Reenable possibility to use an already installed C-Blosc library

[bnavigator]

This fixes #216, by introducing a USE_SYSTEM_BLOSC environment variable. Custom install locations are supported through CMake's standard settable Blosc_ROOT, Blosc_INCLUDE_DIR, etc. variables for finding modules.

For consistency with the standard CMake way, I decided against retaining the defunct BLOSC_DIR variable.

BTW, the build_clib and build_ext commands seems to do nothing on my installs, everything is performed by scikit-build during the build stage. Might have the same reason as #222.

[sreerajkksd]

@FrancescAlted @thewtex Can one of you review this PR ? Does this approach look sane to you guys ?

[thewtex]

Hi @bnavigator @sreerajkksd @FrancescAlted I explained more in #216, but improvements could be made by using a CMake package configuration file.

[FrancescAlted]

Any possibility that the PR adopts the suggestions by @thewtex in #212? Also, the PR has conflicts and needs an update.

[bnavigator]

If I understand correctly, @thewtex suggests to create a CMake package file instead of this PRs FindBlosc.cmake. This would go into c-blosc, not python-blosc. Way out of scope of this PR.

[FrancescAlted]

This looks reasonable to me (unless @thewtex would propose otherwise).

[FrancescAlted]

Let's keep this on-hold until/if @thewtex would chime in.

I have been convinced of the weird CMake name conventions

[FrancescAlted]

I have just a trivial comment

[thewtex]

@thewtex suggests to create a CMake package file instead of this PRs FindBlosc.cmake

Yes, we should definitely be using a CMake package configuration file instead of a Find module. But, if @bnavigator is not willing to make those changes, this is still a step forward. We could merge this and make additional progress later.

[sreerajkksd]

I would recommend merging this now (as we have approval from the maintainers) and creating a new PR for the concern mentioned. I can see that @bnavigator already created an issue for this. I believe this is going to help some people (including me) to compile with an existing c-blosc library.

[sreerajkksd]

@FrancescAlted Can you cut a new release in Pypi ?

[mgorny]

I'm confused now. So building python-blosc against the system library now requires c-blosc with a CMake package file that's not yet provided by c-blosc?

[bnavigator]

python-blosc provides its own FindBlosc.cmake for the time being. All should be there.

[mgorny]

I'm sorry, my bad. However, the FindBlosc module is missing from the pypi sdist. After using the raw git archive from GitHub, it works.

[eamanu]

HI,

I'm packing python-blosc 1.10.6 in Debian. I wanted to be sure if this PR won't
re-introduce the issue #227 and fixed here https://github.com/Blosc/python-blosc/pull/242/files.

Thanks!
Cheers,