Switch from Requires to extensions

[GiggleLiu]

Added two extensions

• TensorInferenceCUDAExt = "CUDA"
• TensorInferenceGTNExt = "GenericTensorNetworks"

[codecov]

Codecov Report

Merging #81 (75a021f) into main (248bdb5) will decrease coverage by 0.41%.
Report is 1 commits behind head on main.
The diff coverage is 50.00%.

@@            Coverage Diff             @@
##             main      #81      +/-   ##
==========================================
- Coverage   84.70%   84.29%   -0.41%     
==========================================
  Files          10       10              
  Lines         536      535       -1     
==========================================
- Hits          454      451       -3     
- Misses         82       84       +2     

Files	Coverage Δ
ext/TensorInferenceGTNExt.jl	90.90% <100.00%> (ø)
src/TensorInference.jl	100.00% <ø> (ø)
src/mar.jl	94.54% <100.00%> (ø)
ext/TensorInferenceCUDAExt.jl	0.00% <0.00%> (ø)
src/utils.jl	88.97% <0.00%> (-0.71%)	⬇️

[mroavi]

I had an issue trying to run the tests on my machine:

(TensorInference) pkg> test
julia> ERROR: Compat `CUDA` not listed in `deps` or `extras` section.

[GiggleLiu]

I had an issue trying to run the tests on my machine:

(TensorInference) pkg> test
julia> ERROR: Compat `CUDA` not listed in `deps` or `extras` section.

What is your Julia version? The extensions feature requires Julia > 1.9. CUDA is listed in weakdeps, however your Julia does not check this section.

[mroavi]

Hmm, I see. I have Julia 1.8.5 installed on my machine. Given the requirement for Julia > 1.9 for the extensions feature, should we consider dropping support for the 1.8.x versions? Or is there a workaround to make it compatible with Julia 1.8.x?

[GiggleLiu]

The backward compatibility is discussed here: https://pkgdocs.julialang.org/v1.9/creating-packages/#Backwards-compatibility

However, I am against using compatibility treatment. For some users who want to stick to older version Julia, they can use older version of TensorInference.

We can either merge this PR directly, or merge it after the next julia LTS release. The previous LTS version is 1.6.

[mroavi]

Since we are planning to drop support for older Julia versions, I propose that we merge this once the next LTS version is released. Is there any way we could improve the error message above? The current error message does not clearly indicate that TensorInference.jl does not support versions prior to 1.9, even though we have specified 'julia = "1.9"' in the Project file.

[GiggleLiu]

Since we are planning to drop support for older Julia versions, I propose that we merge this once the next LTS version is released. Is there any way we could improve the error message above? The current error message does not clearly indicate that TensorInference.jl does not support versions prior to 1.9, even though we have specified 'julia = "1.9"' in the Project file.

I think the package simply does not appear in the General registry in julia version < 1.9. If you are using Julia 1.8, it will fetch older versions for you. FYI: I have droped <1.9 version support in the latest OMEinsum.

But I agree we should at least wait for 1.10 (not nessesaryly LTS), which is another significant improvement comparing with 1.9. Then more people will switch to newer version.

[GiggleLiu]

Hi @mroavi , I factored out the problem specification in GenericTensorNetworks. Now they are specified in ProblemReductions.jl, a light weighted package that contains a lot of models that widely used in statistical physics, such as the SpinGlass and IndependentSet.
So I made it a dependency instead of an extension. It would be great if you could review this PR again and get it merged.

I have also tested the CUDA extension again on my local host to make sure it still works. Also, Julia v1.10 is LST, so I believe most users will accept the extension now. (Instead of Requires.jl)

[GiggleLiu]

Hi Martin, I have to merge this PR and tag a new version since I need to use this feature in a course tomorrow. You can still take time to review this PR, and I will make the change in a new PR. Sorry about that.

[mroavi]

Hi @GiggleLiu . I see you merged it already which is totally fine. I'm currently quite overloaded with my new job as teacher so I won't be able to react promptly.