Consider removing the use of bind()

[damienmg]

The current version force bind statement everywhere. There is no need for them as the exports can point directly to the repository.

[johnynek]

@damienmg there is a need, since users can easily rebind to change the versions we point to. We often use this pattern to reconfigure both temporarily and remote repos that depend on other workspaces.

I am a big fan of bind and it gives us a lot of configuration options. I am very concerned that the bazel developers are generally not familiar with this pattern and are as such neglecting the usefulness.

I have commented at length on this issue: https://github.com/bazelbuild/bazel/issues/1952 but I am not sure I have actually convinced anyone.

While it may be possible for us to discover a new workflow, I don't see a reason to do that just because it seems some folks don't like bind. I'd like to see the case against bind and why it is stronger than the value it provides in configuration (due to bind's ability to be rebound or overridden in WORKSPACEs).

[damienmg]

I am curious to learn about the actual needs that are covered neither by alias nor by --override_repository (we should continue the discussion on bazelbuild/bazel#1952). bind is a very odd mechanisms in Bazel that fills our codebase with tricks.

Anyway regarding this issue, can we just make it an option to bazel-deps?

[johnynek]

Can certainly make it an option. And I am happy to study --override_repository and alias in more detail. Are the docs up to date on that?