Calculates the Levenshtein similarity ratio between commits. A measurement which can be used to identify complex/large commits, which should be checked to ensure commits are concise and only address one issue at a time.
doing git diff --numstat is one such measure, but it can easely be thrown off by something simple as reindenting code.
For an explanation of the sentament: https://wiki.openstack.org/wiki/GitCommitMessages
Question is when one is looking at history, how quickly can these commits be identified.
[/tmp]$: mkdir lsrTest
[/tmp]$: cd $!
[/tmp/lsrTest]$: git init
Initialized empty Git repository in /tmp/lsrTest/.git/
[/tmp/lsrTest(master)]$: echo "readme" >readme
[/tmp/lsrTest(master)]$: git add readme
[/tmp/lsrTest(master)]$: git commit -m "Added initial readme."
[master (root-commit) 817c4bf] Added initial readme.
1 file changed, 1 insertion(+)
create mode 100644 readme
[/tmp/lsrTest(master)]$: # similarity of a commit against itself is clearly 100%
[/tmp/lsrTest(master)]$: git lsr HEAD HEAD
817c4bf 817c4bf = 100.00%
[/tmp/lsrTest(master)]$: echo "a second line" >>readme
[/tmp/lsrTest(master)]$: git commit -am "Updated readme."
[master e19d40a] Updated readme.
1 file changed, 1 insertion(+)
[/tmp/lsrTest(master)]$: git lsr HEAD^ HEAD
817c4bf e19d40a = 72.14%
[/tmp/lsrTest(master)]$: # the ratio is symmetric:
[/tmp/lsrTest(master)]$: git lsr HEAD HEAD^
e19d40a 817c4bf = 72.14%
[/tmp/lsrTest(master)]$: echo "a third line." >>readme
[/tmp/lsrTest(master)]$: git commit -am "Yet another update to readme."
[master 31a6435] Yet another update to readme.
1 file changed, 1 insertion(+)
[/tmp/lsrTest(master)]$: git lsr HEAD HEAD^
31a6435 e19d40a = 80.68%
[/tmp/lsrTest(master)]$: git lsr HEAD HEAD^^
31a6435 817c4bf = 70.39%
[/tmp/lsrTest(master)]$: echo "tempfile" >tempfile
[/tmp/lsrTest(master)]$: git add tempfile
[/tmp/lsrTest(master)]$: git commit -am "Added a tempfile."
[master b9adbb2] Added a tempfile.
1 file changed, 1 insertion(+)
create mode 100644 tempfile
[/tmp/lsrTest(master)]$: git lsr HEAD^ HEAD # we didn't touch the same files, so expect no simularity.
31a6435 b9adbb2 = 0%
[/tmp/lsrTest(master)]$: echo "a second line to the temp file." >>tempfile
[/tmp/lsrTest(master)]$: echo "one last line to the readme." >>readme
[/tmp/lsrTest(master)]$: git commit -am "Updated both tempfile and readme."
[master 927e488] Updated both tempfile and readme.
2 files changed, 2 insertions(+)
[/tmp/lsrTest(master)]$: git lsr HEAD^ HEAD
b9adbb2 927e488 = 34.92%
[/tmp/lsrTest(master)]$: git rm tempfile
rm 'tempfile'
[/tmp/lsrTest(master)]$: git commit -am "Removed tempfile."
[master ab00b79] Removed tempfile.
1 file changed, 2 deletions(-)
delete mode 100644 tempfile
[/tmp/lsrTest(master)]$: git lsr HEAD^ HEAD
927e488 ab00b79 = 75.5%
[/tmp/lsrTest(master)]$: git lsr HEAD^^ HEAD
b9adbb2 ab00b79 = 86.22%
Looking at history, which commits are likely to be less intuative and we need to ensure that they are clean and have good/clear commit messages, or alternatively broken up to smaller commits:
[/tmp/lsrTest(master)]$: git rev-list HEAD | xargs git lsr
ab00b79 927e488 = 75.5%
927e488 b9adbb2 = 34.92%
b9adbb2 31a6435 = 0%
31a6435 e19d40a = 80.68%
e19d40a 817c4bf = 72.14%
We should have a look at what was introduced by b9adbb2 and 927e488. The point is not to eliminate the low numbers, but to make sure we document/use good commit messages when the repository is changing significantly.
The toy repo above does not do this justice, so try it out on a larger repository.
- calc:
sudo apt-get install apcalc - python-levenshtein:
sudo apt-get install python-levenshtein
- git clone this repo.
- create symlinks in ```~/bin/`` for lsr and git-lsr.