Hacker News Reader

The first few lines in the repo you mention illustrate the problem:

> Note that this repository is no longer being maintained. Use this at your own risk, and with the understanding that it is not being maintained, work is being done on its replacement, and that no support or assistance will be offered.

And its replacement syncserver-rs is in this forever-unfinished state so they've effectively killed self-hosting. At least from a support standpoint.

I don't understand. The original was barely 500 lines of code in two files, everything else being packaging machinery. Yes it's python 2, but how hard could have it been to port 500 LOCs to 3.x...? Even assuming some crypto libraries broke, surely there are working equivalents?

Instead, they went for a Big Rewrite in Rust, resulting in a much bigger codebase that will be much harder to maintain... and broke it anyway.

Their issue tracker suggests that it’s mostly due to their team being focused on the main Firefox service:

https://github.com/mozilla-services/syncserver/issues/264#is...

I’m sure it wouldn’t be that hard to port but if they’re committed to a different direction it’s probably best not to give the impression that you should plan to keep using it.

Yeah, but even internally it makes little sense. You've replaced a very simple system built on a widely-known language with a massive ecosystem, something you could even call "boring technology" these days (python), with a larger implementation using a niche language with a small ecosystem and a minuscule pool of skilled developers. It's now even more unlikely you'll ever get any patches from outsiders, it will be harder to hire people to maintain it, harder to monitor and expand the system... And this for a critical system.

I mean, as a job-security device, I'm sure it works out great. As a strategic business decision, though, it looks debatable as a minimum - and I'm being charitable.

Not sure you could call Rust a niche language any more given how many devices it’s running on, but especially at the organization which created it that seems like an understandable choice. I have a lot of experience with Python but can totally sympathize with people running a highly-visible service prioritizing resource utilization.

> Not sure you could call Rust a niche language any more

First comment under "The Adoption of Rust in Business": https://news.ycombinator.com/item?id=34691509

A bit further down: https://news.ycombinator.com/item?id=34691893

Yes, note that’s .gov where they also have trouble hiring technical staff of every sort. I happen to work in .gov and we have multiple people writing Rust so anecdotes are not probably reliable.

This is better - still a self-selected audience but a large one - and it shows around 10%. That’s not huge but I think it’s larger than “niche” given the growing popularity and impact.

https://survey.stackoverflow.co/2022

Good point, I'll update the comment to indicate syncserver is now unmaintained.

Of course, everything self-hosted is at my own risk etc. And official support is a luxury in this kind of FOSS, I generally don't count on it.

But since it works for me, it may be worth sharing the details.

I agree that a lack of support isn't a huge issue for this kind of thing. I'm more worried about the maintenance, like a vulnerability remaining unpatched (it being python 2 doesn't help either) or a protocol change that makes the old server suddenly unviable.

I really wish they'd just finish the Rust version. That would solve everything.

Sounds like classic case of ”Second System Effect”

Better known as "if its not broken, break it"

https://en.wikipedia.org/wiki/Second-system_effect

Have you considered upstreaming the package and module to nixpkgs? It would be awesome to have those one line of config away for all nixpkgs users.

I briefly considered it, but there are some big problems:

- The new replacement package, syncserver-rs, has already been added to Nixpkgs, along with a NixOS module [0]

- Python 2 and the package itself are no longer supported, meaning...

- support in Nixpkgs is pretty rough - dependencies have been updated such that they no longer support Python 2, or have been removed outright, meaning...

- syncserver has already been removed from Nixpkgs

- in order to avoid the problem of unsupported dependencies, I delegated dependency resolution to a community project, poetry2nix[1]. Even so, it was a nightmare to get working, lots of hacks and workarounds were needed

Of course, my repo is public, so anyone is welcome to use the Flake within. If they do, I'd suggest dropping me a line so I don't do anything drastic to the package or module.

[0] https://github.com/NixOS/nixpkgs/pull/176835

[1] https://github.com/nix-community/poetry2nix

Yeah, Python2 alone is enough to not bother. Hopefully the new version "ships" soon.

>Note that Mozilla do provide a script to delete all your data from their servers[1], in case you were using Firefox sync before going self-hosted.

Note : You'll need to disable 2fa first on your account before you can delete the data.

I think Nix, as struggle-full as it was, was the right choice, here, although I don't see the flake.nix file

The flake starts at this level: https://git.sr.ht/~williamvds/config/tree/master/item/nix

Struggle is a good way to put it! I feel I hit one of Nix's pain points here: packaging software whose dependencies are largely missing from Nixpkgs. On top of that, it was a Python app, and the state of Python packaging is quite a mess.

I also feel it Nix was the right choice. My journey was rather harsh, but I do believe in its approach.

The thing that guts me a little is that Nix is also perfect for software preservation, which unfortunately must include Python 2 apps. But meanwhile they've dropped support for Python 2, which means you either have to go with some clever flake solution or something like what you did.

I'm dealing with an app that is breaking every time I try to upgrade a few-years-old version of Node, a version that is no longer on nixpkgs. Right now I just want something that works; I didn't sufficiently budget for ongoing maintenance as the whole site's behind a login.

Another option is to find a Nixpkgs commit which does contain the version of a package that you want. You can pin and import it as a Flake input, and use the old version of the package provided.

So in its own way, Nix and Nixpkgs do serve the purpose of software preservation, as long as you have the full commit history for Nixpkgs.

Preservation doesn't imply maintenance however. It's fair for Nixpkgs to delete packages that are no longer supported, as that's a burden on the limited number of maintainers.

By recreating the package and module in my own Flake, I've taken on responsibility for maintaining it myself, and not adding any work to the Nixpkgs maintainers.

NixOS modules in particular aren't easy to extend, so I ended up copying it entirely. And to be honest, the thought of using an old version of Nixpkgs didn't cross my mind while setting up syncserver. Hopefully it's easier to maintain with poetry2nix now that I've got the package working.