KaiqiJinWow commented on PR #18879:
URL: https://github.com/apache/iceberg/pull/18879#issuecomment-7052612465

   Thanks! All four follow one rule: bounds can be a bit too big, but should 
never be too small, because a box that is too small can skip a file with 
matching rows. So whenever rounding makes a case unclear, the code picks the 
bigger box. The constants aren't proven minimums. They are much larger than 
double rounding error (around 1e-16) and much smaller than anything a query 
cares about (1e-6° is about 0.1 m). I added a short comment for each in 
[91518cf](https://github.com/apache/iceberg/pull/18879/commits/91518cf6878996455ba0046761c3162d028ed25d).
   
   1. `1e-12` arc tolerance: near an endpoint, rounding can flip the sign test 
and reject a vertex that is on the arc. Accepting one just outside only widens 
the box by under a meter.
   2. `1e-6` cross-product cutoff: the length is `sin(theta)`, so a small value 
means the endpoints are almost the same or almost opposite, and the normalized 
plane is unreliable. A short arc can't change latitude by more than its length 
(about 6 m at the cutoff), so we pad by that. A nearly antipodal edge can flip 
to the other side of the globe, so only world bounds are safe.
   3. Latitude padding: `asin` loses precision near the poles, so the relative 
part grows with latitude. The absolute part covers extrema near the equator, 
where a relative margin vanishes. Both are at most about 1 m on Earth.
   4. Longitude: away from the poles, a minor arc stays within the shorter 
circular interval. Near a pole, rounding can pick the wrong side. For example, 
`134.3 - (-45.7)` rounds to exactly `180`, but the arc from `(-45.7, 60)` to 
`(134.3, 60)` passes just beside the pole, near longitude `-135.7°`. So any 
bound that reaches ±90° gets full longitude, and I added this case as a test. 
Exact pole endpoints keep their stored longitude, as you suggested on #17788.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to