> Is this what you're looking for?:

No. :)
(Though you already found this out at the end!)
I was looking for some way to keep the storage classes and any other (meta)data
about the parameters that may be added in the future to the library. I have a
feeling this might be a very big change in how DMD handles parameters, but without
this feature, perfect redirection -- which I think is possible in C++0x -- would
be impossible in D.

My suggestion would be that we treat "ref", "out", and "lazy" as type constructors rather than storage specifiers; that way, we could get rid of the concept of storage classes entirely, and it would unify the concept of references with other languages like C# and C++.

How does this idea sound? :) (I think it's easier said than done, though...)

On 2011-01-19 18:28, %u wrote:
>> Is this what you're looking for?:
>
> No. :)
> (Though you already found this out at the end!)
> I was looking for some way to keep the storage classes and any other (meta)data
> about the parameters that may be added in the future to the library. I have a
> feeling this might be a very big change in how DMD handles parameters, but without
> this feature, perfect redirection -- which I think is possible in C++0x -- would
> be impossible in D.

The ugly solution would be to do this:

void test (ref int a) { }

void main ()
{
    writeln(typeof(&test).stringof);
}

Which will print: "void function(ref int a)" and then parse out what you need.

> My suggestion would be that we treat "ref", "out", and "lazy" as type constructors
> rather than storage specifiers; that way, we could get rid of the concept of
> storage classes entirely, and it would unify the concept of references with other
> languages like C# and C++.
>
> How does this idea sound? :) (I think it's easier said than done, though...)


-- 
/Jacob Carlborg

On 1/19/11, %u <wfunction@hotmail.com> wrote:
>> Is this what you're looking for?:
>
> No. :)
> (Though you already found this out at the end!)
> I was looking for some way to keep the storage classes and any other
> (meta)data
> about the parameters that may be added in the future to the library. I have
> a
> feeling this might be a very big change in how DMD handles parameters, but
> without
> this feature, perfect redirection -- which I think is possible in C++0x --
> would
> be impossible in D.
>
> My suggestion would be that we treat "ref", "out", and "lazy" as type
> constructors
> rather than storage specifiers; that way, we could get rid of the concept of
> storage classes entirely, and it would unify the concept of references with
> other
> languages like C# and C++.
>
> How does this idea sound? :) (I think it's easier said than done, though...)
>

Let me see if I get this straight (correct me if I'm wrong in
understanding this).
So you mean we would have e.g. ref as a type constructor:
ref int x;

The problem with this is that now you're functions look like this:
foo(int x) { }

You can't tell if foo modifies x just by looking at the signature. You would have to look at the calling site to find out which arguments were passed, and then you'd have to look at their declaration. (well, unless you use an IDE of some sort that does this automatically, it's 2011 I guess.. :-) ).

Having "in", "ref", "out".. in function parameters helps in understanding your own code. If you look at e.g. Microsoft header files there are a lot of methods named like so:

foo(/*in*/ param1, /*in*/ param2, /*out*/ param3)

D just took this existing practice and made it a language feature.

Anyway, I don't know of a workaround for the full signature problem. For a second I thought about using a string mixin, but mixins can't be used inside the function signatures, so I can't just glue together "ref" and "int".

A little example ...

> The ugly solution would be to do this:
> void test (ref int a) { }
> void main () { writeln(typeof(&test).stringof); }
> Which will print: "void function(ref int a)" and then parse out what you need.

If you're referring to using mixin() to parse the signature, I've already thought of that, but it won't work because if the type of any parameter is a user-defined type, it will be inaccessible outside the module that provides this facility. This feature will require compiler and/or language modification(s); I don't believe you can add the storage-class feature correctly with the current language (though please correct me if I'm wrong).



> Let me see if I get this straight (correct me if I'm wrong in understanding this). So you mean we would have e.g. ref as a type constructor:
ref int x;

Yes.

> The problem with this is that now you're functions look like this:
> foo(int x) { }

No... why? ref would be just like const -- so your parameter's data type could be
ref(const(string)). Why do you say the functions remove the type constructor from
the header?

[This is a bit unrelated, but while we're on the topic of documenting code through ref/out: I think it would be better if the language required the "ref" and "out" keywords on passing the parameters, like C# does... it prevents a lot of bugs that can happen when a function's header is changed, since now the caller code won't compiler, instead of compiling but working potentially incorrectly.)

I actually noticed a different problem with treating storage classes as type constructors: while "ref" and "lazy" would work well, "out" would make no sense as a type constructor, so this solution probably won't work at all.


Does anyone have any ideas on how to get storage classes to work correctly with ParameterTypeTuple, whether or not it's a dirty solution (so long as the solution works)?

On 1/20/11, %u <wfunction@hotmail.com> wrote:
> No... why? ref would be just like const -- so your parameter's data type
> could be
> ref(const(string)).

ref(const(immutable(string)[])) ?

That's crazy!

Anyway I don't really know what we're talking about anymore. :) s/we're/I'm

The string mixin solution provided by wrzosk is similar to what I came up with. Maybe someone who wrote those templates can help out.

> ref(const(immutable(string)[])) ?
> That's crazy!

Did you mean ref(const(immutable(char)[]))?? :]
Haha... well it was just an idea on how to add the functionality, and like I
mentioned, even if it was fine for 'ref' and 'lazy', it wouldn't make any sense
for 'out' anyway, so I'd say screw my idea; forget it. :)


The trouble with mixin is that it doesn't work the way templates like
ParameterTypeTuple do, because if you dynamically generate the header then one of
these must be true:
1. You can mixin() it in a template in Phobos, but that won't work because any
data types not in the current (Phobos) module can't be referenced, which basically
erases this solution completely.
2. You can just return a string and have the user use mixin() every time, but
that's a very ugly and error-prone solution, since it would litter the user's code
with lots of mixin()s.

Hope that made sense... so my conclusion is that, with the current language/compiler features, it's likely there's no workable solution for this (although I'm happy to be proven wrong!!). :\