't necessarily apply to this specific case, but a lot of 'workarounds' depend on features interacting in the same way across different platforms and configurations as well as potentially different versions of the languages runtime which is not guaranteed, and that is where my comment about things breaking easily comes in, but having a 'concrete' feature instead of a workaround negates this problem since it's no longer dependent on interactions between systems that could change based on countless variables, and I would disagree the existence of work arounds does not diminish the need to come up with a deterministic solution it showcases that people actually want it to be a feature, just because I can use a workaround or more complicated solution to emulate a feature does not mean that I would choose it over the actual feature, I would much rather see something as an actual feature than as a workaround to look like a feature.

[CyrusNajmabadi]

my comment regarding the 'breaking' doesn't necessarily apply to this specific case,

Ok. But we have to evaluate each idea on its merits (and drawbacks). If this doesn't have a breaking concern, that is a point in favor of just using a workaround.

and I would disagree the existence of work arounds does not diminish the need to come up with a deterministic solution it showcases that people actually want it to be a feature,

Ok. Agree to disagree then.

[CyrusNajmabadi]

Proposal #nnn would let you do this a different way.
That's nice. Let me know when that becomes an actual feature of the language.

This is not a good argument for having a language feature. Language features come with an enormous cost. The spec for C# is already massive, and every thing added just adds more and more to an already huge and expensive to maintain language. A first thought I always have whne it comes to language feature requests is "is this really valuable enough to make part of the core language?" as well as "can this already be accomplished in a good enough fashion with what has already shipped up to now. If the answers to those are either "no" or "yes" (respectively), then i would not want the feature added to the language.

Similarly, I often ask myself "who is this benefitting". If this is something that only benefits, say, the authors of APIs, then that's a lot less awesome than something that benefits all the consumers of said APIs. There is usually around a 1000:1 ratio (or more) of code consumers to API authors. So it can be a little more painful for the latter group if we focus more on the former group.

[Corey-M]

GC and capture simply do not play into this at all.

Oh?

public class Program
{
    static Func<string, string, int> fn = null;
    
    public static void Main()
    {
        var w = new Whatever();
        var prop = w.SomeProp;
        fn = (a, b) => prop[a, b];
        w = null;
    }
}

The above captures an instance of the Whatever.MyIndexer structure that contains a reference to the Whatever instance that created it. Now as long as I keep that reference hanging around, in this case closed over by the lambda function, the GC will never collect the Whatever instance as long as the program is running.

That's what the reference capture is all about, and why the GC gets involved.

This is not a good argument for having a language feature.

I wasn't arguing for the feature, I was pointing out that "proposed feature x gives a way to do this" is not a valid argument against the implementation of this feature.

Similarly, I often ask myself "who is this benefitting". If this is something that only benefits, say, the authors of APIs, then that's a lot less awesome than something that benefits all the consumers of said APIs.

Who exactly benefits from attributes on generic parameters? That's in the standard. I've yet to see it used, but it's there. Personally I can't even think of a scenario where I might want to use such a thing, and I've done some pretty weird stuff with code over the years. Should we all be campaigning to have such a useless feature removed in order to help reduce the complexity of the language?

If not, then I consider your objection to have no merit.

[CyrusNajmabadi]

@Thaina no copying should necessary. You can just do things like:

struct Matrix4x4
{
    private readonly float[] values;

    public Matrix4x4(int whatever)
    {
        values = new float[16];
    }

    public Matrix4x4Rows Rows => new Matrix4x4Rows(values);
}

struct Matrix4x4Rows
{
    private readonly float[] values;
    public Matrix4x4Rows(float[] values) => this.values = values;
    public Vector4 this[int index] => new Vector4(values, index);
}

struct Vector4
{
    private readonly float[] values;
    private readonly int rowIndex;

    public Vector4(float[] values, int rowIndex)
    {
        this.values = values;
        this.rowIndex = rowIndex;
    }

    public ref float this[int column] => ref values[rowIndex * 4 + column];
}

class C
{
    static void Main(string[] args)
    {
        var matrix = new Matrix4x4(whatever: 0);
        matrix.Rows[1][2] = 3;
    }
}

[Thaina]

@CyrusNajmabadi This is what Matrix4x4 was implemented

https://github.com/dotnet/runtime/blob/925837e9be56a4b04ad023882634be44ed10c0f8/src/libraries/System.Numerics.Vectors/ref/System.Numerics.Vectors.cs#L54-L72

You don't understand, Matrix4x4 is struct means it is not internally array. Array is reference type. struct MatrixMxN is a bunch of pack of float and it will be passed by value so your example cannot work with it

While indexed property could be just

    public partial struct Matrix4x4 : System.IEquatable<System.Numerics.Matrix4x4>
    {
        public float M11;
        public float M12;
        public float M13;
        public float M14;
        public float M21;
        public float M22;
        public float M23;
        public float M24;
        public float M31;
        public float M32;
        public float M33;
        public float M34;
        public float M41;
        public float M42;
        public float M43;
        public float M44;

        public Vector4 Rows[int i]
        {
            get => switch(i) // just use switch expression for convenience
            {
                0 => new Vector4(M11,M12,M13,M14);
                1 => new Vector4(M21,M22,M23,M24);
                2 => new Vector4(M31,M32,M33,M34);
                3 => new Vector4(M41,M42,M43,M44);
                _ => throw IndexOutOfRange;
            }
            set => switch(i)
            {
                0 => (M11,M12,M13,M14) = value; // assume vector can be deconstruct to 4 floats
                1 => (M21,M22,M23,M24) = value;
                2 => (M31,M32,M33,M34) = value;
                3 => (M41,M42,M43,M44) = value;
                _ => throw IndexOutOfRange;
            }
        }

        public Vector4 Columns[int i] // same as rows just transposing
    }

[CyrusNajmabadi]

You don't understand, Matrix4x4 is struct means it is not internally array. Array is reference type. struct MatrixMxN is a bunch of pack of float and it will be passed by value so your example cannot work with it

Gotcha. Sounds like something ref-fields would help with. you could return then return through a ref struct that points to your values by ref.

[Thaina]

@CyrusNajmabadi Yeah but it also mean this is the same level of impossibility in C#. It need some new feature like ref field and non-copyable type just to make workaround for indexed property with internal struct proxy

And so your workaround is not fully replaceable for indexed property

[CyrusNajmabadi]

And so your workaround is not fully replaceable for indexed property

Yup. But i would champion ref fields more than this. That one seems like a more flexible and more applicable feature to enable lots of scenarios (like yours for example). I'd prefer the broader feature here over the narrower one . Thanks! :)

[BreyerW]

@CyrusNajmabadi wouldnt ref fields require ref struct on containing type effectively making your idea invalid in situations like @Thaina described? MatrixNxN is normally heapable while ref struct would prevent that

[CyrusNajmabadi]

The matrix would be a normal struct. The proxy would contain a ref to it. So the proxy couldn't move to the heap. That would be fine given that you couldn't do that either with a parameterized propriety.

[TobiasKnauss]

Here's another reason for implementing indexed properties: In the likely most used workaround of creating an helper class, e.g. https://stackoverflow.com/a/17618809, you can't create intellisense comments in suitable way.
The helper class is reused at many places, therefore the this property of the helper class should actually not have any intellisense comment at all.
And the property in the main class can have an intellisense comment, but the parameters cannot be described, because this property is just a get {return indexHelper;} and it has no parameters.
If you want to have correct intellisense comments, then you'll have to create a separate helper class for every use case, which is just stupid.

[amshaikh]

We will see one day C# will implement it. Lets remain hopeful.

[jhudsoncedaron]

At this point, if you want it done, fork the compiler. I seriously considered it but it would take more momentum than just myself to maintain a fork.

[acaly]

I remember the C# team have mentioned that they don't recommend us to do that, because it will introduce dialects for the language that make the language much more complicated.

[lcsondes]

They can always add this feature that's already in VB to prevent dialects from forming ;)

[Gadworx]

This is such a useful feature that should be implemented...makes coding faster...easier to debug...more powerful...
This was probably the biggest shock i had when I moved from vb.net to c#. VB has had this for so long I do not remember not using it. I think those that do not see the value in it, have just not really experienced its usefulness in other languages. Of course, there is always ways around not using it, which I do but I cringe every time of the extra work involved to achieve the same outcome....as a programming contractor every extra minute of coding loses money for me. As I write this, I wonder if a class written in VB with multiple get/set properties with their own individual "indexers" is fully accessible from C#? Hmmm...will have to try that someday.

[Gadworx]

This is such a useful feature that should be implemented...makes coding faster...easier to debug...more powerful...
This was probably the biggest shock i had when I moved from vb.net to c#. VB has had this for so long I do not remember not using it. I think those that do not see the value in it, have just not really experienced its usefulness in other languages. Of course, there is always ways around not using it, which I do but I cringe every time of the extra work involved to achieve the same outcome....as a programming contractor every extra minute of coding loses money for me. As I write this, I wonder if a class written in VB with multiple get/set properties with their own individual "indexers" is fully accessible from C#? Hmmm...will have to try that someday.

[SerratedSharp]

I have a couple dozen methods that have this pattern:

public string Style(string propertyName) => FuncString("css", propertyName);
public JQueryBox Style(string properyName, string value) => FuncJQuery("css", properyName, value);

These would more naturally be implemented as an indexer. This required an additional class, additions to the constructors to initialize an indexer proxy class for each pair of these methods. Feels like alot of unnecessary scaffolding and instantiation. It feels like a couple functions which fit a get/set pattern well, in the effort to reduce them down to an indexer, ballooned out quite a bit.

public JQueryBox()
{
    styles = new JQIndexer(this, "css");
    properties = new JQIndexer(this, "property");
    classes = new JQIndexer(this, "class");
    // times a dozen more
}

//public JQIndexer Styles { get; set; } = new JQIndexer(this, "css");// can't reference this, I have to move this initilizer to the ctor
public JQIndexer Styles { get => styles;}        
private JQIndexer styles;

public  class JQIndexer
{
    private readonly JQueryBox jQueryBox;
    private readonly string funcName;

    public JQIndexer(JQueryBox jQueryBox, string funcName)
    {
        this.jQueryBox = jQueryBox;
        this.funcName = funcName;
    }

    public string this[string name] {
        get => jQueryBox.FuncString(funcName, name);
        set => jQueryBox.FuncString(funcName, name, value);
    }

}

[SunSi12138]

like this , you can have a type seperate indexer, but the T1 can't be the same with T2, we need a clean way to access both setter and getter:

public class BiDiction<T1, T2>
{
    readonly Dictionary<T1, T2> _dict1 = [];
    readonly Dictionary<T2, T1> _dict2 = [];

    public void Add(T1 key1, T2 key2)
    {
        _dict1.Add(key1, key2);
        _dict2.Add(key2, key1);
    }

    public T2 this[T1 key1]
    {
        get { return _dict1[key1]; }
        set { _dict1[key1] = value; }
    }

    public T1 this[T2 key2]
    {
        get { return _dict2[key2]; }
        set { _dict2[key2] = value; }
    }
}

if there's a named indexer, it would be helpful

[StellarWolfEntertainment]

@alrz to provide a 'real' use case where an 'indexed property' is not the same as a property with an indexer (this is 'theorized' from a project I am working on), and yes I know that I can always just use 'Get/Set' methods, but this would be much more useful to me

namespace ChangedForAtLeastALittleSecrecy;

public class Tile
{
    [NonSerialized] private DirectionArray<Tile> m_Neighbors = new DirectionArray<Tile>();

    public Tile Neighbor[Direction direction]
    {
        get {
            return m_Neighbors[direction];
        }

        set {
            if (m_Neighbors[direction] != null)
                return;

            m_Neighbors[direction] = neighbor;

            if (neighbor != null)
                neighbor.m_Neighbors[direction.Opposite()] = this; // a property with an indexer would require a 'second' parameter for this part to work
        }
    }
#endregion
}

And yes I do know that this is just ONE use case, but it does demonstrate that an indexed property is not the same, and I also (after about thirty minutes) realized that the posted code doesn't show why not just use an indexer directly, and that is because I actually have more than one stored DirectionArray each of which have 'special' requirements for setting and all would benefit from a named indexer

[CyrusNajmabadi]

Struct proxy types don't always work.

I'd be fine making that work (it's just removing an error) versus all the work to design and implement indexed properties.

[jnm2]

I'm not convinced it's as simple as just removing the error. The error is definitely desired here, and it's not just about inline arrays:

var c = new C();

// GOOD error CS0131: The left-hand side of an assignment must be a variable, property or indexer
c.InlineArray[42] = new object();

class C
{
    public InlineArray43<object> InlineArray { get; set; }
}

[System.Runtime.CompilerServices.InlineArray(43)]
struct InlineArray43<T> { public T Element0; }

If you want to mutate it, it's like with any other mutable-struct-valued property. You need to be doing this:

var temp = c.InlineArray;
temp[42] = new object();
c.InlineArray = temp; // This is the key part

Therefore, I don't see a path toward enabling zero-alloc simulation of named indexers, certainly not a path that is necessarily less work than exposing the named indexer IL feature in C#.

Status quo: Named indexer simulation currently requires class-based wrappers.

[jnm2]

💡 What if the error is keyed on whether the struct is readonly? If the struct is readonly, you're definitely not in need of a followup assignment back to the property.

var c = new C();

// Proposal: no error because SimulatedNamedIndexer returns a readonly struct.
c.SimulatedNamedIndexer[42] = new object();

class C
{
    public WrapperStruct SimulatedNamedIndexer => new(this);

    public readonly struct WrapperStruct(C c)
    {
        public object this[int index]
        {
            // Indexer accesses private state or calls private methods in 'c'
            get => ...;
            set => ...;
        }
    }
}

[jnm2]

The inline array example is a different error, but hopefully the point still stands if you replace an inline array with some other mutable struct.

[jnm2]

There's an existing discussion for this! #8364