missing tensorflow-2.18 on linux-aarch64

[dslarm]

Solution to issue cannot be found in the documentation.

• I checked the documentation.

Issue

conda install tensorflow=2.18
Channels:
 - conda-forge
 - bioconda
 - defaults
Platform: linux-aarch64
Collecting package metadata (repodata.json): done
Solving environment: failed

PackagesNotFoundError: The following packages are not available from current channels:

  - tensorflow=2.18*

Current channels:

  - https://conda.anaconda.org/conda-forge
  - https://conda.anaconda.org/bioconda
  - defaults

To search for alternate channels that may provide the conda package you're
looking for, navigate to

    https://anaconda.org

and use the search bar at the top of the page.

the package is there for other architectures, according to anaconda

Installed packages

(banana) [rocky@ip-10-11-36-193 ~]$ conda list
# packages in environment at /home/rocky/miniforge3/envs/banana:
#
# Name                    Version                   Build  Channel

Environment info

(banana) [rocky@ip-10-11-36-193 ~]$ conda info

     active environment : banana
    active env location : /home/rocky/miniforge3/envs/banana
            shell level : 2
       user config file : /home/rocky/.condarc
 populated config files : /home/rocky/miniforge3/.condarc
                          /home/rocky/.condarc
          conda version : 24.11.3
    conda-build version : not installed
         python version : 3.12.8.final.0
                 solver : libmamba (default)
       virtual packages : __archspec=1=neoverse_n1
                          __conda=24.11.3=0
                          __glibc=2.34=0
                          __linux=5.14.0=0
                          __unix=0=0
       base environment : /home/rocky/miniforge3  (writable)
      conda av data dir : /home/rocky/miniforge3/etc/conda
  conda av metadata url : None
           channel URLs : https://conda.anaconda.org/conda-forge/linux-aarch64
                          https://conda.anaconda.org/conda-forge/noarch
                          https://conda.anaconda.org/bioconda/linux-aarch64
                          https://conda.anaconda.org/bioconda/noarch
                          https://repo.anaconda.com/pkgs/main/linux-aarch64
                          https://repo.anaconda.com/pkgs/main/noarch
                          https://repo.anaconda.com/pkgs/r/linux-aarch64
                          https://repo.anaconda.com/pkgs/r/noarch
          package cache : /home/rocky/miniforge3/pkgs
                          /home/rocky/.conda/pkgs
       envs directories : /home/rocky/miniforge3/envs
                          /home/rocky/.conda/envs
               platform : linux-aarch64
             user-agent : conda/24.11.3 requests/2.32.3 CPython/3.12.8 Linux/5.14.0-427.37.1.el9_4.aarch64 rocky/9.4 glibc/2.34 solver/libmamba conda-libmamba-solver/24.9.0 libmambapy/1.5.12
                UID:GID : 1000:1000
             netrc file : None

[hmaarrfk]

We've never built them. PRs welcome.

[dslarm]

Ah. Thanks Mark. I see they're "pkgs/main" - which I hadn't noticed (anaconda?). I can take a look but I'm probably behind you guys on what's going on.

[dslarm]

whilst I can build outside of conda using tensorflow's instructions, I see this currently after adding linux_aarch64 to my local fork:

Collecting h5py==3.11.0 (from -r /tmp/tmpch8cnpfy (line 1))
  Using cached h5py-3.11.0.tar.gz (406 kB)
  Installing build dependencies ... error
  error: subprocess-exited-with-error
  
  × pip subprocess to install build dependencies did not run successfully.
  │ exit code: 1
  ╰─> [3 lines of output]
      Ignoring oldest-supported-numpy: markers 'python_version == "3.8"' don't match your environment
      ERROR: Could not find a version that satisfies the requirement Cython<4,>=0.29.31 (from versions: none)
      ERROR: No matching distribution found for Cython<4,>=0.29.31
      [end of output]
  
  note: This error originates from a subprocess, and is likely not a problem with pip.
error: subprocess-exited-with-error

tried a couple of things so far:

• after first failing with the h5py, I tried to bump to 3.12.0 - but that involves a bit intrusive package and sha256sum edits to dependencies. That got to the next missing package, but seemed like a wrong direction,
• installing h5py==3.11.0 via pip on the command line outside of conda causes pip to need to build the wheel - and it needed
  hdf5 dev libraries. With 3.12.0 there was a precompiled 3.12.0 for aarch64 on pypi (ie. there was not a precompiled 3.11.0 - is that the cause of the error about Cython?)

(any help / ideas on the Cython error? cython is in the meta.yaml and it claims:

(/home/conda/feedstock_root/build_artifacts/tensorflow-split_1742491401195/_build_env) [conda@45391984a0d7 pypi_h5py]$ cython --version
Cython version 3.0.12 

[hmaarrfk]

The best way to try to get help is to make a PR here with the suggested changes so we can all be on the page about what is going on.

To build locally:

1. Manually add aarch migration.
2. Rerender
3. python build-locally.py select the appropriate build.

[hmaarrfk]

Sorry for not recognizing you from the perl work on arm.

A PR here showing the "startup" challenges on tensorflow is very helpful.

Once the "initial build" passes the first few phases, often it needs like 10hours to complete, which times out the CIs.

Those can then be run on your machine.

[dslarm]

Made a start with #426