https://issues.dlang.org/show_bug.cgi?id=20809

moonlightsentinel@disroot.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |moonlightsentinel@disroot.o
                   |                            |rg

--- Comment #1 from moonlightsentinel@disroot.org ---
Digger blames this PR: https://github.com/dlang/dmd/pull/10577

--

https://issues.dlang.org/show_bug.cgi?id=20809

hsteoh@quickfur.ath.cx changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hsteoh@quickfur.ath.cx

--- Comment #2 from hsteoh@quickfur.ath.cx ---
Please copy-n-paste the code in this bug, so that it won't get lost in the ether in the future.

--

https://issues.dlang.org/show_bug.cgi?id=20809

--- Comment #3 from hsteoh@quickfur.ath.cx ---
Here, let me do it for you:

------
import std.stdio;
@safe:
struct S
{
    @safe:
    int a;
    ~this()
    {
        a = 0;
    }

    ref val()
    {
        return a;
    }
}

S bar()
{
    return S(2);
}

int foo()
{
    return bar.val; // The return value of val is saved locally as ref int and
then the destructor of S is called (set the local cache to 0). Now the ref
value is dereferenced and returned
}

void main()
{
    assert(foo() == 2); // foo() is 0
}
------

Just please keep in mind next time to post the entire body of code in the bug so that it will never get lost in the future.

--

https://issues.dlang.org/show_bug.cgi?id=20809

--- Comment #4 from dlang@foerdi.net ---
(In reply to hsteoh from comment #3)

> Just please keep in mind next time to post the entire body of code in the bug so that it will never get lost in the future.

Thank you, I will keep this in my mind for the next time.

--

https://issues.dlang.org/show_bug.cgi?id=20809

Ketmar Dark <ketmar@ketmar.no-ip.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ketmar@ketmar.no-ip.org

--

https://issues.dlang.org/show_bug.cgi?id=20809

welkam <wwwelkam@gmail.com> changed:

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

--- Comment #5 from welkam <wwwelkam@gmail.com> ---
Even more reduced test case
----------------------------
struct S
{
    int a;
    ~this()
    {
        a = 0;
    }
}

void main()
{
    assert(foo() == 2); // foo() is 0
}

int foo()
{
    //bug is here
    return S(2).a; // destructor of the struct is called in this scope and sets
a to 0
}

--

https://issues.dlang.org/show_bug.cgi?id=20809

--- Comment #6 from welkam <wwwelkam@gmail.com> ---
This function works as expected
int foo()
{
    S test = S(2);
    return test.a;
}

--

https://issues.dlang.org/show_bug.cgi?id=20809

timon.gehr@gmx.ch changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |timon.gehr@gmx.ch

--- Comment #7 from timon.gehr@gmx.ch ---
While this is a regression, the underlying bug existed already in 2.089.1:

@safe:
struct S{
    @safe:
    int[8] a;
    ~this(){ a[] = 0; }
    ref val(){ return a; }
}
S bar(){ return S([2,2,2,2,2,2,2,2]); }
int[8] foo(){ return bar.val; }

void main(){ assert(foo() == [2,2,2,2,2,2,2,2]); } // error

I guess this is why review did not catch the problem (the pull request just generalized an already existing wrong transformation).

--

https://issues.dlang.org/show_bug.cgi?id=20809

Dennis <dkorpel@live.nl> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |safe
                 CC|                            |dkorpel@live.nl
           Hardware|x86_64                      |All
                 OS|Linux                       |All

--- Comment #8 from Dennis <dkorpel@live.nl> ---
This is important for making Phobos's tempCString @safe:

```
    // $(RED WARNING): $(RED Incorrect usage)
    auto pInvalid1 = str.tempCString().ptr;
    const char* pInvalid2 = str.tempCString();
    // Both pointers refer to invalid memory here as
    // returned values aren't assigned to a variable and
    // both primary expressions are ended.
```

--

https://issues.dlang.org/show_bug.cgi?id=20809

--- Comment #9 from foerdi <dlang@foerdi.net> ---
Recently I discovered an interesting fact: this error or regression has also affected the std RefCounter.

https://run.dlang.io/is/PMyy0W
```d
import std;

int foo()
{
    return refCounted(2); //(alias this) .refCountedPayload has ref return
}

void main()
{
    writeln(foo()); // foo() is corrupted
    assert(foo() == 2); // foo() is corrupted
}
```

I am surprised that this issue has not been more noticeable.

--