Fix reverse of ranges of unsigned integers

[mbauman]

Fixes #29576, works around improves #29810.

Edit: well, works around might be a little strong. It changes a segfault to an inexact error, so this doesn't add support for reversing ranges of pointers (hence the lack of tests thereof).

Closes #42718

[chethega]

Cool, inexact is very preferable to segfault, and your PR points out the problem:

function unsafe_length(r::StepRange)
    n = Integer(div((r.stop - r.start) + r.step, r.step))
    isempty(r) ? zero(n) : n
end

should maybe become

function unsafe_length(r::StepRange)
    if r.step > 0
        n = Integer(div((r.stop - r.start) + r.step, r.step))
    else
       n = Integer(div(-(r.stop - r.start) - r.step, -r.step))
    end
    isempty(r) ? zero(n) : n
end

which would fix the pointer range. On the other hand, this still smells very brittle. Do we need to support ranges with larger than Int64 length?

[vtjnash]

I've expanded this more, and think we should merge, but adding triage label in case someone wants to contest it. I noticed after pushing this that there might also be a case in intersect that I missed switching from -step to negate(step), which might be good to add here too.

[vtjnash]

:D

Error in testset sorting:
Error During Test at /buildworker/worker/tester_linux32/build/share/julia/test/sorting.jl:220
 Unexpected Pass
 Expression: length(reverse(0x01:0x02)) == 2
 Got correct result, please change to @test if no longer broken.

[oscardssmith]

Is this ready to merge?

[vtjnash]

It has the triage label, but we could also merge now, and then see if anyone has feedback