This was broken by the Tensorflow version bump, and is non-trivial to
correct - it now depends on dm-tree, which requires some work to build
due to a C++ extension which appears to require Bazel to compile.
flat hashes can be substituted through hashed-mirrors, while recursive
hashes can’t. This is especially important for Bazel since the bazel
fetch dependencies can come from multiple different methods (git,
http, ftp, etc.). To do this, we create tar archives from the
output/external directory, which is then extracted to build. All of
the Bazel hashes are all updated.
This unbreaks the build by updating to a newer version, adding new
dependencies and backporting some patches.
0.9 is released already, but that version is no longer compatible with
tf1. 0.8.0 is compatible with tf1 and tf2.
Fixes the tensorflow-probability built by first building the wheel with
bazel. This actually creates the dist-info folder, allowing the package
to be picked up as a pip dependency.
