Float.init shouldn’t be nan

[drpriver]

See title

[IgorDeepakM]

Yes, this is something we can now change. I'm kind of indifferent if default should be Nan or 0 when it comes to computations but I don't use floats that often.

However, there are other reasons I think init for floats should be 0.

1. About in all other languages, the default value is 0 including C#. This gives programmers a familiarity that also should be found in OpenD.
2. It reduces code size. CPU often have special 0 instructions and float zero areas can be coalesced with other zero memory areas.
3. There are probably other benefits too, which you probably know more about.

EDIT. I can now edit my own posts. Finally. Such luxury!

[Lyratelle]

On the contrary.
int.init should be also NaN = 0x8000_0000, and int.min changed to 0x8000_0001, so that the range is symmetric (abs(int.min)==int.max) and we have an invalid value for integers that would often be useful.
Bitmanipulation should be reserved for unsigned types (which are less useful for general calculations), and only there using all possible bitpatterns as valid values makes sense.

[adamdruppe]

I'm inclined to make it zero... moving int to a nan is also a good idea but can you imagine how annoying that would be to users who are used to the simple zero?

[Lyratelle]

For D users? Can't be so bad, as it is the same with float and char... :-)

[quickfur]

This would be a big breaking change, since a lot of existing code implicitly relies on int.init being 0. I know my code has a lot of that. I know, I know, it's a bad habit... but I wouldn't like it if suddenly my existing projects start having subtle breakages because int.init is no longer 0.

With float.init the breakage is less severe because generally if your project is already producing results it's generally not banking on float.init being NaN (since presumably the output wouldn't be all NaN's). Changing it to 0 might break some projects, but probably not across the board like changing int.init would.

[Lyratelle]

Oh, the break wouldn't be subtle. If your code suddenly starts producing a lot of NaNs, maybe you should add " = 0" near the end of your variable declarations. Will even make your code more readable.
If this change is inconvenient for you, I hope you will never be confronted with abominations like QAC, POLYSPACE or MISRA. :-)