SPARSE-CHECKOUT(7)     Miscellaneous Information Manual     SPARSE-CHECKOUT(7)

NAME
     Sparse Checkout – a cool git feature

DESCRIPTION
     I was going to write a post about git-subtree(1) (and I still plan to!)
     but while talking about it with a friend I came across another command:
     git-sparse-checkout(1).  I got pretty excited because I already had a use
     case for it.

     git-sparse-checkout(1) does pretty much what it sounds like.  It lets you
     only have a subset of files in the repository actually “checked out”.
     This is really useful for huge respositories where you are only
     interested in some part of it.  Any operation touching the working tree
     is much faster because it can skip all the files you don't care about.

     My use case is with the FreeBSD ports(7) tree, which recently moved to
     git and contains almost 14 thousand files.  Working with the whole
     repository was super painful.  git-status(1), which I run as a habit when
     my shell is idle, would take dozens of seconds to check the whole working
     tree and report back.  (I didn't get any real time measurements before
     enabling git-sparse-checkout(1), and I'm not about to disable it now,
     since it'd have to check out all those files again.)  I'm only actually
     working on a small handful of ports, so all that work is wasted.  Time to
     turn on sparse checkout:

           git sparse-checkout init --cone

     The --cone option here (which I keep reading as “clone” because it's git)
     restricts the kinds of patterns you can use to select files to check out,
     but makes the calculation more efficient.  Basically it means you can
     only select paths along with everything below them, which I think is
     pretty much always what you want anyway.  Enabling sparse checkout can
     take quite a while because it has to do a lot of un-checking-out.  I
     should mention that you can pass --sparse to git-clone(1) to avoid ever
     checking out the whole tree.

     The default selection when you run init is to check out all the files at
     the root of the repository, but none of the subdirectories.  For
     ports(7), I also want to check out the shared scripts and Makefiles:

           git sparse-checkout add Keywords Mk Templates Tools

     And then I can selectively check out just the ports I'm working on:

           git sparse-checkout add irc/catgirl irc/pounce

     After enabling sparse checkout, git-status(1) takes what I'd call a
     normal amount of time.  I also did this on a couple-weeks-out-of-date
     copy of the ports(7) tree, and when I ran git-pull(1) it was also really
     quick, because it didn't have to bother updating all those files I'm not
     interested in.  It still downloads all the git objects, of course, and
     you can just add any new paths you need to the sparse checkout list.  My
     disk usage also went down by about a gigabyte.

     I'm super pleased to discover this part of git, because it makes working
     with huge and/or monorepo-style repositories so much more feasible!  You
     can see how I came across it, since git-subtree(1) is also a useful tool
     for monorepos.  Stay tuned for that post, I guess :)

AUTHORS
     june <june@causal.agency>

Causal Agency                    June 9, 2021                    Causal Agency