https://issues.dlang.org/show_bug.cgi?id=23300

RazvanN <razvan.nitu1305@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |razvan.nitu1305@gmail.com

--- Comment #1 from RazvanN <razvan.nitu1305@gmail.com> ---
(In reply to Ate Eskola from comment #0)
> This should compile but does not on dmd 2.100, using -preview=dip1000 flag.
> 
> -------------
> @safe int[] fun()
> { import std : array, map;
>   scope r = [1,2,3].map!"a+3";
>   return r.array;
> }
> -------------
> 
> Were I returning `r` directly, this would be escaping a reference to scoped data, but `array` copies the data. Returning `array`ed range should be allowed, `scope` or no.

The problem here is that `r` is `scope` so for it to be able to be passed to `array`, `array` also needs to receive a scope parameter. However, `array`'s parameter cannot be `scope` because it is copied into allocated memory.

By the current scope rules this seems to be correct behavior, making this bug report invalid.

--

https://issues.dlang.org/show_bug.cgi?id=23300

Dennis <dkorpel@live.nl> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |dkorpel@live.nl

--- Comment #2 from Dennis <dkorpel@live.nl> ---
(In reply to RazvanN from comment #1)
> However, `array`'s
> parameter cannot be `scope` because it is copied into allocated memory.

The parameter to `array` should infer `scope` unless the range has non-scope range primitives. It's not escaping the input array, it's dereferencing it and copying the elements into a new GC array which is not scope.

`scope` inference fails and it falls back on `return scope` inference from `pure`. By making it impure, the newest DMD reports:

test_.d(12): Error: scope variable `r` assigned to non-scope parameter `r`
calling `array`
std/array.d(112):        which is not `scope` because of `__r353 = r`

https://github.com/dlang/phobos/blob/b578dfad94770574d7e522557a77276c35943daa/std/array.d#L112

So the underlying problem is that `scope` inference is defeated by `foreach` on a struct with range primitives.

--

https://issues.dlang.org/show_bug.cgi?id=23300

--- Comment #3 from Ate Eskola <Ajieskola@gmail.com> ---
I narrowed down what causes the inference to fail

------
auto fun(int* ptr)
{ // comment out this line and this compiles
  // as does if you explicitly mark either it or the argument scope
  auto r = ptr;
  return new int;
}

@safe int* gun(scope int* ptr){return ptr.fun;}
------

Is the scope inference supposed to behave like this?

--

https://issues.dlang.org/show_bug.cgi?id=23300

ag0aep6g <ag0aep6g@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ag0aep6g@gmail.com

--- Comment #4 from ag0aep6g <ag0aep6g@gmail.com> ---
(In reply to Ate Eskola from comment #3)
> ------
> auto fun(int* ptr)
> { // comment out this line and this compiles
>   // as does if you explicitly mark either it or the argument scope
>   auto r = ptr;
>   return new int;
> }
> 
> @safe int* gun(scope int* ptr){return ptr.fun;}
> ------
> 
> Is the scope inference supposed to behave like this?

That's issue 20674 ("[DIP1000] inference of `scope` is easily confused").

--

https://issues.dlang.org/show_bug.cgi?id=23300

--- Comment #5 from Ate Eskola <Ajieskola@gmail.com> ---
Thanks. The current attitude of Walter seems to be that yes, the inference is working as it should. I do aknowledge that this isn't necessarily set in stone - we might still reach consensus to change the language.

But as long as the status quo stays what it is now, this is a Phobos bug. `array` should be changed so that it compiles my first example with the current language rules.

--

https://issues.dlang.org/show_bug.cgi?id=23300

Dennis <dkorpel@live.nl> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |johanengelen@weka.io

--- Comment #6 from Dennis <dkorpel@live.nl> ---
*** Issue 23657 has been marked as a duplicate of this issue. ***

--

https://issues.dlang.org/show_bug.cgi?id=23300

Dlang Bot <dlang-bot@dlang.rocks> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |pull

--- Comment #7 from Dlang Bot <dlang-bot@dlang.rocks> ---
@dkorpel updated dlang/dmd pull request #14492 "Fix 20674, 23208, 23300 - improve `scope` inference" fixing this issue:

- Fix 20674, 23208, 23300, 23294 - improve `scope` inference

https://github.com/dlang/dmd/pull/14492

--

https://issues.dlang.org/show_bug.cgi?id=23300

johanengelen@weka.io changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |industry

--

https://issues.dlang.org/show_bug.cgi?id=23300

--- Comment #8 from johanengelen@weka.io ---
Note that this bug does not require DIP1000, and breaks compilation since
2.101.
https://issues.dlang.org/show_bug.cgi?id=23657

--

https://issues.dlang.org/show_bug.cgi?id=23300

Ate Eskola <Ajieskola@gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Depends on|                            |24003


Referenced Issues:

https://issues.dlang.org/show_bug.cgi?id=24003
[Issue 24003] return/scope inference does not end up in type to some degree
--