Test out unbreak of R build

[johnkerl]

Context:

• Nightly feedstock build failed #126
• [r] Use arrow via Suggests: only easing the build requirements single-cell-data/TileDB-SOMA#2415

[johnkerl]

@jdblischak at
https://dev.azure.com/TileDB-Inc/CI/_build/results?buildId=38877&view=logs&j=58ac6aab-c4bc-5de2-2894-98e408cc8ec9&t=933f325c-924e-533d-4d95-e93b5843ce8b

- package libtiledbsoma-1.9.5rc0-haede419_0 requires tiledb >=2.21.2,<2.22.0a0, but none of the providers can be installed

-- I think I need a tiledb/for-cloud feedstock PR perhaps ... 👀

[johnkerl]

Derp, nope ... https://github.com/TileDB-Inc/tiledb-feedstock-for-cloud/pull/22 🤔

[johnkerl]

@jdblischak additionally I'm confused by the solver error

└─ libtiledbsoma 1.9.5rc0 haede419_0 is not installable because it requires
   └─ tiledb >=2.21.2,<2.22.0a0 , which requires
      └─ aws-crt-cpp >=0.26.4,<0.26.5.0a0 , which conflicts with any installable versions previously reported.

since in the recipe/meta.yaml in this repo we have only

tiledb >=2.21.0,<2.22

🤔

[johnkerl]

Also puzzled this is mac-only, not Linux ... :think

[johnkerl]

At https://anaconda.org/conda-forge/tiledb/files I see for example

| [osx-64/tiledb-2.21.2-h0d80af6_0.conda](https://anaconda.org/conda-forge/tiledb/2.21.2/download/osx-64/tiledb-2.21.2-h0d80af6_0.conda)

🤔

[johnkerl]

• This has 2.21.* https://github.com/conda-forge/r-tiledb-feedstock/blob/main/recipe/meta.yaml#L34
• This PR has 2.21.*

@jdblischak @ihnorton where does

└─ libtiledbsoma 1.9.5rc0 haede419_0 is not installable because it requires
   └─ tiledb >=2.21.2,<2.22.0a0 , which requires

come from?

[jdblischak]

tiledb 2.21.2 comes from the "run exports" of the tiledb recipe. Since libtiledbsoma was built against tiledb 2.21.2, it requires tiledb >=2.21.2,<2.22. This is the same problem we had last week with the cloud Docker files (which were pinning tiledb 2.21.1). For the cloud Docker files, we just had to update the TileDB env var pin to 2.21.2. In this case we'll need to migrate the aws version, so hopefully the solver error will be fixed by conda-forge/tiledb-feedstock#271

[jdblischak]

I just merged conda-forge/tiledb-feedstock#271. The build to monitor is at conda-forge/tiledb-feedstock@95e102d

[johnkerl]

@jdblischak OK thank you!

Meanwhile TileDB-SOMA -- as a package outside of Conda -- depends on 2.21.1.

I tried single-cell-data/TileDB-SOMA#2424 but I found that TileDB-Py https://github.com/TileDB-Inc/TileDB-Py/releases/tag/0.27.1 is getting 2.21.1 and there is as yet no TileDB-Py 0.27.2 using 2.21.2. I'm not sure what's going to happen next but I'll continue working on this.

[jdblischak]

Meanwhile TileDB-SOMA -- as a package outside of Conda -- depends on 2.21.1.

If that is so, we would need to update all the pins in meta.yaml to reflect this stricter requirement

I found that TileDB-Py https://github.com/TileDB-Inc/TileDB-Py/releases/tag/0.27.1 is getting 2.21.1 and there is as yet no TileDB-Py 0.27.2 using 2.21.2

Have we confirmed that is a strict requirement? While tiledb-py 0.27.1 bundles tiledb 2.21.1 (eg in its wheels uploaded to PyPI), it's my understanding that 0.27.1 should be compatible with any tiledb 2.21.*. That is how the tiledb-py-feedstock operates, so if my assumption is untrue, we should also fix that recipe

https://github.com/conda-forge/tiledb-py-feedstock/blob/0656ff28f3f855c63d993790c38c1733bfa1e55e/recipe/meta.yaml#L35

As further evidence, the only difference between tiledb-py 0.27.0 and 0.27.1 is the bundled version of tiledb, which increases my confidence in my assertion that the 0.27.1 release was purely a mechanism for deploying tiledb 2.21.1 via the PyPI wheels

TileDB-Inc/TileDB-Py@0.27.0...0.27.1

[ihnorton]

it's my understanding that 0.27.1 should be compatible with any tiledb 2.21.*

Correct

[johnkerl]

I found that TileDB-Py https://github.com/TileDB-Inc/TileDB-Py/releases/tag/0.27.1 is getting 2.21.1 and there is as yet no TileDB-Py 0.27.2 using 2.21.2

Have we confirmed that is a strict requirement? While tiledb-py 0.27.1 bundles tiledb 2.21.1 (eg in its wheels uploaded to PyPI), it's my understanding that 0.27.1 should be compatible with any tiledb 2.21.*. That is how the tiledb-py-feedstock operates, so if my assumption is untrue, we should also fix that recipe

@jdblischak your theoretical reasoning is 100% solid & I wish it were that way. However on single-cell-data/TileDB-SOMA#2424 (comment) I found that 2.21.1 was actually getting found (admittedly, outside of Conda).

We need to thread today's needle with regard to:

• [c++] Use core 2.21.2 [pending TileDB-Py 0.27.2] single-cell-data/TileDB-SOMA#2424 (comment)
• Rebuild for aws_crt_cpp 0.26.6 conda-forge/tiledb-feedstock#271 (thank you!)
• The problem originally reported on this PR

and see if we can get green CI on this feedstock PR. It's not clear to me yet how this will all unfold.

[jdblischak]

I found that 2.21.1 was actually getting found (admittedly, outside of Conda).

I don't think that should affect the conda builds. That upstream CI job installed tiledb-py from a PyPI wheel, which bundles its own copy of libtiledb. I agree this is a general problem with non-conda builds, and something I continuously struggle with in my various nightly builds (ie the nightly builds only make sense if all the clients are using the same underlying version of the nightly libtiledb). But with the various conda builds, they will all be linked to the conda-installed libtiledb

Using cached tiledb-0.27.1-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl.metadata (3.6 kB)

[johnkerl]

This seems to be superseded by #129 so I'll abandon this attempt.

[johnkerl]

single-cell-data/TileDB-SOMA#2414

[sc-44854]