View mode: basic / threaded / horizontal-split · Log in · Help
May 29, 2007
Template method and overloading
Why
"template test.A.Foo(T) conflicts with test.A.Foo(T) at ..."
?

public class A
{
	public void Foo(T)(T p1)
	{
	}

	public void Foo(T)(T p1, bool b)
	{
	}
}

The two methods has different signatures.
I'm not really familiar with c++ templates, but c#/.net can handle this.

What's the reason behind this, and will the behaviour change?

Thanks,
gerleim
May 29, 2007
Re: Template method and overloading
gerleim wrote:
> Why
> "template test.A.Foo(T) conflicts with test.A.Foo(T) at ..."
> ?
> 
> public class A
> {
> 	public void Foo(T)(T p1)
> 	{
> 	}
> 
> 	public void Foo(T)(T p1, bool b)
> 	{
> 	}
> }
> 
> The two methods has different signatures.
> I'm not really familiar with c++ templates, but c#/.net can handle this.
> 
> What's the reason behind this, and will the behaviour change?
> 
> Thanks,
>  gerleim

Yeh that's not allowed.  I presume the reason was it was easier to 
implement this way.
The workaround is to give them different *template* parameter lists. 
Something like:
 	public void Foo(T)(T p1)
 	{
 	}

 	public void Foo(T,dummy=float)(T p1, bool b)
 	{
 	}

It's silly that you have to do that, but since that workaround is 
available, I don't think Walter considers fixing it a very high priority.

--bb
May 29, 2007
Re: Template method and overloading
Oh, I see, thanks. Didn't know that template parameter lists have to differ.
(Is that documented?)

I tried to workaround with a tuple template parameter.

gerleim

> Yeh that's not allowed.  I presume the reason was it was easier to 
> implement this way.
> The workaround is to give them different *template* parameter lists. 
> Something like:
>   	public void Foo(T)(T p1)
>   	{
>   	}
> 
>   	public void Foo(T,dummy=float)(T p1, bool b)
>   	{
>   	}
> 
> It's silly that you have to do that, but since that workaround is 
> available, I don't think Walter considers fixing it a very high priority.
> 
> --bb
May 29, 2007
Re: Template method and overloading
"gerleim" <elf_qt@_deletethisifyouarenotaspammer_yahoo.com> wrote in message 
news:f3hvvt$2q18$1@digitalmars.com...
> Oh, I see, thanks. Didn't know that template parameter lists have to 
> differ.
> (Is that documented?)
>

It's because templated functions are really just syntactic sugar.  So:

void foo(T)(int x){}
void foo(T)(int x, float y){}

Is really seen by the compiler as:

template foo(T) { void foo(int x){} }
template foo(T) { void foo(int x, float y){} }

Now it looks obvious that there should be an error, because it's the 
templates that are at top-level, not the functions.
May 29, 2007
Re: Template method and overloading
"Jarrett Billingsley" <kb3ctd2@yahoo.com> wrote in message 
news:f3i3uc$2vkn$1@digitalmars.com...
> "gerleim" <elf_qt@_deletethisifyouarenotaspammer_yahoo.com> wrote in 
> message news:f3hvvt$2q18$1@digitalmars.com...
>> Oh, I see, thanks. Didn't know that template parameter lists have to 
>> differ.
>> (Is that documented?)
>>
>
> It's because templated functions are really just syntactic sugar.  So:
>
> void foo(T)(int x){}
> void foo(T)(int x, float y){}
>
> Is really seen by the compiler as:
>
> template foo(T) { void foo(int x){} }
> template foo(T) { void foo(int x, float y){} }
>
> Now it looks obvious that there should be an error, because it's the 
> templates that are at top-level, not the functions.

But after posting that, I now don't know how the compiler thinks this is any 
easier to deal with:

template foo(T) { void foo(int x){} }
template foo(T, dummy = void) { void foo(int x, float y){} }

How does it find the foo(int x, float y)?  Maybe it's buried in weird 
template rules I don't entirely understand..
May 29, 2007
Re: Template method and overloading
Jarrett Billingsley wrote:
> "Jarrett Billingsley" <kb3ctd2@yahoo.com> wrote in message 
> news:f3i3uc$2vkn$1@digitalmars.com...
>> "gerleim" <elf_qt@_deletethisifyouarenotaspammer_yahoo.com> wrote in 
>> message news:f3hvvt$2q18$1@digitalmars.com...
>>> Oh, I see, thanks. Didn't know that template parameter lists have to 
>>> differ.
>>> (Is that documented?)
>>>
>> It's because templated functions are really just syntactic sugar.  So:
>>
>> void foo(T)(int x){}
>> void foo(T)(int x, float y){}
>>
>> Is really seen by the compiler as:
>>
>> template foo(T) { void foo(int x){} }
>> template foo(T) { void foo(int x, float y){} }
>>
>> Now it looks obvious that there should be an error, because it's the 
>> templates that are at top-level, not the functions.
> 
> But after posting that, I now don't know how the compiler thinks this is any 
> easier to deal with:
> 
> template foo(T) { void foo(int x){} }
> template foo(T, dummy = void) { void foo(int x, float y){} }
> 
> How does it find the foo(int x, float y)?  Maybe it's buried in weird 
> template rules I don't entirely understand.. 

I think that's the SFINAE thing ("Somethinerother Failure Is Not An 
Error").  It tries one template, and that fails to work, so it just goes 
on to try the next until it finds one that works.  Or maybe it finds all 
the ones that work then picks the "best match" of the lot.  (if you have 
a foo(int T) and a foo(float T) for instance).  Either way in the above 
example only one template instantiation works so it's not hard to decide.

--bb
May 30, 2007
Re: Template method and overloading
Bill Baxter escribió:
> Something like:
>      public void Foo(T)(T p1)
>      {
>      }
> 
>      public void Foo(T,dummy=float)(T p1, bool b)
>      {
>      }
> 
> It's silly that you have to do that, but since that workaround is 
> available, I don't think Walter considers fixing it a very high priority.

!

This is so wrong. The language is trying to be elegant with "foreach", 
"delegate" and some other stuff, while there are holes in many other 
places, like this one.

I'm not saying that it's bad the problem exists. I think it's bad fixing 
this kind of things has low priority, while adding new features to the 
language has high priority. I know fixing is boring, and new stuff is 
fun. But... Workaround Oriented Programming is not fun. :-(
May 30, 2007
Re: Template method and overloading
Ary Manzana wrote:
> Bill Baxter escribió:
>> Something like:
>>      public void Foo(T)(T p1)
>>      {
>>      }
>>
>>      public void Foo(T,dummy=float)(T p1, bool b)
>>      {
>>      }
>>
>> It's silly that you have to do that, but since that workaround is
>> available, I don't think Walter considers fixing it a very high priority.
> 
> !
> 
> This is so wrong. The language is trying to be elegant with "foreach",
> "delegate" and some other stuff, while there are holes in many other
> places, like this one.
> 
> I'm not saying that it's bad the problem exists. I think it's bad fixing
> this kind of things has low priority, while adding new features to the
> language has high priority. I know fixing is boring, and new stuff is
> fun. But... Workaround Oriented Programming is not fun. :-(

I'm pretty sure this hasn't been fixed not because it's boring, but
because it's hard.  As Jarret said, the original code expands to two
templates with exactly the same name and set of arguments.

I mean, would you call it a bug that you can't have two methods with the
same name and same arguments, but differ by their contents?  This isn't
so much a bug with D, but just a limitation.

I vaguely remember someone (possibly Walter) saying at some point that
fixing this would require changing the compiler to distinguish between
templates and templated functions; which would require duplicating a lot
of code, and making the front end a lot messier.  When there's a simple
(if ugly) workaround available, I don't see how this is more important
than getting good const semantics.

This you can work around.  Working around a lack of good const is much
harder :)

	-- Daniel

-- 
int getRandomNumber()
{
   return 4; // chosen by fair dice roll.
             // guaranteed to be random.
}

http://xkcd.com/

v2sw5+8Yhw5ln4+5pr6OFPma8u6+7Lw4Tm6+7l6+7D
i28a2Xs3MSr2e4/6+7t4TNSMb6HTOp5en5g6RAHCP  http://hackerkey.com/
May 30, 2007
Re: Template method and overloading
I thought maybe the compiler could merge all the templates with the same 
name and template parameters... I think this is ok as long as you don't 
use mixins, because you'll end up mixing more stuff than what you really 
want...

Unless... the merge is documented somewhere, and so there will be no 
confusion at all. Most of the time you won't have two templated 
functions and then would want to mix them, or have a templated function 
and a raw template with the same name.

Is this possible?

Daniel Keep escribió:
> 
> Ary Manzana wrote:
>> Bill Baxter escribió:
>>> Something like:
>>>      public void Foo(T)(T p1)
>>>      {
>>>      }
>>>
>>>      public void Foo(T,dummy=float)(T p1, bool b)
>>>      {
>>>      }
>>>
>>> It's silly that you have to do that, but since that workaround is
>>> available, I don't think Walter considers fixing it a very high priority.
>> !
>>
>> This is so wrong. The language is trying to be elegant with "foreach",
>> "delegate" and some other stuff, while there are holes in many other
>> places, like this one.
>>
>> I'm not saying that it's bad the problem exists. I think it's bad fixing
>> this kind of things has low priority, while adding new features to the
>> language has high priority. I know fixing is boring, and new stuff is
>> fun. But... Workaround Oriented Programming is not fun. :-(
> 
> I'm pretty sure this hasn't been fixed not because it's boring, but
> because it's hard.  As Jarret said, the original code expands to two
> templates with exactly the same name and set of arguments.
> 
> I mean, would you call it a bug that you can't have two methods with the
> same name and same arguments, but differ by their contents?  This isn't
> so much a bug with D, but just a limitation.
> 
> I vaguely remember someone (possibly Walter) saying at some point that
> fixing this would require changing the compiler to distinguish between
> templates and templated functions; which would require duplicating a lot
> of code, and making the front end a lot messier.  When there's a simple
> (if ugly) workaround available, I don't see how this is more important
> than getting good const semantics.
> 
> This you can work around.  Working around a lack of good const is much
> harder :)
> 
> 	-- Daniel
>
May 30, 2007
Re: Template method and overloading
I've just tried it, it works.

---
module main;

import std.stdio;

template foo(T) {
	void foo(T x, int y) {
		writefln("In first with %s: %s", typeid(T), y);
	}
	void foo(T x, char[] y) {
		writefln("In second with %s: %s", typeid(T), y);
	}
}

void main() {
	int[] x;
	double[] y;
	
	foo!(int[]).foo(x, 1);
	foo!(double[]).foo(y, "hello");
}
---

Output:
In first with int[]: 1
In second with double[]: hello

The only problem is that you have to explicitly say which is T, because 
the "one member" rule dosen't apply anymore. Maybe the rule could be 
changed for functions, to allow this kind of overloads.

Is there any problem with this I'm not being able to see?

Ary Manzana escribió:
> I thought maybe the compiler could merge all the templates with the same 
> name and template parameters... I think this is ok as long as you don't 
> use mixins, because you'll end up mixing more stuff than what you really 
> want...
> 
> Unless... the merge is documented somewhere, and so there will be no 
> confusion at all. Most of the time you won't have two templated 
> functions and then would want to mix them, or have a templated function 
> and a raw template with the same name.
> 
> Is this possible?
> 
> Daniel Keep escribió:
>>
>> Ary Manzana wrote:
>>> Bill Baxter escribió:
>>>> Something like:
>>>>      public void Foo(T)(T p1)
>>>>      {
>>>>      }
>>>>
>>>>      public void Foo(T,dummy=float)(T p1, bool b)
>>>>      {
>>>>      }
>>>>
>>>> It's silly that you have to do that, but since that workaround is
>>>> available, I don't think Walter considers fixing it a very high 
>>>> priority.
>>> !
>>>
>>> This is so wrong. The language is trying to be elegant with "foreach",
>>> "delegate" and some other stuff, while there are holes in many other
>>> places, like this one.
>>>
>>> I'm not saying that it's bad the problem exists. I think it's bad fixing
>>> this kind of things has low priority, while adding new features to the
>>> language has high priority. I know fixing is boring, and new stuff is
>>> fun. But... Workaround Oriented Programming is not fun. :-(
>>
>> I'm pretty sure this hasn't been fixed not because it's boring, but
>> because it's hard.  As Jarret said, the original code expands to two
>> templates with exactly the same name and set of arguments.
>>
>> I mean, would you call it a bug that you can't have two methods with the
>> same name and same arguments, but differ by their contents?  This isn't
>> so much a bug with D, but just a limitation.
>>
>> I vaguely remember someone (possibly Walter) saying at some point that
>> fixing this would require changing the compiler to distinguish between
>> templates and templated functions; which would require duplicating a lot
>> of code, and making the front end a lot messier.  When there's a simple
>> (if ugly) workaround available, I don't see how this is more important
>> than getting good const semantics.
>>
>> This you can work around.  Working around a lack of good const is much
>> harder :)
>>
>>     -- Daniel
>>
« First   ‹ Prev
1 2
Top | Discussion index | About this forum | D home