turbodbc
turbodbc copied to clipboard
blue-yonder
Distribution of more flavours of wheels
It would be really nice to have also wheels for Linux (manylinux1_x86_64) and (cp34m) for Turbodbc on PyPI. Many Cloudera and in general Linux distributions only provide Python 3.4. Thanks a bunch for your great work!
Hi Florian! You are absolutely right, those wheels would make installing turbodbc that much easier.
@MathMagique We would then build wheels that have unixodbc bundled? Then I could build a quick Dockerfile that will emit these wheels and also make macOS wheels via a separate script.
@FlorianWilhelm For the time being, I've updated the conda packages to the latest version: https://github.com/conda-forge/turbodbc-feedstock / https://anaconda.org/conda-forge/turbodbc
@xhochy Yes, I think that would be the most reliable thing to do. Many Linux variants use unixodbc 2.3.x in the meantime (Ubuntu 16+, Debian 8+), so I would suggest to bundle the latest officially released version (2.3.4) with the wheels.
Probably it makes sense to add some build information to turbodbc so users can query the unixodbc version used during the build.
+1 on this. @MathMagique , out of curiosity, did the conversation continue on elsewhere, was looking for more info on where this issue ended up/
A couple of us are looking to use turbodbc, albeit linux wheel files would make our lives easier.
Hi nrhy! Sadly, the conversation did not continue with much momentum. In the meantime, you can use the conda package manager to install the precompiled turbodbc conda packages.
That's unfortunate. In this case, I guess that until official manylinux wheels can be provided, you would need to create wheels for your specific operating system (pip wheel should help). Just make sure you build the wheel in an environment with NumPy/Arrow installed so these add-ons are built as well.
You can then distribute the wheel among your colleagues, e.g., via devpi or file transfer. If they share the same OS, you should be safe, although not very convenient.
@MathMagique Yeah, it's what I've been doing. It'll have to do for now.
Thank you!
+1 on this. Having official Windows + Linux wheels available on pypi.org for at least Python 3.5+ would make installing turbodbc much easier. I remember when pyarrow starting putting out wheels on pypi in addition to their conda-forge builds.
@mpeleshenko Can you help with implementing this? For the Linux wheels, a base would be to construct a docker image that contains all dependencies compiled for the manylinux1 image as in seen in pyarrow: https://github.com/apache/arrow/tree/master/python/manylinux1
@xhochy Sure. I’ll give it a shot. By the way, I saw you had previously opened https://github.com/blue-yonder/turbodbc/pull/108. Is that one no longer viable?
Azure Functions (Consumption Plan, Linux) also cannot install turbodbc through requirements.txt, probably missing the non-pythonic dependencies. I'm all new to this, but I came across conda-press and conda-smithy, maybe that could help with building and distributing "fat" wheels?
