Re: [boost] Current Guidance on Compiler Warnings?

Subject: Re: [boost] Current Guidance on Compiler Warnings?
From: Peter Dimov (lists_at_[hidden])
Date: 2018-11-27 00:24:40

Gavin Lambert wrote:

> In this case for any non-massive array you're fairly safe (esp. when 
> someone fixes the call site) with a simple:
>
>    assert(i < size());

That is correct. If you have an upper bound, `i < n` for unsigned `i` is 
equivalent to `i >= 0 && i < n` with a signed `i`, so in this specific case 
you can use either.

Signed is still preferable though because it's less surprising. `x > -1` 
gives you the normal answer, for instance. But, you'll say, this will be a 
warning. Well yes, this specific case will be, but not all will. Some 
signed/unsigned mismatches don't warn on purpose because there are too many 
false positives (https://godbolt.org/z/c1rzjS), and in some cases, such as 
with the difference between (a-b)/(c-d) and (b-a)/(d-c), unsigned finds a 
way to ruin your day without any signed/unsigned mismatches at all.

So it's a long-standing guideline to never use `unsigned`, except for 
bitwise operations and modular arithmetic. For numbers, signed is the way to 
go.