Re: [PATCH v2 1/2] gdb: Avoid undefined shifts

[Tom Tromey <tom@tromey.com>]

>>>>> "Pedro" == Pedro Alves <pedro@palves.net> writes:

Pedro> Compilers don't error out on such shifts, at best they warn, so I
Pedro> think GDB should just continue doing the shifts anyhow too.

Pedro> - Adjust scalar_binop to avoid the undefined paths, either by adding
Pedro>   explicit result paths, or by casting the lhs of the left shift to
Pedro>   unsigned, as appropriate.  For the shifts by negative amount or a
Pedro>   too-large amount, I made the result be 0 (and warn).  This is
Pedro>   undefined, so we're free to pick any value, and 0 seems reasonable,
Pedro>   as it's what you'd get if you split a too-large shift by a number of
Pedro>   smaller valid shifts, and if you imagine a shift by a negative value
Pedro>   as a shift by a very-large unsigned value, you get the same too.

Pedro> - Make GDB warn for the cases that are still undefined in C++20.  The
Pedro>   warnings' texts are the same as what GCC prints.

I wonder whether any of the supported languages define shifts differently.
Rust is a bit vague about these cases, but I know Java had different
semantics (but OTOH Java would have had to implement its own expression
node for this anyway).

Anyway what I'm wondering is whether the warnings are always appropriate.

Pedro> +      warning (_("right shift count is negative"));

Pedro>  	    case BINOP_LSH:
Pedro> -	      v = v1 << v2;
Pedro> +	      if (!check_valid_shift_rhs (result_type, v2))

This is a left shift but the warning will mention a right shift.

Tom