capture version_id after upload

[shcheklein]

Similar to fsspec/gcsfs#718

To be able to refer to a file properly (in case multiple uploads happening simultaneously to the same path), we want to grab generation from the response. S3 already does this.

[shcheklein]

@martindurant do you know by chance who is maintaining the project and could release this change?

[kyleknap]

@shcheklein thanks for contribution! Populating the updated version_id after a completed upload seems reasonable to me especially since s3fs is already doing this and you have a similar change out for gcsfs.

I'll have more cycles next week to take a deeper review but some initial things we probably want to address now are:

1. Let's populate the version_id only if the upstream file system is version_aware. Looks like s3fs has similar logic
2. It would be great if we can get test cases added to make sure we the version_id is getting set for the different upload branches where it is being set now for version aware filesystems. It would also be good to have a test case too where we don't set the version_id if the filesystem is not version aware.
3. Add a small changelog entry here for the enhancement.

[shcheklein]

Hi @kyleknap , thanks for the feedback. I've addressed all the items.

Azurite doesn't support versioning - thus we use mocks in tests. I think that should be fine.

[kyleknap]

Thanks @shcheklein! I agree mocks suffice in this case. I'll take a deeper look at the PR this week.

[kyleknap]

This looks good! I pulled down the PR and tried it out and it was working well. I just had one question related to append blobs that I ran into.

[kyleknap]

For append blobs, were you able to get version id's returned for the upload blob method call? Trying it out myself, I'm not sure if an x-ms-version-id header is even returned back when a PUT is called on an append blob. So the version_id was consistently None even though for block blobs it was being returned back. That being said for HEAD calls on the append blob, the x-ms-version-id was present.

I'm thinking if the API is not returning version id's for PUTs on append blobs, let's remove this logic for now to avoid code paths that won't be hit and avoid having tests implying this should work. We can always add it in if version id's are returned in the future.