www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.bugs - overloading template functions it not always allowed

reply szali <bszalkai0 gmail.com> writes:
In one of my classes, I created two overloads to opIndexAssign (the
second one was made for better performance, because in most cases
only one index is used):

public final T opIndexAssign(T value, int[] args ...)
public final T opIndexAssign(T value, int i)

These are allowed by the compiler. But these are not:

public T opIndexOpAssign(string op)(T value, int i)
public T opIndexOpAssign(string op)(T value, int[] args ...)

Seems I cannot overload them just because these are template
functions. But I dont see the rationale behind this.

And when I want to make these "final", like this:

public final T opIndexOpAssign(string op)(T value, int i)

the compiler complains because it thinks i want to apply the final
keyword to "string op" (why would I? the keyword is at a completely
different position). That is kind of strange.
Dec 31 2010
parent reply "Steven Schveighoffer" <schveiguy yahoo.com> writes:
On Fri, 31 Dec 2010 09:09:23 -0500, szali <bszalkai0 gmail.com> wrote:

BTW, this list is generally not used for questions (it's auto-generated  
 from bugzilla reports), d.learn is a better place, but no worries, here  
are your answers:


 In one of my classes, I created two overloads to opIndexAssign (the
 second one was made for better performance, because in most cases
 only one index is used):

 public final T opIndexAssign(T value, int[] args ...)
 public final T opIndexAssign(T value, int i)

 These are allowed by the compiler. But these are not:

 public T opIndexOpAssign(string op)(T value, int i)
 public T opIndexOpAssign(string op)(T value, int[] args ...)


It's a limitation of the way templates are specified.  To the compiler,  
both are the same template.

The way around this is to change the template parameters:

public T opIndexOpAssign(string op)(T value, int i)
public T opIndexOpAssign(string op, bool variadic=true)(T value, int[]  
args ...)

It's a crappy requirement, I think this is a well-known bug.

BTW, you gain very very very little by having both these functions, the  
variadic one is all you need.

Also, you may have an issue with using a variadic, as I think you can call  
with zero indexes (not sure how that would look).  You may want to replace  
both with this one function:

public T opIndexOpAssign(string op)(T value, int idx0, int[] idxN...)


 And when I want to make these "final", like this:

 public final T opIndexOpAssign(string op)(T value, int i)

 the compiler complains because it thinks i want to apply the final
 keyword to "string op" (why would I? the keyword is at a completely
 different position). That is kind of strange.


All template functions are final.  They cannot be virtual, so even though  
I feel this is a bug (it should be silently ignored), you can fix it by  
just removing final.

-Steve
Dec 31 2010
parent reply bearophile <bearophileHUGS lycos.com> writes:
Steven Schveighoffer:


 so even though I feel this is a bug (it should be silently ignored),


Generally silently ignoring attributes is exactly the opposite you want from a
modern compiler. See bug 
3934.

Bye,
bearophile
Dec 31 2010
parent "Steven Schveighoffer" <schveiguy yahoo.com> writes:
On Fri, 31 Dec 2010 10:11:02 -0500, bearophile <bearophileHUGS lycos.com>  
wrote:


 Steven Schveighoffer:


 so even though I feel this is a bug (it should be silently ignored),


 Generally silently ignoring attributes is exactly the opposite you want  
 from a modern compiler. See bug
 3934.


In this case though, you are asking for a function which is already final  
to be final.  The compiler can safely ignore the request because the  
request is already satisfied.

If you asked for a virtual function to be final, and the compiler ignored  
the request, I'd say it was bad.

-Steve
Dec 31 2010