• Teodor Dutu (30/30) Oct 31 2023 Hi,

Teodor Dutu <teodor.dutu gmail.com> writes:

Hi,

Continuing my work on fixing [this 
bug](https://issues.dlang.org/show_bug.cgi?id=24159), it turned 
out that I was incorrectly modifying the original expression like 
this:

```d
arr ~= a + b;

// was lowered to:
__tmp = a + b; _d_arrayappendcTX(arr, 1), arr[$ - 1] = tmp;

// but was also changed to:
arr ~= __tmp;
```

Creating the temporary variable was necessary for situations such 
as `arr ~= arr[$ - 1]`. Without it, the lowering would have been 
`_d_arrayappendcTX(arr, 1), arr[$ - 1] = arr[$ - 1]`. Note that 
in the second expression `arr[$ - 1]` is no longer referring to 
the same element as in the original expression, but to the newly 
created element, which is uninitialised.

However, the fix was too hard for the problem it was trying to 
solve. Simply modifying the `IndexExp` in the lowering from `$ - 
1` to `$ - 2` was enough and more efficient at the same time. 
This got rid of the backend bug where the `__tmp` variable was 
unrecognised.

However, now this [function from 
DRuntime](https://github.com/dlang/dmd/blob/49690857b45f3f378ff84ff12ece51f6248a0303/druntime/src/co
e/demangle.d#L2141) seems to be visited twice in the glue layer. This makes its
`vthis` to be visited twice as [this
`assert`](https://github.com/dlang/dmd/blob/49690857b45f3f378ff84ff12ece51f6248a0303/compiler/
rc/dmd/glue.d#L949) is triggered. I haven't found the source of this error yet.

I also spent some time on convering `_d_newarraym{i,}TX` to 
templates and hope I'll be able to fix the final bugs and create 
a PR for these hooks by the end of this week.

Thanks,
Teodor