I want to associate custom metadata with a git commit
. Specifically to record a review ID from a code review but it could be anything. Tags seem a natural way to do that but I expect to have a review for every commit and I don't want to clutter gitk
with tons of tags. Is there some other mechanism to add custom metadata? Can I make certain tags invisible? If I could tell gitk
not to display tags matching some pattern or RE, that would likely work but I don't see a way to do that.
With git notes
you can add a “note” to a commit. You can also add them
to other Git objects, but let’s just focus on commits since that is what
the question is about.
A note is a Git object, and can in principle be “whatever” (arbitrary data). But we’ll focus on something simple and textual for our purposes.
The question mentions review ids, so let’s make up some way to represent such a thing. I don’t know what review ids really look like, but hopefully the following would be sensible:
Review-id: 42
So this is effectively a key-value pair. Let’s add the above string to the current commit:
git notes add -m "Review-id: 42"
If you run git log
the note will be shown inline:(†1)
Author: Victor Version Control <[email protected]>
Date: Tue Nov 8 21:10:25 2016 +0100
Implement feature x
Notes:
Review-id: 42
Of course you can add more “subnotes” to this note (we will stick with
the simple key: value
syntax, one value per line). For example, if you
found out three months later that the commit message got something
wrong, just append the correction to the note:
git notes append -m "Errata: It was actually feature y."
git log
:
Author: Victor Version Control <[email protected]>
Date: Tue Nov 8 21:10:25 2016 +0100
Implement feature x
Notes:
Review-id: 42
Errata: It was actually feature y.
We use git notes append
in order to easily add this extra data to the
note. You could also use git notes edit
in order to edit the file
directly.
Of course, since a Git note is just a single mutable file, you can run into merge conflicts. To make that less likely, you can:
man git-notes
, section “Notes
merge strategies”.The OP asked:
> Can I make certain tags invisible?
By default, git log
only shows one note, namely
.git/refs/notes/commits
. commits
is just one note in the namespace.
Maybe you want issues to be in their own namespace:
git notes --ref=issues add -m "Fixes: #32"
Since this is stored in .git/refs/notes/issues
and not in
.git/refs/notes/commits
, “Fixes: #32” won’t show up when you run
git log
. So you have effectively made such notes invisible by default.
If you want it to be shown, pass --notes=issues
to git log
:
$ git log --notes=issues
Author: Victor Version Control <[email protected]>
Date: Tue Nov 8 21:10:25 2016 +0100
Implement feature x
Notes (issues):
Fixes: #32
But now .git/refs/notes/commits
are hidden. That one can easily be
included as well:
$ git log --notes=issues --notes=commits
Author: Victor Version Control <[email protected]>
Date: Tue Nov 8 21:10:25 2016 +0100
Implement feature x
Notes (issues):
Fixes: #32
Notes:
Review-id: 42
Errata: It was actually feature y.
There are variables to configure which notes are shown by default; see
man git-config
.
Metadata can of course be recorded in the commit message directly.(†2) But
commit messages are immutable, so to change them really means to make a
whole new commit, with all the rippling consequences that that entails.
Git-notes on the other hand are mutable, so you are always able to
revise them. And each modification of a note is of course version
controlled. In our case, for .git/refs/notes/commits
:
$ git log refs/notes/commits
Author: Victor Version Control <[email protected]>
commit 9f0697c6bbbc6a97ecce9834d4c9afa0d668bcad
Date: Tue Nov 8 21:13:52 2016 +0100
Notes added by 'git notes append'
commit b60997e49444732ed2defc8a6ca88e9e21001a1d
Author: Victor Version Control <[email protected]>
Date: Tue Nov 8 21:10:38 2016 +0100
Notes added by 'git notes add'
Your notes aren’t shared by default; you have to do so explicitly. And compared to other refs, sharing notes isn’t very user-friendly. We have to use the refspec syntax:
git push refs/notes/*
The above will push all of your notes to your remote.
It seems that fetching notes is a bit more involved; you can do it if you specify both sides of the refspec:
git fetch origin refs/notes/*:refs/notes/*
So that’s definitely not convenient. If you intend to use Git-notes regularly, you’ll probably want to set up your gitconfig to always fetch notes:
[remote "origin"]
…
fetch = +refs/notes/*:refs/notes/*
Git has the inconvenient default that notes are not carried over when a commit is rewritten. So if you for example rebase a series of commits, the notes will not carry over to the new commits.
The variable notes.rewrite.<command>
is by default set to true
, so one might
assume that notes are carried over. But the problem is that the variable
notes.rewriteRef
, which determines which notes will be carried over, has no
deafult vaule. To set this value to match all notes, execute the following:
git config --global notes.rewriteRef "refs/notes/*"
Now all notes will be carried over when doing rewrite operations like git
rebase
.
If you are using git format-patch
to format your changes to be sent as emails,
and you have some metadata stored as Git notes, you can pass the --notes
option to git format-patch
in order to append the notes to the email draft.
† 1: “This is the default for git log
[…] when there is no --pretty
,
--format
, or --oneline
option given on the command line.” ― man git-log
, git version 2.10.2
† 2: One practice/convention for metadata-in-commit-messages that is used in projects like e.g. Git and the Linux kernel is to add key–value pairs in the “trailer” of the commit message, i.e. at the bottom. See for example this commit by Linus Torvalds:
mm: remove gup_flags FOLL_WRITE games from __get_user_pages()
This is an ancient bug that was actually attempted to be fixed once
(badly) by me eleven years ago in commit 4ceb5db9757a ("Fix
get_user_pages() race for write access") but that was then undone due to
problems on s390 by commit f33ea7f404e5 ("fix get_user_pages bug").
In the meantime, the s390 situation has long been fixed, and we can now
fix it by checking the pte_dirty() bit properly (and do it better). The
s390 dirty bit was implemented in abf09bed3cce ("s390/mm: implement
software dirty bits") which made it into v3.9. Earlier kernels will
have to look at the page state itself.
Also, the VM has become more scalable, and what used a purely
theoretical race back then has become easier to trigger.
To fix it, we introduce a new internal FOLL_COW flag to mark the "yes,
we already did a COW" rather than play racy games with FOLL_WRITE that
is very fundamental, and then use the pte dirty flag to validate that
the FOLL_COW flag is still valid.
Reported-and-tested-by: Phil "not Paul" Oester <[email protected]>
Acked-by: Hugh Dickins <[email protected]>
Reviewed-by: Michal Hocko <[email protected]>
Cc: Andy Lutomirski <[email protected]>
Cc: Kees Cook <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Cc: Willy Tarreau <[email protected]>
Cc: Nick Piggin <[email protected]>
Cc: Greg Thelen <[email protected]>
Cc: [email protected]
Signed-off-by: Linus Torvalds <[email protected]>
See:
man git-interpret-trailers