Good ways to manage a changelog using git?

Question

I've been using Git for a while now, and I recently started using it to tag my releases so that I could more easily keep track of changes and be able to see which version each of our clients are running (unfortunately the code currently mandates that each client have their own copy of the PHP site; I'm changing this, but it's slow-going).

In any case, we're starting to build some momentum, I thought it would be really good to be able to show people what has changed since the last release. Problem is, I haven't been maintaining a changelog because I don't have a good idea of how to go about it. For this particular time, I can run through the log and manually create one, but that will get tiring very quickly.

I tried googling "git changelog" and "git manage changelog" but I didn't find anything that really talked about the workflow of code changes and how that coincides with the changelog. We're currently following Rein Henrichs' development workflow and I would love something that went along with that.

Is there a standard approach that I am missing, or is this an area where everybody does their own thing?

Thanks very much for your comments/answers!

Answer 1

For a GNU style changelog, I've cooked the function

gnuc() {
  {
    printf "$(date "+%Y-%m-%d")  John Doe  <[e-mail]>\n\n"
    git diff-tree --no-commit-id --name-only -r HEAD | sed 's/^/\t* /'
  } | tee /dev/tty | xsel -b
}

With this:

• I commit my changes periodically to backup and rebase them before doing the final edit to the ChangeLog
• then run: gnuc

and now my clipboard contains something like:

2015-07-24  John Doe  <[e-mail]>

        * gdb/python/py-linetable.c (): .
        * gdb/python/py-symtab.c (): .

Then I use the clipboard as a starting point to update the ChangeLog.

It is not perfect (e.g. files should be relative to their ChangeLog path, so python/py-symtab.c without gdb/ since I will edit the gdb/ChangeLog), but is a good starting point.

More advanced scripts:

• https://github.com/tromey/git-gnu-changelog by Tromey, a GDB maintainer
• GCC's contrib/mklog
• https://github.com/davidmalcolm/gcc-refactoring-scripts/blob/1fc7fe038bffbb2b48d0a7a300625639f01aefa1/generate-changelog.py

I have to agree with Tromey though: duplicating git commit data in the ChangeLog is useless.

If you are going to make a changelog, make it a good summary of what is going on, possibly as specified at http://keepachangelog.com/

Answer 2

You can use some flavor of git log to help you out:

git log --pretty=%s                 # only print the subject

If you name your branches nicely, so that a merge to master shows up as something like "Merged branch feature-foobar", you can shorten things by only showing that message, and not all the little commits that you merged, which together form the feature:

git log --pretty=%s --first-parent  # only follow first parent of merges

You might be able to augment this with a script of your own, which could do things like strip out the "Merged branch" bits, normalize formatting, etc. At some point you have to write it yourself though, of course.

Then you could create a new section for the changelog once per version:

git log [opts] vX.X.X..vX.X.Y | helper-script > changelogs/X.X.Y

and commit that in your version release commit.

If your problem is that those commit subjects aren't anything like what you'd want to put in a changelog, you pretty much have two options: keep doing everything manually (and try to keep up with it more regularly instead of playing catch-up at release time), or fix up your commit message style. One option, if the subjects aren't going to do it for you, would be to place lines like "change: added feature foobar" in the bodies of your commit messages, so that later you could do something like git log --pretty=%B | grep ^change: to grab only those super-important bits of the messages.

I'm not entirely sure how much more than that git could really help you create your changelogs. Maybe I've misinterpreted what you mean by "manage"?

Answer 3

Since creating a tag per version is the best practice, you may want to partition your changelog per version. In that case, this command could help you:

git log YOUR_LAST_VERSION_TAG..HEAD --no-merges --format=%B

Answer 4

This was about 3-4 years ago, but for the sake of future searchers, it's now possible to generate gorgeous logs with:

git log --oneline --decorate

Or, if you want it even prettier (with color for terminal):

git log --oneline --decorate --color

Piping that output to ChangeLog is what I currently use in all my projects, it's simply amazing.

Answer 5

The gitlog-to-changelog script comes in handy to generate a GNU-style ChangeLog.

As shown by gitlog-to-changelog --help, you may select the commits used to generate a ChangeLog file using either the option --since:

gitlog-to-changelog --since=2008-01-01 > ChangeLog

or by passing additional arguments after --, which will be passed to git-log (called internally by gitlog-to-changelog):

gitlog-to-changelog -- -n 5 foo > last-5-commits-to-branch-foo

For instance, I am using the following rule in the top-level Makefile.am of one of my projects:

.PHONY: update-ChangeLog
update-ChangeLog:
    if test -d $(srcdir)/.git; then                         \
       $(srcdir)/build-aux/gitlog-to-changelog              \
          --format='%s%n%n%b%n' --no-cluster                \
          --strip-tab --strip-cherry-pick                   \
          -- $$(cat $(srcdir)/.last-cl-gen)..               \
        >ChangeLog.tmp                                      \
      && git rev-list -n 1 HEAD >.last-cl-gen.tmp           \
      && (echo; cat $(srcdir)/ChangeLog) >>ChangeLog.tmp    \
      && mv -f ChangeLog.tmp $(srcdir)/ChangeLog            \
      && mv -f .last-cl-gen.tmp $(srcdir)/.last-cl-gen      \
      && rm -f ChangeLog.tmp;                               \
    fi

EXTRA_DIST += .last-cl-gen

This rule is used at release time to update ChangeLog with the latest not-yet-recorded commit messages. The file .last-cl-gen contains the SHA1 identifier of the latest commit recorded in ChangeLog and is stored in the Git repository. ChangeLog is also recorded in the repository, so that it can be edited (e.g. to correct typos) without altering the commit messages.

Answer 6

DISCLAIMER: I'm the author of gitchangelog of which I'll be speaking in the following.

TL;DR: You might want to check gitchangelog's own changelog or the ascii output that generated the previous.

If you want to generate a changelog from your git history, you'll probably have to consider:

• the output format. (Pure custom ASCII, Debian changelog type, Markdow, ReST...)
• some commit filtering (you probably don't want to see all the typo's or cosmetic changes getting in your changelog)
• some commit text wrangling before being included in the changelog. (Ensuring normalization of messages as having a first letter uppercase or a final dot, but it could be removing some special markup in the summary also)
• is your git history compatible ?. Merging, tagging, is not always so easily supported by most of the tools. It depends on how you manage your history.

Optionaly you might want some categorization (new things, changes, bugfixes)...

With all this in mind, I created and use gitchangelog. It's meant to leverage a git commit message convention to achieve all of the previous goals.

Having a commit message convention is mandatory to create a nice changelog (with or without using gitchangelog).

commit message convention

The following are suggestions to what might be useful to think about adding in your commit messages.

You might want to separate roughly your commits into big sections:

• by intent (for example: new, fix, change ...)
• by object (for example: doc, packaging, code ...)
• by audience (for example: dev, tester, users ...)

Additionally, you could want to tag some commits:

• as "minor" commits that shouldn't get outputed to your changelog (cosmetic changes, small typo in comments...)
• as "refactor" if you don't really have any significative feature changes. Thus this should not also be part of the changelog displayed to final user for instance, but might be of some interest if you have a developer changelog.
• you could tag also with "api" to mark API changes or new API stuff...
• ...etc...

Try to write your commit message by targeting users (functionality) as often as you can.

example

This is standard git log --oneline to show how these information could be stored::

* 5a39f73 fix: encoding issues with non-ascii chars.
* a60d77a new: pkg: added ``.travis.yml`` for automated tests. 
* 57129ba new: much greater performance on big repository by issuing only one shell command for all the commits. (fixes #7)
* 6b4b267 chg: dev: refactored out the formatting characters from GIT.
* 197b069 new: dev: reverse ``natural`` order to get reverse chronological order by default. !refactor 
* 6b891bc new: add utf-8 encoding declaration !minor 

So if you've noticed, the format I chose is:

{new|chg|fix}: [{dev|pkg}:] COMMIT_MESSAGE [!{minor|refactor} ... ]

To see an actual output result, you could look at the end of the PyPI page of gitchangelog

To see a full documentation of my commit message convention you can see the reference file gitchangelog.rc.reference

How to generate exquisite changelog from this

Then, it's quite easy to make a complete changelog. You could make your own script quite quickly, or use gitchangelog.

gitchangelog will generate a full changelog (with sectioning support as New, Fix...), and is reasonably configurable to your own committing conventions. It supports any type of output thanks to templating through Mustache, Mako templating, and has a default legacy engine written in raw python ; all current 3 engines have examples of how to use them and can output changelog's as the one displayed on the PyPI page of gitchangelog.

I'm sure you know that there are plenty of other git log to changelog tools out there also.