by my definition is a ground breaking addon to git to help archive / replicate and optionally add metadata to large files by the esteemed Joey Hess.

It doesn't do several things which I've learnt the hard way:

  1. doesn't play well with S3 if you care about using original filenames
  2. doesn't actually efficiently store your binary files or diff them
  3. taxes my puny earth brain

Here are my complaints in detail:

Difficult to deploy

It's a large, monolothic, complex Haskell implementation. Basically you need the static binary release. Ideally it was just some C extending the existing git C code base.

I still haven't managed to get it running on my alarm device.

Making modifications to files is really difficult without replicating beforehand

If you are in the school of thought of making changes in your media file and not externally, you will hit the problems I have with git-annex.

Say you want to describe your images by adding exif tags to your images, you will need unlock and lock the file. Very cumbersome.

Direct mode sucks

Direct Mode makes it simple to change files, but direct mode has very bad disadvantages. For example, AFAIK if you delete the file in direct mode, you lose the file. No way to undo, unless you taken the time to replicate the original beforehand. ARGH!

One workaround is setting tags with git annex metadata. However what happens if you want to next edit (e.g. crop) the image? You will face the same problems... i.e. no way to back out sanely!

Furthermore I have an issue whereby when you a copy in a file to a direct mode git-annex repository, you can't seem to check whether it's tracked by git-annex or not, before making changes. I don't want to discard a file I don't have a back of yet.

The transcoder workflow

If you want to transcode a to film.mp4... that basically replaces film.mp4... again you will have a LOT of pain doing so.

Create an annex from a S3 bucket and vice versa


  • Make bucket public and add corresponding URLs as alternative downloads
  • Configure bucket policy to expire to glacier

The big workaround

Tbh I just want to use git commands without having to worry about:

For me, I require my files to be on S3 on a sane URL, like

My approach until maybe git-lfs matures, is to just check the large file. The big draw back is that I can't drop the files to free up disk space on my limited SSD on my laptop.

I might be able to work around that by working from an attached USB3 2TB drive. I will have an non-bare repository that I will s3cmd sync to S3.


Powered by Vanilla PHP feedback form