[REG 2.112.0] Duplicate __aaget declaration for simple expression with associative array

[vindexbit]

void main() {
    int[][string] aarr;
    aarr["key"].length++;
}

Compilation fails:

../probe/bug_aaget.d(3): Error: declaration `bug_aaget.main.__aaget3` is already defined
    aarr["key"].length++;
    ^
../probe/bug_aaget.d(3):        `variable` `__aaget3` is defined here
    aarr["key"].length++;
    ^

Version 2.111.0 didn't have this fail. Unfortunately, the ldc2 and gdc compilers were too quick to incorporate the changes from 2.112+.

[vindexbit]

This version of the example is better:

void main() {
    int[][string] aarr;
    aarr["key"] = [];
    aarr["key"].length++;
}

But the error is the same.

[rainers]

The lowering of array.length++ seems broken, it evaluates the array expression twice, even without associative arrays involved:

int[] arr()
{
	import core.stdc.stdio;
	printf("hi\n");
	return null;
}

void test2() {
	arr().length++;
}

prints "hi" twice. arr().length += 1 is ok.

[vindexbit]

The lowering of array.length++ seems broken, it evaluates the array expression twice, even without associative arrays involved:

int[] arr()
{
import core.stdc.stdio;
printf("hi\n");
return null;
}

void test2() {
arr().length++;
}

prints "hi" twice. arr().length += 1 is ok.

I doubt this behavior is a problem.
arr().length++ is equal to arr().length = arr().length + 1; (likely)

The regression I described is most likely related to the new implementation of associative arrays

[rainers]

It is lowered to tmp = arr().length, ++arr().length, tmp, but arr() is still unexpected to be called twice. The issue is triggerd by the new AA lowering, but I suspect you can get similar compile errors with lowerings involving temporaries.

Here is a fix: #23189