The expression "a.b >>>=0" or "a[b] >>>=0" will leave a lingering "a" loaded on the operand stack when generating bytecode, in many cases throwing off stack maps and ultimately failing the verifier.

This is specific to the >>>= operator, in case where the LHS is a base expression (property or element access) and the RHS is a literal zero.

The bug is in a special-case handling of RHS zero (we have a special path for "coerce to UInt32, which is basically what >>>=0 is doing) that is not loading LHS with "base already on stack" flag set and thus resulting in an additional, unnecessary load of the base.