The other day at work I had a mildly annoying problem – merging only selected files between a test and production branch in github. My particular use case was I had a test branch that needed to only interact with a test database, and the master branch needed to talk to the prod database. So I had particular config files with essentially different SQLAlchemy connection strings, but nothing else. Note I did not want these files ignored, just not merged between branches. (If I edit them I will need to make sure to edit both master & test branches in the end.)

I often use the GitHub desktop GUI to commit changes (when working on my local laptop). You can use the GUI to make a pull request, but when accepting the pull request in the browser I think it is all or nothing. I also need an entire command line solution for when I am working on a remote headerless machine with no GUI as well anyway. So here are my notes on how I solved the issue.

So just for illustration I added a test branch to my Blog_Code repository, and then some junk files just to illustrate. Via the git bash shell, if you navigate to your repository and do git diff master test --name-only, it shows you the different files in the two branches:

So you can see that I have 5 different files in total. Two config files and three different text files. If we do git diff master test -- special_config1, we can see the more specific differences between those two config files:

So you can see that the test branch version (in red) and the master branch version (in green), just have a minor difference. But in the end I want to keep those two files different between the branches, and not merge this config file (along with the other config file).

So here is the particular logic I put together, piping a bunch of commands together:

git switch master
git diff master test --name-only |
grep -v 'special_config1' |
grep -v 'Python/special_config2' |
sed 's/.*/"&"/' |
xargs git checkout test

The first line git switch is pretty self explanatory – I switch to the master branch (I will typically be doing work on test). Second I grab all the files that are different using git diff branch1 branch2, and only print out the file names. Third/Fourth lines I use grep to get rid of my specific config files out of that resulting list of files. You could also do grep -v 'file1.txt|file2.txt' |, but in this case this was giving me fits (maybe due to the forward slash not being escaped the right way for grep?).

The fifth sed line I wrap the files in quotes (if you have a file that has a space it will cause problems otherwise).

Sixth line I then use xargs to pass git checkout from the test environment, and pass in all of my files (minus my two config files). This is advice taken from this blog, just a slicker way to grab all of the files that are different minus a few specific config files. So instead of typing git checkout test file1.txt file2.txt etc. and typing the files by hand, I just grab all the files that are different and check them all out together.

Then once that is done it is the usual to commit the updated files. And then here in the end I switch the active environment back to test.

git commit -m 'Example only merging select files'
git push
git switch test

Maybe one of these days I will entirely ditch the GUI behind. But for now will just have to get by with my limited command line fu compared to these real computer programmers I work with more regularly.