[CLN-list] cln 1.3.5

[Richard B. Kreckel]

Hi,

It's good hearing from you again, Alexey!

On 12/4/19 6:19 PM, Alexey Sheplyakov wrote:
> This code relies on integer multiplication overflow to find out
> the bit size of various integer types. However singed integer overflow
> is an undefined behavior. Apparently in some contexts GCC uses mod 2^N
> arithmetics to evaluate the signed integer expressions. As a result
> `typedef int[2*((T)((T)2 * ... * (T)2) == 0) - 1]` trick works with both
> signed and unsigned integers (and gives the correct result).
>  
> Howerver GCC considers an expression involving an integer overflow as
> a non-constant, and refuses to use it in `static_assert`

Right. According to section 5.19 in the C++11 standard (but the
equivalent section 8.20 in C++17 is much more clear), an "integral
constant" is implicitly converted to a prvalue where the converted
expression is a "core constant". And by the same standard sections it's
not a "core constant" if it has undefined behavior like signed integer
overflow. I guess that explains why static_assert doesn't work here.

But I find it interesting that the C++ requirements on "integral
constants" / "core constants" do not seem to apply to the "condition" in
typedef int[2*(condition) - 1] trick. I'm unsure if this is really
compliant.

Anyway, I've applied your patch.

All my best,
  -richy.
-- 
Richard B. Kreckel
<https://in.terlu.de/~kreckel/>