sparse_set_iterator::operator[] causes warnings when used with size_t about entt

[skypjack]

Interesting. I've clang-tidy enabled and cannot reproduce it though.
Can I ask you what param triggers the error? Maybe it's not the only one.

[jonesmz]

It's this PR: https://github.com/TheOpenSpaceProgram/osp-magnum/pull/269/files , the file test/tasks/main.cpp which is the very bottom-most file in the review (at the time of writing).

The deprecation message says to switch from ::at(index) to ::begin()[index], but i was actually able to switch to simply ::operator[](index) directly.

However, originally (before i just switched to ::operator[](index))when i switched from ::at(index) to ::begin()[index] i got the warning from clang tidy i reported in my initial comment.

The types are:

'entt::internal::sparse_set_iterator<std::vector<osp::TaskId>>::difference_type', where osp::TaskId is an enum-class backed by a uint32_t.

https://github.com/TheOpenSpaceProgram/osp-magnum/blob/a747cab17ae3f5bea02a800c8b9cfbe8a1e345c3/src/osp/tasks/tasks.h#L67-L72

difference_type should be std::ptrdiff_t but apparently cpp reference thinks the only requirement is that it's a signed integer. Regardless, it's being calculated on my platform as long.

rRand is type std::mt19937 &rRand defined here: https://en.cppreference.com/w/cpp/numeric/random/mersenne_twister_engine

which apparently returns std::uint_fast32_t and i guess on the github actions runner is of type unsigned long.

runTasksLeft is auto const runTasksLeft = rExec.tasksQueuedRun.size(); which i imagine is std::size_t, though i'm not 100% sure what the type of tasksQueuedRun is off of the top of my head.

The full function is here:

https://github.com/TheOpenSpaceProgram/osp-magnum/blob/a747cab17ae3f5bea02a800c8b9cfbe8a1e345c3/test/tasks/main.cpp#L51-L73

since unsigned long can't convert to long without precision loss, we get the warning about narrowing conversions.

My recommendation is that you can either make the functions for sparse_set_iterator templated on the std::integral concept, or you can define two versions, one for signed integers and another for unsigned.

But don't fix this on my account, since ::operator[](index) works for me just fine. I just wanted to report the warning in case you wanted to fix it.

[skypjack]

Ohoo, I see, so the error is at the call site in your case and not in the body of the function, is it?
I got it the other way around initially. This would explain why it doesn't trigger on the CI at least.

[skypjack]

I checked the standard and a few implementations for the major compilers.
As expected, operator[] for random access iterators is always defined in terms of difference_type.
Therefore, I think I'll stick with the current implementation. Thanks for bringing it up though. 👍

[jonesmz]

Do you mind sharing some of the random access iterators that you were looking at? I'm curious.

[skypjack]

You can check a few implementations here (working draft for the language). Then I checked the MSVC one from within VS (just look into the implementation of a container like the vector) and the ones offered by clang (easily available online). 👍