[postgis-users] snapToGrid strange behaviour

[Michael Fuhr]

On Thu, Jan 26, 2006 at 02:25:58PM -0800, Kevin Neufeld wrote:
> Can somebody please explain why we are getting these results?
[...]
> -[ RECORD 1 ]----------------------------------------
> geom     | 0101000000C3F5285C8FC2F33F1B2FDD2406C12340
> geom     | 0101000000C2F5285C8FC2F33F1A2FDD2406C12340
> astext   | POINT(1.235 9.877)
> astext   | POINT(1.235 9.877)
> equals   | f
> ?column? | f

I think you're encountering the usual inaccuracies of representing
floating-point numbers in binary.  Notice the difference in the
internal representations of the two geometries: in the first the
y-coordinate is 1B2FDD2406C12340; in the second it's 1A2FDD2406C12340.
Since the data is little-endian (as indicated by the leading 01),
the values are actually 0x4023C10624DD2F1B and 0x4023C10624DD2F1A,
which differ only in the lowest bit.  So instead of being exactly
9.877 the values are more like 9.8770000000000007 and 9.8769999999999989.
The calculations that yielded those numbers must have been slightly
different, resulting in that one-bit discrepancy.

-- 
Michael Fuhr