Closed jchamberlain closed 10 months ago
jchamberlain
commented
1 year ago There's more going on here than just precision. I tried getting the TestGenerateTilesForBounds test passing without fully understanding it, but now realize it's failing on arm64 but not amd64 because overflow of unsigned ints is implementation-specific (see https://github.com/golang/go/issues/52943). We'll need a different solution for that test case.
gdey
commented
1 year ago Sounds like there is still an issue here. Were you looking for advice or looking to find a new way to write this test case?
jchamberlain
commented
1 year ago Looking for feedback, including advice on other approaches. I've never used Tegola or anything go-spatial before a couple weeks ago, so I'm assuming there are probably obvious things I'm missing. Any pointers are appreciated.
TestGenerateTilesForBounds is failing in master because of a bug in go-spatial/geom which is fixed in more recent versions, but it's not a simple upgrade because many of the functions Tegola relies on have been removed. I've worked through most of it, but the one piece I'm missing now is the ability to transform a geom.Extent of SRID 3857 to SRID 4326. The geom package doesn't seem to include that capability. If such a capability already exists, can you point me to it?go-spatial/geom upgrade. geom must be doing math differently enough that floats end up being substantially different in some cases (e.g., -1.0021200154904785e+07 instead of -1.017529720390625e+07).I've opened #951 to track this better.
jchamberlain
commented
1 year ago This address the first part of #951.
I've removed the changes to TestGenerateTilesForBounds as that is handled in #952. Now this contains only precision formatting changes.
coveralls
commented
10 months ago | Totals | |
|---|---|
| Change from base Build a6f1b0cd3: | 0.0% |
| Covered Lines: | 6571 |
| Relevant Lines: | 14024 |
ARolek
commented
10 months ago I just merged this into master but the PR didn't close. ref: https://github.com/go-spatial/tegola/commit/b7ad33bafd713e72254c01786c45dc441b7c3406
The purpose of this PR is to make tests pass on Apple silicon without breaking tests for x86. The assumption (possibly incorrect) is that 8 places of precision is enough for scale demoninator, pixel width and height, and bounding box.