With the Rails 3.1 Asset Pipeline, there’s the rake assets:precompile command which precompiles assets into public/assets along with a manifest.yml file which rails uses to put the paths to the assets in its templates. Precompiling generates a few files, which can clutter commit logs if they’re committed. For this reason I might add public/assets to my .gitignore and build them on the server, rather than committing them to git and uploading them. Or, if the server had trouble compiling them, I might precompile them locally, and scp them up rather than committing them to git, because they can make a lot of noise in my git logs.

On Heroku, though, git is the only way to deploy. So if I’m having trouble having heroku precompile them, I need to commit them and push them via git. This could clutter up my commit log.

There is a way around this, though: use a branch! If I create a branch just for heroku, and just commit them to the branch, I can keep my commit logs free of changes to public/assets.

Here is a sketch of this process:

  1. Add public/assets to my .gitignore and commit it
  2. Create a heroku branch with git checkout -b heroku
  3. remove public/assets from my .gitignore and commit it. This just affects my heroku branch.
  4. Run RAILS_ENV=production rake assets:precompile there.
  5. Run git add . and git commit -m 'precompile assets' to commit the files to my heroku branch
  6. Run git push heroku heroku:master to push my heroku branch to heroku as master (heroku only uses the master branch)
  7. Run git checkout master and continue development on my master branch.

A deploy rake script could be used to automate this.

The heroku branch wouldn’t need to be pushed to github, because it doesn’t contain any information that can’t be regenerated. The only reason the history would need to be preserved is so it can be cleanly pushed to the heroku app. Fortunately, if the local repo was lost, it’s possible to pull from the heroku repository. Heroku shouldn’t be used as a place to store git repositories because the lifecycle for a heroku app is often different than that of a repository. For this purpose, though, the usefulness of the old assets commits is tied to the life of the Heroku app so it works out well.

The other day I went back to an old branch that had about five files that were changed and I only wanted to incorporate two of them into my master branch. In order to just merge in a couple, I used git merge with the –no-commit option.

git merge --no-commit planner

This merges in all changes but doesn’t commit it. Since it isn’t committed, it’s easy to roll back individual files. I ran git status to get a list of files and rolled back several files with git checkout. When I was done, I had only the two files I wanted to merge showing up when I ran git status.