> as long as you enable LFS support

You still end up with two copies of the same file, one in the local LFS “server”, one in the work tree, no? (I only played with LFS a bit many years ago when it first came out, so I could be wrong.) Unless you take into account deduplication built into certain filesystems.

You don't get copies until you need them - thats the point entirely. More details here:

https://www.atlassian.com/git/tutorials/git-lfs

Also saves you from beating up your index with every change.

I've been using GIT LFS for the last 6 months with an Unreal Engine project, with multiple gigabytes of files being tracked, and it really is painless.

> You don't get copies until you need them

I know. But you do need them, and files in your work tree don’t magically disappear when you commit them in (presumably). So either you delete the work tree copy immediately after pushing it to LFS server, and duplicate the server copy every time you need to access it, in which case the file is only duplicated then but comes with elevated cost of access, or the latest copy sits around costing double the amount of space at all times.

I don't see the issue? Either you want to use Git or not. I have gigabyte-scale files in my 6-month old repo's and haven't ever run into any issues. Of course this may be because my git server is right next to my desk and I'm on gigabit ethernet ..

or just use a bare repo?

Aside: are there non-binary and/or non-large blobs? I'm thinking along the lines of ATM machine / PIN number but maybe BLOB no longer implies "binary large" without being explicit.