Incorrect results with fixed-precision buffer

[dbaston]

Copying an old issue over from the mailing list:

I confirmed this is still a bug in JTS 1.14 (trunk).

Not sure what's going on, but there's a lot of complex processing in the
buffer algorithm, and lower precision models tends to stress the robustness
of some of the processes.

On Tue, Jul 28, 2015 at 9:25 AM, Daniel Baston <dbaston@...> wrote:

Hi Martin,

I came across a strange result from a basic buffer operation today, using
a buffer distance of 3.6e-4 and the following LineString:

LINESTRING (-89.188142 48.482882, -89.186677 48.483336, -89.181966
48.483669, -89.173652 48.483456, -89.17351 48.474992, -89.167407 48.474742,
-89.171975 48.47285, -89.173831 48.471316, -89.175295 48.468853, -89.177579
48.459002, -89.178864 48.457182, -89.183325 48.454576, -89.184967
48.454683, -89.18561 48.455397, -89.193498 48.451257, -89.197424 48.454612,
-89.19878 48.455468, -89.199708 48.457039, -89.199744 48.458252, -89.2016
48.459038, -89.201814 48.460037, -89.203349 48.461001, -89.20342 48.462964,
-89.202599 48.463678, -89.202563 48.464998, -89.201171 48.465284,
-89.199958 48.466604, -89.200922 48.468817, -89.201243 48.469781,
-89.197959 48.471173, -89.194997 48.472779, -89.194854 48.473529,
-89.192641 48.475349, -89.190357 48.47592, -89.190357 48.47724, -89.188893
48.477562, -89.187716 48.479239, -89.188142 48.482882)

Running this through TestBuilder in 1.13, I get seemingly correct results
with a PrecisionModel of 1e5 or 1e7 (or full double-precision), but the
result is incorrect (missing a hole) with a PrecisionModel of 1e6. Any
thoughts on this?

Thanks,
Dan