Ecosystem-wide testing

[tiemvanderdeure]

As I proposed in the geospatial community meeting, we could use this package for ecosystem-wide testing.

For example:

• packages do not export the same names (Name overlaps #1)
• conversion between types works (e.g. call Raster on a YAXArray)
• interfaces are correctly implemented (e.g. writing a lazy raster doesn't use chunks rafaqz/Rasters.jl#889)

We don't have to add all of this at once - if an ecosystem-related bug comes we could add a test here and incrementally build a test suite.

[rafaqz]

It would be nice if we could run @JuliaGeoBot test in a branch of any package and run the Geo tests with the package devved to the branch. So multi-package tests don't all run all the time but so we can check them when a PR is done

Not sure how much work it is to write such a bot

[asinghvi17]

Yeah I've been mulling over this as well, might try it.

[evetion]

That's actually a reverse dependency test bot, those exist for testing new Julia versions, and we really need one for certain jll packages.

[felixcremer]

For the tests we should set up common test data artifacts. For small data we could set up a dedicated JuliaGeoRasterTestData and JuliaGeoVectorTestData repositories where we could capture all datasets and reuse them from different packages.
And for larger datasets that could live on an S3 these could be added to our institutes S3 server. There is no promise that this serer will run forever, but it ran for multiple years already and the chances are high that it just keeps set up.

We could capture the datasets that would need to find a better home here so that we can then add them to our S3.
@meggart

[asinghvi17]

Agreed, although I wonder if we should just keep a single JuliaGeoTestData.jl? Name may also need a rethink (GeoTestData?) because I don't think you can register a package with Julia in the name...

[felixcremer]

This would not be registered as a package. We would use this to provide a tarball which can be used in Artifacts. toml files in package tests.

[asinghvi17]

Ah, my thought was that we have everything as a lazy artifact that can be addressed by name in the package - then it becomes extremely easy to load just by GeoTestData.artifact_name. No mucking about with changing checksums or whatnot if a new version of the data is released

[asinghvi17]

fwiw, we also have https://github.com/JuliaGeo/JuliaGeoData that already exists - perhaps we could update that...