tabulate is used by functions that extract tables from Microsoft documents, but there is nothing explicitly requiring the library. This was not caught by tests, because for some reason, tabulate is in base.txt.
This PR adds the dependency to base.in (which also puts it in setup.py), and recompiles the dependencies.
Updated to the the latest version of unstructured-inference. detectron2 now gets implemented with onnxruntime, yay!
---------
Co-authored-by: Matt Robinson <mrobinson@unstructured.io>
Addresses #631.
* Uses constraints to keep dependency versions more consistent.
* Moves all dependencies to .in files which are then ingested by setup.py.
* Adds script to check consistency of all extras.
* Adds consistency check to CI.
I should note that while it shouldn't be possible to cause a conflict between base.txt and any of the extras (because base.txt constrains all the extras) it is possible to get a conflict between two of the extras files. There are ways of trying to avoid that (like constraining each file by all the files that have already been processed before it in the order given in the make pip-compile target) but the ones I could think of seemed a little overwrought, and come with problems of their own. If a conflict arises, it should be flagged by CI or locally with make check-deps. When/if that happens, you can resolve the conflict by adding appropriate global constraints in requirements/constraints.txt.
Also note that if fileA.in is constrained by fileB.txt, then fileB.in should be compiled before fileA.in in the make pip-compile target. Otherwise fileA.in will be compiled with the old version of fileB.txt which can cause conflicts or keep dependencies from being updated properly.
Update versions of dependencies, including unpinning the unstructured-inference dependency that's causing conflicts in repos like pipeline-oer that want the newer version.
So as you may see this is a pretty big PR, that basically adds an "adapter" to easily plug in any connector with an available fsspec implementation. This is a way to standardize how the remote filesystems are used within unstructured.
I've additionally renamed s3_connector.py to s3.py for readability and consistency and tested that the current approach works as expected and is aligned with the expectations.
