I just hit a problem in my serialization library, Orange, being integrated as std.serialization. The format of .stringof changed in git HEAD. Now, the question is should the format of .stringof be defined and reliable or not defined at all? Either way, this should be clearly stated in the documentation, which it's currently not.

Below is what's changed. It's managed to stay unchanged for the last 6-7 years.

Before the change:

struct Foo
{
    int a;
}

static assert(Foo.tupleof[0].stringof == "(Foo).a");

After the change:

static assert(Foo.tupleof[0].stringof == "a");

-- 
/Jacob Carlborg

On Friday, 13 September 2013 at 07:06:27 UTC, Jacob Carlborg
wrote:
> I just hit a problem in my serialization library, Orange, being integrated as std.serialization. The format of .stringof changed in git HEAD. Now, the question is should the format of .stringof be defined and reliable or not defined at all? Either way, this should be clearly stated in the documentation, which it's currently not.
>
> Below is what's changed. It's managed to stay unchanged for the last 6-7 years.
>
> Before the change:
>
> struct Foo
> {
>     int a;
> }
>
> static assert(Foo.tupleof[0].stringof == "(Foo).a");
>
> After the change:
>
> static assert(Foo.tupleof[0].stringof == "a");

Funny. I just ran into that with msgpack-d in my project... So
that's two libraries broken.

On 2013-09-13 09:11, Brian Schott wrote:

> Funny. I just ran into that with msgpack-d in my project... So
> that's two libraries broken.

I guess __traits(identifier, Foo.tupleof[0]) is more reliable than .strinfof. But it would be nice to once and for all get an official agreement on this issue.

-- 
/Jacob Carlborg

On Friday, 13 September 2013 at 07:06:27 UTC, Jacob Carlborg wrote:
> I just hit a problem in my serialization library, Orange, being integrated as std.serialization. The format of .stringof changed in git HEAD. Now, the question is should the format of .stringof be defined and reliable or not defined at all? Either way, this should be clearly stated in the documentation, which it's currently not.
>
> Below is what's changed. It's managed to stay unchanged for the last 6-7 years.
>
> Before the change:
>
> struct Foo
> {
>     int a;
> }
>
> static assert(Foo.tupleof[0].stringof == "(Foo).a");
>
> After the change:
>
> static assert(Foo.tupleof[0].stringof == "a");

Aye, I had my own code break from this change as well (though it was a very easy fix in my case). I really prefer the new format however. Including the type made reading people's code very confusing with the random T.tupleof[T.stringof + 3 .. $] if you didn't understand the exact format. That being said, this is something that's commonly enough used that it should be defined whether or not the format could be relied upon.