Beta 2.103.0

[Dmitry Olshansky]

On Thursday, 2 March 2023 at 15:56:35 UTC, ryuukk_ wrote:
> On Thursday, 2 March 2023 at 14:40:04 UTC, Iain Buclaw wrote:
>> Glad to announce the first beta for the 2.103.0 release, ♥ to 
>> the 43 contributors.
>>
>> This release comes with 9 major changes, including:
>>
>> - In the compiler, `-preview=dip25` is now enabled by default.
>> - In the standard library, std.uni Grapheme functions have 
>> been updated to conform to Unicode 15
>> - In dub, the `--color` argument now accepts the values 
>> `auto`, `never`, and `always`.
>>
>> http://dlang.org/download.html#dmd_beta
>> http://dlang.org/changelog/2.103.0.html
>>
>> As usual please report any bugs at https://issues.dlang.org
>>
>> -Iain
>> on behalf of the Dlang Core Team
>
> ```D
> @safe ref int wrongIdentity(ref int x) {
>     return x; // ERROR! Cannot return a ref, please use "return 
> ref"
> }
> @safe ref int identity(return ref int x) {
>     return x; // fine
> }
> ```
> a keyword to return a value
>
> ``return 5;``
>
> and a keyword to tell that a reference is returnable
>
> ``return ref int x``
>
> that's dumb, why?

return ref is probably the tipping point for me. Even if better 
keyword is used the accrual of complexity never ends. Though 
arguably transitive qualifiers and inout were the point of no 
return. Recently I had a case where in 64-bit mode a 
data-structure could be implicitly converted to immutable while 
in 32-bit it wouldn’t. After some mucking around a seemingly 
unrelated change fixed it, the error message gave no clue of 
what’s wrong.

Anyway I’d have a hard time selling transitive qualifiers in 
particular shared, as most of the time shared object is taken off 
the queue and should become thread-local. The type system doesn’t 
want to hear any of it. Seems like ownership system beats 
transitive qualifiers here, albeit both add significant friction.

—
Dmitry Olshansky