[HLSL][AST] ArrayToPointer decay for array indexing?

Hi,

Found this while looking at the SGEP implementation on the frontend.
The following HLSL code:

void main() {
  uint array[4];
  array[2] = 12;
}

Yields this AST:

  | `-BinaryOperator 0x56193b16b770 <line:6:3, col:14> 'uint':'unsigned int' lvalue '='
  |   |-ArraySubscriptExpr 0x56193b16b718 <col:3, col:10> 'uint':'unsigned int' lvalue
  |   | |-ImplicitCastExpr 0x56193b16b700 <col:3> 'uint *' <ArrayToPointerDecay>
  |   | | `-DeclRefExpr 0x56193b16b648 <col:3> 'uint[4]' lvalue Var 0x56193b16b5c8 'array' 'uint[4]'
  |   | `-IntegerLiteral 0x56193b16b680 <col:9> 'int' 2
  |   `-ImplicitCastExpr 0x56193b16b758 <col:14> 'uint':'unsigned int' <IntegralCast>
  |     `-IntegerLiteral 0x56193b16b738 <col:14> 'int' 12

Is it by design that we have this ArrayToPointerDecay cast? When working on the SGEP emission (see https://github.com/llvm/llvm-project/pull/177332 ), I realized I needed to bypass those when emitting HLSL.

Seems to me this might be an oversight of the behavior we inherit from C/C++ no? Shouldn't we avoid those ArrayToPointer decay in HLSL since we have no pointers?

Reproduce the HLSL example and inspect the generated AST, then read the frontend SGEP implementation discussed in PR 177332. Determine whether ArrayToPointerDecay is intended for HLSL array indexing and document or implement the agreed behavior so AST construction and SGEP emission are consistent.