ObservationFromGrid is centered on an odd point

[TGlas]

Let (x,y,z) denote the position of the character as reported by the XPos, YPos, and ZPos observations. These are floating point numbers, not integer-valued block coordinates. I assume that the numbers refer to the character's feet; at least the y coordinate is an integer while standing on a solid block.

Apparently (I have collected quite some data), the grid of blocks returned by ObservationFromGrid is centered on the block with the following coordinates:

           (floor(x), round(y), floor(z))

In other words, truncation is applied in horizontal directions, while rounding to the closest integer is applied in the vertical direction. This odd combination resulted in some very subtle bugs in my code (trying to perform temporal integration).

I cannot be 100% sure that rounding is what's actually going on here. It may also be that the grid is centered on a point located rougly half a block above the feet. However, it doesn't seem to be the eye's, which would be the other logical point, since that should be about 1.5 blocks above the feet. That's why I presume a rounding issue.

I would be very much surprised if the current behavior was by design, so I assume it is a bug. I'd like to strongly suggest to apply the floor operation relative to the XPos, YPos, and ZPos observations in all three coordiantes. This seems to be the most natural behavior, probably expected by most users.

[DaveyBiggers]

That does indeed sound like a bug. I'll take a look.

[DaveyBiggers]

Yes, for some reason 0.5 gets added to the y value before the floor takes place.
I've fixed this - ca9f416 - hopefully no one is relying on the wrong behaviour...

[TGlas]

Thanks a lot, works perfectly!