-
Notifications
You must be signed in to change notification settings - Fork 31
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
use the new NetCDF_jll; require julia 1.3 #88
Conversation
I wanted to give the new NetCDF_jll from JuliaPackaging/Yggdrasil#1090 a test run. Since this relies on Artifacts being available it needs julia 1.3 minimum and therefore a minor version bump, to 0.11. https://github.com/JuliaBinaryWrappers/NetCDF_jll.jl Also see Alexander-Barth/NCDatasets.jl#88
Hmm on Windows this is fine apart from the compression issue mentioned above, but on Linux and OSX it complains with
|
Ok, after enabling |
Thank you for trying. Yes, I see this issue too:
I guess this is the problem: |
Hmm so the native ARM build named 1.10.5 is actually 1.10.4? |
Yes, that seems like the best option. |
@simonbyrne is it easy to test Clima on this branch with @Alexander-Barth with regard to the approach you sketch out in JuliaPackaging/Yggdrasil#1112 (comment), there is some doubt about the benefits here: JuliaPackaging/Yggdrasil#1496 (comment). |
Yes, wine is not a great solution. Here are the options as far as I can see:
Did I forget one option? I would say I favor a combination of option 2 + 3 and stop extracting binaries from msys2,h5py... Use true native builds for Windows, Mac OS and Linux x84 and qemu for Linux on ARM, aarch64 and PowerPC |
Difficult, none seem ideal. I mainly want to get off of Conda.jl and start using Artifacts, which will provide a much better user experience for installation. Yggdrasil supports just distributing prebuilt binaries, like is done now for HDF5. I guess for NetCDF in Yggdrasil we can just follow whatever is done in the HDF5 Yggdrasil build? I have been doing quite some NetCDF IO locally using this branch without any issues, so perhaps the differences in patch number are ok? That said, looking at JuliaPackaging/Yggdrasil#1496 and HDFGroup/hdf5#2 it's a bit scary. Although we could just specify that the NetCDF_jll is only compatible with a specific minor version of the HDF5_jll (1.10). |
To clarify my point 3. This is also what we can do in the future, but instead of getting the binaries from some several 3rd parties we would get it from "our" HDF5 binary build repository (for Windows and Mac OS). As far as I understand, this already the case for the native build for arm. Speaking of which, does somebody known if there is a public repo for this artefact? Another use-case to think is that a user might want to use HDF5 and NetCDF files in the same julia session. Therefore we need to use the same HDF5 libaries in NetCDF (as far as I understand dynamic linking). |
I wanted to give the new NetCDF_jll from JuliaPackaging/Yggdrasil#1090 a test run. Since this relies on Artifacts being available it needs julia 1.3 minimum and therefore a minor version bump, to 0.11. https://github.com/JuliaBinaryWrappers/NetCDF_jll.jl Also see Alexander-Barth/NCDatasets.jl#88
I wanted to give the new NetCDF_jll from JuliaPackaging/Yggdrasil#1090 a test run. Locally all tests pass except the ones updating the compression level to 4, they seem to stay at 9. Since this relies on Artifacts being available it needs julia 1.3 minimum and therefore a minor version bump, to 0.11. https://github.com/JuliaBinaryWrappers/NetCDF_jll.jl
This reverts commit b4cbe9f.
I assume this will be more efficient, although I haven't done benchmarks. This should be still done throughout this file.
to see if otherwise CI passes
This is now updated with a new NetCDF_jll build, that uses a new HDF5_jll build by @musm that upgraded things consistently to 1.12, so there are no header mismatch error anymore here. This gives all green CI, with the note that I marked a test as broken. This is the test at the end: # check compression
isshuffled,isdeflated,deflate_level = deflate(v)
@test isshuffled == true
@test isdeflated == true
@test deflate_level == 9
# change compression
deflate(v,false,true,4)
isshuffled,isdeflated,deflate_level = deflate(v)
@test isshuffled == false
@test isdeflated == true
@test_broken deflate_level == 4 Instead the I just looked around and found Unidata/netcdf-c#1713. So it's a known issue in 4.7.4 that will be fixed in the next version. Since calling |
Fantastic! |
This is great ! Thanks a lot @visr ! |
* use the new NetCDF_jll; require julia 1.3 I wanted to give the new NetCDF_jll from JuliaPackaging/Yggdrasil#1090 a test run. Since this relies on Artifacts being available it needs julia 1.3 minimum and therefore a minor version bump, to 0.11. https://github.com/JuliaBinaryWrappers/NetCDF_jll.jl Also see Alexander-Barth/NCDatasets.jl#88 * update CI versions * update docs projects This adds NetCDF itself as a dependency, just like is done in PkgTemplates. That means Conda is still a dependency for now until we tag a new release. * add NetCDF_jll to deps * update Documenter to 0.25 With prettyurls set to false for local builds, such that you don't need to run a local server to click around.
I wanted to give the new NetCDF_jll from JuliaPackaging/Yggdrasil#1090 a test run.
Locally all tests pass except the ones updating the compression level to 4, they seem to stay at 9.
Since this relies on Artifacts being available it needs julia 1.3 minimum and therefore a minor version bump, to 0.11.
https://github.com/JuliaBinaryWrappers/NetCDF_jll.jl