bisect-demo

Demoing git bisect

MIT License

Stars
5
Committers
1

A demo of git bisect

square.py squares a number you give it. But it's broken with negative numbers! I know I checked this with the first version I wrote. When did I break it?

$ python3 square.py 5
25
$ python3 square.py -6
0

git bisect can help. It goes through the history of git commits to find the first one where something doesn't work.

This example demonstrates how to use git bisect, but it doesn't really show why. It's easy here to look at the script and see what's going wrong. Bisect is more useful when you have a big complicated codebase, and it's not obvious where a bug is coming from.

The git bisect documentation has much more detail about what it can do.

Bisecting manually

To start git bisecting, we have to tell it a point when the code is bad— this is the current commit—and one where it's good, i.e. before it got broken. In this repository, I've made a tag my-first-code when I know this was working, but it can be trickier to figure out in real life.

$ git bisect start
$ git bisect bad
$ git bisect good my-first-code 
Bisecting: 7 revisions left to test after this (roughly 3 steps)
[84e9e6343a2cb00b12439c4a9022f9666b9ac0df] Use argparse for command line

Now it's checked out a new revision, half way between our good and bad ones. We need to try it and tell git if it's good (working) or bad (broken).

$ python3 square.py -6
0
$ git bisect bad
Bisecting: 3 revisions left to test after this (roughly 2 steps)
[ebe7c91e9d7c558a61ebed24b45adf181f5020f7] Multiplication is just repeated addition

Each time we say git bisect bad or git bisect good, git will give us a new commit to try. After a few times, it has narrowed it down to one commit—this one.

ebe7c91e9d7c558a61ebed24b45adf181f5020f7 is the first bad commit
commit ebe7c91e9d7c558a61ebed24b45adf181f5020f7
Author: Thomas Kluyver
Date:   Wed Nov 29 18:34:06 2017 +0000

Multiplication is just repeated addition

:100644 100644 f68ade3b07c88b099742b55b9cf4de591687b941 03989097f0eb54a10646b6b02c89079b2fc56b20 M	square.py

Finally, finish bisecting and go back to where we were:

$ git bisect reset
Previous HEAD position was 59f2008... Check that we get an integer
Switched to branch 'master'
Your branch is up-to-date with 'origin/master'.

Bisecting automatically

If we bisect manually, we still have to figure out if it's working each time. That can be tedious, so can we make the computer do it for us?

There's a test in this folder, test_square.py. You can run it with py.test, and see that it fails. We can tell git bisect to run this automatically to check for working code.

First, make a copy of the test file. The test was only added after the bug was introduced, so when git goes back in history, it will delete the file that's tracked in git. Making a copy gets around this.

$ cp test_square.py test_square2.py

Start bisecting like before:

$ git bisect start
$ git bisect bad
$ git bisect good my-first-code 
Bisecting: 7 revisions left to test after this (roughly 3 steps)
[84e9e6343a2cb00b12439c4a9022f9666b9ac0df] Use argparse for command line

Now we tell git to run py.test:

$ git bisect run py.test

Your terminal will rapidly fill up with information as git jumps around and re-runs your tests. When it stops, you should see the same result as manual bisecting.

ebe7c91e9d7c558a61ebed24b45adf181f5020f7 is the first bad commit
commit ebe7c91e9d7c558a61ebed24b45adf181f5020f7
Author: Thomas Kluyver
Date:   Wed Nov 29 18:34:06 2017 +0000

Multiplication is just repeated addition

:100644 100644 f68ade3b07c88b099742b55b9cf4de591687b941 03989097f0eb54a10646b6b02c89079b2fc56b20 M	square.py
bisect run success

Don't forget to finish bisecting and go back to where you were:

$ git bisect reset
Previous HEAD position was 59f2008... Check that we get an integer
Switched to branch 'master'
Your branch is up-to-date with 'origin/master'.

Git can run any script you give it. The only requirement is that it exits successfully (exit code 0) if your code is good, and errors (exit code not 0) if your code is bad.