The open source Git project just released Git 2.17.0, with features and
bugfixes from over 60 contributors.
Here’s our look at some of the most interesting new features from the
last few versions of Git.
Coloring moved code
When you’re reviewing a commit that moves some code, its diff will show
a big chunk of lines being deleted in one place and a big chunk of
lines appearing elsewhere. The commit message may claim that this was
pure code movement, but how can you verify that? And if some lines needed
to be changed, how can you easily see which lines?
Git v2.15 offers a new
which colors groups of moved lines differently. Use this option
for any command that generates a diff:
git log -p,
git, and so on.
The blue lines show the blocks that were moved verbatim, while the red
and green lines show the corrected typo. Note that these aren’t the
default colors; we’ve just used them to highlight this simple example. By
default, old and new blocks get their own colors, and non-adjacent
groups of lines are colored independently. This helps when there are
several moved blocks in a single diff. You can configure the
to match your preferences.
Speeding up status with watchman
One reason that Git is fast is avoiding unnecessary disk reads.
For example, when you run
git status, Git
stat() system call to check whether each file’s size or
timestamp has changed since the last time Git checked. If not, it
assumes that its contents haven’t changed and avoids reading the file
contents at all.
This helps a lot, but it still requires Git to call
stat() once for
every file in your working copy. If you have a lot of files, that work
adds up. This can be further exacerbated if you’re on an operating
stat() is more expensive.
Git v2.16 adds a new
that relies on the operating system to tell us what has changed. Doing
so allows us to avoid the repeated
stat() calls, making the work
proportional to the number of changed files, not the total size of the
The hook provides a generic interface into which you can plug any
tool, but most people will want to use the
watchman tool, which runs on
Linux, macOS, and Windows. Git ships with a
Finding objects in history
Have you ever gotten hold of an object hash for a blob or tree like
c1626490c2 and had no idea what path it corresponds to, or in which
commits it could be found? Perhaps somebody else mentioned the object,
or a Git command lke
fsck complained about it.
There’s no single defined answer for these questions. Because Git
commits are snapshots of the tree, an object may appear in many commits:
it may get introduced in commit
A, then persist in commit
only touches another file, and then finally go away in commit
Likewise, an object may appear at multiple paths if it’s renamed or
Getting this information used to require a pipeline of arcane commands.
But now you can use
--find-object to select the commits you want and
display them however you like.
Here we’ve used
--raw to easily show that these commits are the ones
where the object comes and goes (and that Git omitted any in the middle
which didn’t affect it). You might be more likely to use
-p to show
the full diff.
Grepping with function context
When you search for a string with
git grep, it’s often helpful to see
the surrounding context. You can use
-C 2 to tell Git to show you two
lines of context on either side, but it’s hard to know the right number
in advance without ending up with too much or too little context.
For many years,
git grep has had a
--function-context option (and
its shorter alias
-W) to show you the whole function surrounding each
match. This uses language-specific patterns if you have the appropriate
defined, or it can fall back to some generic heuristics.
Since Git v2.16, it can now also find and show comments before the start
of a function, which provides even more context for your matches.
-W, we’re left with very little information about how the
function is used. With it, we can see the whole function, including its
- The diff
--indent-heuristicoption we mentioned in
our v2.11 post
has now become the default. If you’ve never tried it, you may now see your
diffs subtly improved. Automatically!
- Our v2.11
also talked about a new protocol that lets Git more efficiently
invoke filters like Git LFS. That protocol has been extended to allow
filters to return results asynchronously. That lets Git proceed with
the work of checking out files while a filter like Git LFS is waiting
on the network.
- If you build Git with support for Perl Compatible Regular
Expressions, it now supports libpcre2, as well as JIT support for
both libpcre1 and libpcre2. This can improve the performance of
up to 70%.
- Our v2.13 post
described Git’s pathspecs and how globs like
*.cno longer require
you to enter
--on the command line. That convenience has been
extended to “magic” pathspecs, letting you more easily write exotic
git ls-files ':attr(filter=lfs)'.
- Have you ever found yourself in the middle of a conflicted rebase and
needed to see the original patch? And you couldn’t remember that the
magic invocation is
cat .git/rebase-apply/patch? Or sometimes
/rebase-merge/, depending on how rebase was invoked? Now you can use
git rebase --show-current-patch, and even tab-complete it.
- Some projects put trailers like “Signed-off-by:” or “Reported-by:” in
their commit messages to give additional information. These are meant
to be machine-readable, but parsing them can be non-trivial. Now you
git log --format='%h%n%(trailers:only,unfold)'to get just
key: valuetrailers, with any whitespace normalized.
- The epic quest
to transition Git off of the SHA-1 algorithm has continued. There’s a
public transition plan,
and the internal code cleanups to prepare for an alternate hash
function are nearly complete.
- Have you ever enabled automatic line conversion or another
after you’ve already run
git addon the files? Fixing that used to be
but since Git v2.16 you can just use
git add --renormalize.
- You can enable or disable the pager for an individual command like
git logby setting the
pager.logconfig option. But this was
useless for commands like
git tag, which sometimes generate a lot
of output (when listing tags) but other times generate none (when
creating a tag, in which case having to quit the pager is an
configcommands have all
learned only to invoke the pager in their “list” modes, and now
default to paging in those modes.