FramePos improvements

[Holzhaus]

Some minor improvements to the FramePos class added in #4043.

[Holzhaus]

This should ensure that all calculations with frame positions are well defined, and we never end up with weird Nan values in the engine.

[uklotzde]

I am still worried about the inconsistencies introduced by the special equality comparisons ==/!= that are incompatible with </<=/>/>=.

How about reverting them and only adding the assertion DEBUG_ASSERT(lhs.isValid() == rhs.isValid())? No, that doesn't make any sense ;)

[Holzhaus]

Actually the striked through suggestion from your comment is already the case.

The comparison operators all use value() which raises a debug assertion if the value is invalid. Exceptions are == and !=. There return true if both positions are valid and their value matches, or both are invalid. Otherwise they return false.

[Holzhaus]

What kind of issues do you expect?

IMHO allowing comparisons of valid and invalid values is more confusing. If you
make a check loopStart <= loopEnd and due to some engine bug, loopEnd is set to infinity (and the FramePos is invalid), the check would return true and you won't notice. With this implementation, a debug assertion is raised, which indicates your assumption that both values are valid is wrong.

Do you have a use case for comparisons between valid and invalid values, or two invalid values?

[uklotzde]

I didn't consider the debug assertions that already prevent using </<=/>/>= with invalid values. Then there should be no inconsistencies.

[uklotzde]

Thank you! LGTM