Do you know if it's possible to replace  typeof(f1)  with  f1.typeof  in D (for symmetry with sizeof too)?

import std.stdio: writeln;
struct Foo {
    int x;
}
void main() {
    Foo f1;
    int fsize = f1.sizeof; // OK
    alias typeof(f1) T1;   // OK
    alias f1.typeof T2;    // ERR
}


[What I'd like is this, but this is for another thread:
Type T2 = f1.typeof;
]

Bye,
bearophile

On 16/05/10 21:43, bearophile wrote:
> Do you know if it's possible to replace  typeof(f1)  with  f1.typeof  in D (for symmetry with sizeof too)?
>
> import std.stdio: writeln;
> struct Foo {
>      int x;
> }
> void main() {
>      Foo f1;
>      int fsize = f1.sizeof; // OK
>      alias typeof(f1) T1;   // OK
>      alias f1.typeof T2;    // ERR
> }
>
>
> [What I'd like is this, but this is for another thread:
> Type T2 = f1.typeof;
> ]
>
> Bye,
> bearophile

The closest I've managed with my quick attempt is:
----
mixin template typeOf()
{
	alias typeof(this) typeOf;
}
struct Foo
{
	int x;
	mixin typeOf;
}
void main() {
    Foo f1;
    auto fsize = f1.sizeof; // OK
    alias typeof(f1) T1;   // OK
    alias f1.typeOf T2;    // OK
}
----

It's not ideal, but it works. You can choose a capitalization to suit you. If you want the capitalization to match you can make a similar template for sizeof to keep things consistent.

A couple of other things to note, I changed fsize to auto, as the type of .sizeof is different on x86_64, this isn't an issue now, but it's nice to keep code portable! Another thing is template bloat - I haven't looked properly, it seems that the template is optimized out though, so using it doesn't add any overhead... I could be wrong though.

Robert

Robert Clipsham:

> The closest I've managed with my quick attempt is:
> ----
> mixin template typeOf()
> {
> 	alias typeof(this) typeOf;
> }
> struct Foo
> {
> 	int x;
> 	mixin typeOf;
> }

Thank you for your code :-) Your example shows this feature can become a built-in.


> I changed fsize to auto, as the type of .sizeof is different on x86_64, this isn't an issue now, but it's nice to keep code portable!

You are right, thank you. At the moment it's a size_t, that is an unsigned CPU word. (I have recently asked in bugzilla for all those values to become signed words, because unsigned are too much unsafe in a language that doesn't catch signed-unsigned conversions and has no integral overflows tests).

Bye,
bearophile