Search git history for code within a file

My class was missing a method! So I had to search git history for one particular file for actuall code changes! This is easy to do:

git grep "<regex>" $(git rev-list --all -- <file>)

where <regex> is what you are searching for, and where <file> is your file


Stop git committing chmod changes

Pretty self explanatory. Just do this:

git config core.fileMode false

The documentation says this about it:

If false, the executable bit differences between the index and the 
working copy are ignored; useful on broken filesystems like FAT. 
See git-update-index(1). True by default.

Here’s a warning from a guy on StackOverflow:

core.fileMode is not the best practice and should be used carefully. This setting only covers the executable bit of mode and never the read/write bits. In many cases you think you need this setting because you did something like chmod -R 777, making all your files executable. But in most projects most files don’t need and should not be executable for security reasons.

The proper way to solve this kind of situation is to handle folder and file permission separately, with something like:

find . -type d -exec chmod a+rwx {} \; # Make folders traversable and read/write
find . -type f -exec chmod a+rw {} \; # Make files read/write

If you do that, you’ll never need to use core.fileMode, except in very rare environment.

Line Endings in Git with Windows

Devving on Windows is a PITA.

Anyway, ever seen a message like this?

warning: LF will be replaced by CRLF in tests/unit/Del/Console/CommandTest.php.
The file will have its original line endings in your working directory.
warning: LF will be replaced by CRLF in tests/unit/Del/Console/CommandTest.php.
The file will have its original line endings in your working directory.

We only want LF. To squelch this crap, run the following:

git config core.autocrlf false


Use Git bisect to find dodgy commits

Today I had the mammoth task of checking through 512 git commits to find a piece of code that broke something.

Usually i would git log, look back however many commits, and then do a git reset –hard COMMIT_NUMBER, then check if it worked. If it did, I would git pull back to the HEAD again, and try a resetting back to a more recent commit, until I found the bad code.

Never again! Git bisect to the rescue!

Find any good commit in the past, and note the commit number. Find any bad commit where the code is broken, and note the commit number.

Now, do the following:

git bisect start
git bisect good 514d83c
git bisect bad b27f38e


Git checks out the middle commit between the good and bad ones. At this point I reloaded my page to see if the code was working or not. The code was working, so I then told git that it was good:

git bisect good


Again the code was working, so as you can see I ran it again. Each time, git bisect jumps half way, iterating and narrowing down the options. Keep doing this and checking your code until it breaks, then say:

git bisect bad

Here’s the rest of the output:


Now we have the exact commit number, and can do a git diff to find out what you did wrong! ๐Ÿ™‚

Once you have the commit in question, tell git bisect that you are finished:

git bisect reset

I am amazed that I’m only just finding out about this awesome feature of Git! I’m sure you’ll love it too, try it!

Have fun!

Updating a forked GitHub Repo

At the PHPNW 2014 conference, they had a ‘Hackathon’, where everyone got together and started hacking away at projects. I had a shot at the JoindIn website.

Anyway, I forked the repositories so I could work on my own branch. Now though, my fork is outdated, and needs updating. So in order to remedy this, we add another remote.

From your local machine’s cloned repo, you simply say:

git remote add upstream
git checkout master
git pull upstream master
git push origin master

Thanks to Derick Rethans of XDebug fame for this info ๐Ÿ˜€

Fixing Git Submodule Issues

If you’ve set up a Git ‘Push to Deploy’ setup for your site based on my earlier post, you may or may not have run into some anomalous behaviour! Namely, when you run git submodule init, you get a message similar to this:

No submodule mapping found in .gitmodules for path 'js/jquery-browser-fingerprint'

Or even :

fatal: reference is not a tree: 48d1407779fb1e8dc6d3c6c6fb87df4c1cbc65ed
Unable to checkout '48d1407779fb1e8dc6d3c6c6fb87df4c1cbc65ed' in submodule path 'cms/js/flipswitch'

Thankfully, these issues are easily fixed. modules should be listed in a file in your root .gitmodules, and look like this:

[submodule "jquery-browser-fingerprint"]
 path = /js/jquery-browser-fingerprint
 url = git://

Listing your submodules in here will fix your first problem. The second issue is caused by referencing a commit that hasn’t been pushed! Did you copy project files from somewhere else? Try pushing the commit from your other project.

A clearer way to sort these issues however, is to clear whatever cache git has for those submodules. Typing this clears it.

git rm --cached js/jquery-browser-fingerprint

This should allow you to run git submodule init, and git submodule update. One last point. You don’t remember ever having to create this .gitmodules file before, so why now? Basically it allows you to set custom folders etc for the submodule. If you’re happy with the defaults, you should theoretically be able to skip the submodules file, and ย can combine both git commands into this:

git submodule update --init

This updates everything and initialises in one! So to sum up, try removing the cached stuff in git, then run the combined update init file! Happy pushing, pulling, and deploying!