`ModuleNotFoundError`: 1.4.0 (and below, possibly) missing an explicit `future` dependency

As of right now, new installations of tensorflow-transform==1.4.0 fail to import with a ModuleNotFoundError because the future library is not marked as a dependency in the setup.py.

This likely wasn’t a problem before because some recursive dependency of tft depended on future, masking the issue. But future is a hard dependency of tft 1.4.0, as it is imported on import tensorflow_transform, so it should probably be added as a dep in a 1.4.1 release (assuming this team does back-releases of that kind).

For people looking for a hotfix in your own project, adding future as a dependency of your project should be enough.

This isn’t a problem in 1.5.0 because the only usage of future was removed in this commit.

[jiyongjung0]

FYI, tensorflow-transform 1.4.1 was released with future as an additional dependency.

[davidxia]

My team’s using tfx==1.3.4 and also hit this today. This tfx version pulls in tensorflow-transform==1.3.0 which explicitly depends on future here but doesn’t declare it as a dependency.

We worked around by adding future==0.18.2 to our own requirements.txt.

The error messages will say Exception: The full "tfx" package must be installed to use this functionality. But if you look more closely, you’ll see it’s a future import failure.