Hi,

Just noticed that using UFCS does not work for nested functions, and was wondering whether that's intended, and what the rationale behind it is:


class X {
	void foo() {}
}

void main() {
	// moving bar to module scope solves the error below
	void bar(X x) {}

	X x;
	x.foo(); // ok
	bar(x); // ok
	x.bar(); // Error: no property 'bar' for type 'nested.X'
}

Steffen Wenz:

> Just noticed that using UFCS does not work for nested functions,

Right.


> and was wondering whether that's intended, and what the rationale behind it is:

Currently it's intended. Because doing otherwise causes other problems with struct/class member functions. Perhaps there are ways to design around this problem, but so far no one has suggested a good way to design it (and the good Kenji has worked on this problem, so probably there are no simple solutions).

Once you understand this problem space well, if you find a good design solution you can submit it to Bugzilla.

Bye,
bearophile

On Sunday, 18 May 2014 at 08:15:08 UTC, Steffen Wenz wrote:
> Hi,
>
> Just noticed that using UFCS does not work for nested functions, and was wondering whether that's intended, and what the rationale behind it is:

I just had the same question.

I can imagine that the context pointer of nested functions complicates things, but making `bar` `static` does not help. Has anything changed in recent years regarding the difficulty of implementing UFCS for nested functions? Would it be easier to only support static nested functions?

```
void main() {
    static void bar(int x) {}

    int x;
    x.bar(); // Error: no property 'bar' for type 'int'
}
```

On 6/18/18 7:16 AM, Bastiaan Veelo wrote:
> On Sunday, 18 May 2014 at 08:15:08 UTC, Steffen Wenz wrote:
>> Hi,
>>
>> Just noticed that using UFCS does not work for nested functions, and was wondering whether that's intended, and what the rationale behind it is:
> 
> I just had the same question.
> 
> I can imagine that the context pointer of nested functions complicates things, but making `bar` `static` does not help. Has anything changed in recent years regarding the difficulty of implementing UFCS for nested functions? Would it be easier to only support static nested functions?
> 
> ```
> void main() {
>      static void bar(int x) {}
> 
>      int x;
>      x.bar(); // Error: no property 'bar' for type 'int'
> }
> ```

It's never been supported, and likely will not be. I think the idea is that you can override expected behavior inside by accidentally defining some function locally with the same name.

-Steve

On Monday, 18 June 2018 at 14:19:30 UTC, Steven Schveighoffer wrote:
> On 6/18/18 7:16 AM, Bastiaan Veelo wrote:
>> On Sunday, 18 May 2014 at 08:15:08 UTC, Steffen Wenz wrote:
>>> Hi,
>>>
>>> Just noticed that using UFCS does not work for nested functions, and was wondering whether that's intended, and what the rationale behind it is:
>> 
>> I just had the same question.
>> 
>> I can imagine that the context pointer of nested functions complicates things, but making `bar` `static` does not help. Has anything changed in recent years regarding the difficulty of implementing UFCS for nested functions? Would it be easier to only support static nested functions?
>> 
>> ```
>> void main() {
>>      static void bar(int x) {}
>> 
>>      int x;
>>      x.bar(); // Error: no property 'bar' for type 'int'
>> }
>> ```
>
> It's never been supported, and likely will not be. I think the idea is that you can override expected behavior inside by accidentally defining some function locally with the same name.
>
> -Steve

Wondering how this is different than with non-nested functions? If a global function has the same name as a member function then the member function takes precedence. So wouldn't the same thing just apply here if it were supported?

Cheers,
- Ali

On Monday, 18 June 2018 at 17:16:29 UTC, aliak wrote:
> On Monday, 18 June 2018 at 14:19:30 UTC, Steven Schveighoffer wrote:
>> On 6/18/18 7:16 AM, Bastiaan Veelo wrote:
>>> On Sunday, 18 May 2014 at 08:15:08 UTC, Steffen Wenz wrote:
>>>> Hi,
>>>>
>>>> Just noticed that using UFCS does not work for nested functions, and was wondering whether that's intended, and what the rationale behind it is:
>>> 
>>> I just had the same question.
>>> 
>>> I can imagine that the context pointer of nested functions complicates things, but making `bar` `static` does not help. Has anything changed in recent years regarding the difficulty of implementing UFCS for nested functions? Would it be easier to only support static nested functions?
>>> 
>>> ```
>>> void main() {
>>>      static void bar(int x) {}
>>> 
>>>      int x;
>>>      x.bar(); // Error: no property 'bar' for type 'int'
>>> }
>>> ```
>>
>> It's never been supported, and likely will not be. I think the idea is that you can override expected behavior inside by accidentally defining some function locally with the same name.
>>
>> -Steve
>
> Wondering how this is different than with non-nested functions? If a global function has the same name as a member function then the member function takes precedence. So wouldn't the same thing just apply here if it were supported?
>
> Cheers,
> - Ali

I second this.

On 6/18/18 1:25 PM, bauss wrote:
> On Monday, 18 June 2018 at 17:16:29 UTC, aliak wrote:
>> On Monday, 18 June 2018 at 14:19:30 UTC, Steven Schveighoffer wrote:
>>> On 6/18/18 7:16 AM, Bastiaan Veelo wrote:
>>>> On Sunday, 18 May 2014 at 08:15:08 UTC, Steffen Wenz wrote:
>>>>> Hi,
>>>>>
>>>>> Just noticed that using UFCS does not work for nested functions, and was wondering whether that's intended, and what the rationale behind it is:
>>>>
>>>> I just had the same question.
>>>>
>>>> I can imagine that the context pointer of nested functions complicates things, but making `bar` `static` does not help. Has anything changed in recent years regarding the difficulty of implementing UFCS for nested functions? Would it be easier to only support static nested functions?
>>>>
>>>> ```
>>>> void main() {
>>>>      static void bar(int x) {}
>>>>
>>>>      int x;
>>>>      x.bar(); // Error: no property 'bar' for type 'int'
>>>> }
>>>> ```
>>>
>>> It's never been supported, and likely will not be. I think the idea is that you can override expected behavior inside by accidentally defining some function locally with the same name.
>>>
>>
>> Wondering how this is different than with non-nested functions? If a global function has the same name as a member function then the member function takes precedence. So wouldn't the same thing just apply here if it were supported?
>>
> 
> I second this.

What then can happen is that your local calls can get hijacked from outside the module, if someone happens to define something later that you happened to import. D tries to avoid such possibilities.

There's not much precedent for local symbols being overridden by module-level symbols.

-Steve

On Monday, 18 June 2018 at 17:58:11 UTC, Steven Schveighoffer wrote:
> On 6/18/18 1:25 PM, bauss wrote:
>> On Monday, 18 June 2018 at 17:16:29 UTC, aliak wrote:
>>> On Monday, 18 June 2018 at 14:19:30 UTC, Steven Schveighoffer wrote:
>>>> On 6/18/18 7:16 AM, Bastiaan Veelo wrote:
>>>>> On Sunday, 18 May 2014 at 08:15:08 UTC, Steffen Wenz wrote:
>>>>>> Hi,
>>>>>>
>>>>>> Just noticed that using UFCS does not work for nested functions, and was wondering whether that's intended, and what the rationale behind it is:
>>>>>
>>>>> I just had the same question.
>>>>>
>>>>> I can imagine that the context pointer of nested functions complicates things, but making `bar` `static` does not help. Has anything changed in recent years regarding the difficulty of implementing UFCS for nested functions? Would it be easier to only support static nested functions?
>>>>>
>>>>> ```
>>>>> void main() {
>>>>>      static void bar(int x) {}
>>>>>
>>>>>      int x;
>>>>>      x.bar(); // Error: no property 'bar' for type 'int'
>>>>> }
>>>>> ```
>>>>
>>>> It's never been supported, and likely will not be. I think the idea is that you can override expected behavior inside by accidentally defining some function locally with the same name.
>>>>
>>>
>>> Wondering how this is different than with non-nested functions? If a global function has the same name as a member function then the member function takes precedence. So wouldn't the same thing just apply here if it were supported?
>>>
>> 
>> I second this.
>
> What then can happen is that your local calls can get hijacked from outside the module, if someone happens to define something later that you happened to import. D tries to avoid such possibilities.
>
> There's not much precedent for local symbols being overridden by module-level symbols.
>
> -Steve

I don't understand. What local symbol would be overwritten by which module-level symbol?

Whatever the concerns, what is the difference regarding these concerns between this:
```
// Valid today
void bar(int) {}
void main() {
    int x;
    b.bar;
}
```
and this:
```
\\ Invalid today
void main() {
    static void bar(int) {}
    int x;
    x.bar;
}
```

On Monday, 18 June 2018 at 17:58:11 UTC, Steven Schveighoffer wrote:
> What then can happen is that your local calls can get hijacked from outside the module, if someone happens to define something later that you happened to import. D tries to avoid such possibilities.
>
> There's not much precedent for local symbols being overridden by module-level symbols.
>
> -Steve

I thought that happens already with non-nested functions:

module a;

struct A {
  void f(); // assume it's added later
}

module b;
import a;

void f(A) { }
void g() {
  auto x = A();
  a.f(); // this would be calling local f until someone added A.f
}

Or I misunderstood what you said?

Cheers,
- Ali

PS: This is something I've worried about before actually [1] when I was more of a noob than now, but I've come to accept I guess :) ... though I could still be misunderstanding things of course :/

https://forum.dlang.org/post/crcbaautgmrglhzvxmvk@forum.dlang.org

On 6/18/18 2:58 PM, aliak wrote:
> On Monday, 18 June 2018 at 17:58:11 UTC, Steven Schveighoffer wrote:
>> What then can happen is that your local calls can get hijacked from outside the module, if someone happens to define something later that you happened to import. D tries to avoid such possibilities.
>>
>> There's not much precedent for local symbols being overridden by module-level symbols.
> 
> I thought that happens already with non-nested functions:
> 
> module a;
> 
> struct A {
>    void f(); // assume it's added later
> }
> 
> module b;
> import a;
> 
> void f(A) { }
> void g() {
>    auto x = A();
>    a.f(); // this would be calling local f until someone added A.f
> }
> 
> Or I misunderstood what you said?

It's the same in the fact that your call is silently switched to a different call. However, in the current syntax, an external entity CANNOT override a local function. When you call the nested function, it's the nested function, no matter what else occurs outside (even in the local module). There is no precedent for local functions to be overridden by module-level functions.

So if we allow this, we break a guarantee of which function is called, albeit via a different syntax. Generally, the local function takes precedence, then the member functions, then module-level functions. Making the module level functions override the local functions is not normal or expected. Generally you are defining locals to override what you see outside the function.

But with UFCS, it's treated as part of the API of the type. The type defines the API first, and then you can add to it, you can't override it.

Part of this is historical in nature -- UFCS came after member functions, and so they had to be lower priority.

> PS: This is something I've worried about before actually [1] when I was more of a noob than now, but I've come to accept I guess :) .... though I could still be misunderstanding things of course :/
> 
> https://forum.dlang.org/post/crcbaautgmrglhzvxmvk@forum.dlang.org

I think the current state of affairs still leaves some hijacking doors open, depending on your point of view. It's certainly not perfect. The only way to be sure you are doing things correctly is to use member syntax when you know it's a member, and function syntax otherwise.

IMO, UFCS for locals isn't going to change, but I could also be wrong. It's not really up to me what goes into the language, I just am trying to help explain the rationale behind the current rules.

-Steve