py3.6 >= support

[isidentical]

I see that 3.6 support dropped in this commit though it is not clear about reasoning. I would assume it is because it's EOL coming in 9 months, though 3.6 might still be one of the most used versions. Would it make sense to bring the support back?

[martindurant]

The happened after the confluence of a couple of things

• a general move in the pydata ecosystem (like pandas, dask and many more)
• a specific async issue in fsspec that I could not resolve for py 3.6

The latter point might have been avoided in fsspec/filesystem_spec#572 , but I'd like to avoid possible differences in asyncio API for py36.

3.6 might still be one of the most used versions

This maybe true, but those on py36 probably don't install the latest releases of packages like s3fs.

[isidentical]

a general move in the pydata ecosystem (like pandas, dask and many more)

When would you think that filesystem-spec will drop the 3.6 support though?

[martindurant]

That is up for debate. If we are to have it, then the CI should be made to run on 3.6. I don't know if the async thing will work or not. Most implementations don't need it, but HTTP, abfs, s3fs and gcsfs do.

[isidentical]

That is up for debate. If we are to have it, then the CI should be made to run on 3.6. I don't know if the async thing will work or not. Most implementations don't need it, but HTTP, abfs, s3fs and gcsfs do.

I can try to send a pull request that adds workflows, though I guess we should move the discussion over there?

[martindurant]

Sounds good