[REG master] `aa[key] ~= null` discarded

[CyberShadow]

void main()
{
	int[][][string] aa;  // AA with nested array value type
	aa["x"] = null;
	aa["x"] ~= null;     // This append is silently lost!
	assert(aa["x"].length == 1);  // FAILS: length is still 0
}

Introduced in #21066, I think.

[thewilsonator]

cc @rainers

[rainers]

The type of the appended null is ambiguous, it used to be interpreted as int[], but now it is int[][], so an empty array is appended instead of adding an element. The latter is still happening when appending null to int[][] a, so it must be somewhere in the lowering of the AA...

A workaround is to cast null to the desired type.

[CyberShadow]

I'm pretty sure the new behavior is wrong because it is inconsistent with

void main()
{
	int[][] a;
	a ~= null;
	assert(a.length == 1);
}

so this is just a clear-cut regression rather than maybe-breaking-improvement.

[rainers]

Sure, I tested that, too. I'll have to investigate...