Skip to content

Latest commit

 

History

History
134 lines (102 loc) · 4.55 KB

README.md

File metadata and controls

134 lines (102 loc) · 4.55 KB
Raw

TF Protobuf definitions

This directory contains the forked subtrees of the tensorflow and tensorflow_serving projects which provide the project with TensorFlow protobuf definitions.

They should only be updated to pull changes from the original upstream projects using git subtree commands.

How to update the Protobuf definitions (to pull changes from upstream)

Set up the upstream (do this the first time you want to update)

  1. Clone this repo
  2. Set the upstream remote to the parent tensorflow repo and/or tensorflow_serving repo
git remote add upstream-tf [email protected]:tensorflow/tensorflow.git
git remote add upstream-serving [email protected]:tensorflow/serving.git

This should set up a new master branch for each upstream (upstream-tf/master and upstream-serving/master) and copy all of the upstream refs into your local.

Updating with upstream changes

This is a very similar process for tensorflow and tensorflow_serving - the main difference is that tensorflow is much bigger than tensorflow_serving, so it will take longer (possibly up to an hour).

tensorflow
  1. Pull the updated upstream refs:
git fetch upstream-tf

Now upstream-tf is up to date.

Also fetch the upstream tags (since we use two upstreams, it's a good idea to use a refspec to keep tags organised.

git fetch upstream-tf '+refs/tags/*:refs/rtags/upstream-tf/*'
  1. Check out the ref you want to update to:
git checkout refs/rtags/upstream-tf/<version>
  1. Create a upstream-tf-subtree branch containing the subtree from upstream-tf. This command will take a while - tensorflow/tensorflow is big.
git subtree split --prefix tensorflow --onto upstream-tf-subtree
  1. Create a new branch in the min-tfs-client tree with clean protobuf_srcs/tensorflow.

There doesn't appear to be a way to use git subtree commands to cleanly pull the subtree changes into a branch, especially since tensorflow repos do their own fancy git rework which complicates things. Therefore the recommended approach is to use base git to ensure a clean history.

git checkout master
git checkout -b <new-branch>
rm -rf protobuf_srcs/tensorflow
git add --all && git commit -m "[WIP]"
  1. Merge the subtree into protobuf_srcs:
git subtree add --squash --prefix protobuf_srcs/tensorflow upstream-tf-subtree
  1. Clean the commit history:
git reset --soft master
git add protobuf_srcs/tensorflow && git commit

It helps to include a descriptive PR commit message which mentions the upstream git ref used, e.g.

Squashed commit of tensorflow changes

tensorflow/tensorflow commit [<version>]: <commit SHA>
  1. Push and raise a PR for <new-branch> against master
git push -u origin HEAD
tensorflow_serving
  1. Pull the updated upstream refs:
git fetch upstream-serving

Now upstream-serving is up to date.

Also fetch the upstream tags (since we use two upstreams, it's a good idea to use a refspec to keep tags organised.

git fetch upstream-serving '+refs/tags/*:refs/rtags/upstream-serving/*'
  1. Check out the ref you want to update to:
git checkout refs/rtags/upstream-serving/<version>
  1. Create a upstream-serving-subtree branch containing the subtree from upstream-serving.
git subtree split --prefix tensorflow_serving --onto upstream-serving-subtree
  1. Create a new branch in the min-tfs-client tree with clean protobuf_srcs/tensorflow_serving.

There doesn't appear to be a way to use git subtree commands to cleanly pull the subtree changes into a branch, especially since tensorflow repos do their own fancy git rework which complicates things. Therefore the recommended approach is to use base git to ensure a clean history.

git checkout master
git checkout -b <new-branch>
rm -rf protobuf_srcs/tensorflow_serving
git add --all && git commit -m "[WIP]"
  1. Merge the subtree into protobuf_srcs:
git subtree add --squash --prefix protobuf_srcs/tensorflow_serving upstream-serving-subtree
  1. Clean the commit history:
git reset --soft master
git add protobuf_srcs/tensorflow_serving && git commit

It helps to include a descriptive PR commit message which mentions the upstream git ref used, e.g.

Squashed commit of tensorflow serving changes

tensorflow/tensorflow_serving commit [<version>]: <commit SHA>
  1. Push and raise a PR for <new-branch> against master
git push -u origin HEAD