I like flakes a lot, speaking as a type of user which I’ll call a “casual packager”.
This is to say that I like being a good citizen and sharing my packaging efforts… while also simultaneously feeling totally uninterested in becoming an owner-for-life. Flakes let me share a package without those pesky strings – when the user installs a package using one of my flakes, a personal lockfile gets generated at the latest git commit and that’s that. If the user doesn’t like the version they get, then the power is in their hands to choose a different git ref via their own generated lockfile.
Obviously this is something of a user footgun, especially for consumers of high-impact or security-critical applications, but most of those things are already important enough to get packaged. When it comes to niche, infrequently updated stuff, this approach works super well and helps to draw many reluctant packagers like myself out into the open.
I think that Nix flakes have some considerable benefits, such as:
Convenient pinning of evaluation-time dependencies
Eliminating pointless rebuilds of code by only including tracked files in builds
Making Nix code, on average, much more reproducible by pervasive pinning
Allegedly caching evaluation
Possibly making Nix easier to learn by reducing the amount of poking at strange attribute sets and general NIX_PATH brokenness
However, at the same time, there are a few things that one might be led to think about flakes that are not the most effective way of doing things. I personally use flakes relatively extensively in my own work, but there are several ways I use them that are not standard, with reason.
Flakes are optional, and as much as some people whose salary depends on it might say otherwise, they are not the (only) future of Nix: they are simply a special entry point for Nix code with a built in pinning system, nothing more, nothing less.
Nix continues to gather a reputation for bad documentation, in part because the official documentation for nixpkgs and NixOS is de facto not allowed to talk about flakes, as a policy. This situation is certainly partially due to a divide between Nix developers and nixpkgs developers, which are groups with surprisingly little overlap.
Flakes also are a symptom or cause of much intra-community strife between “pro-flakes” and “anti-flakes” factions, but this situation is at some level a sign of broken consensus processes and various actors trying to sidestep them, an assumption by many people that the docs are “outdated” for not using flakes, and the bizarre proliferation of flakes everywhere in blog posts or tutorials leading to a belief that they are required for everything.
This post is about how to architect Nix projects in general, with a special eye on how to do so with flakes while avoiding their limitations. It tries to dispel misconceptions that can develop in such a monoculture.
Read all of Flakes aren’t real and cannot hurt you: a guide to using Nix flakes the non-flake way
Flakes are optional
I think the bigger point is that they are experimental. Who wants to go into production with experimental stuff? Unfortunately, if my employer were to read a suggestion of using experimental stuff in production, it would be immediately rejected.
nixitself, however, has been getting some people interested at work and if things go well, might be integrated into production solutions we sell to customers.They really should be moved out of that category soon. They’ve been so fully embraced and used by the community that they’re effectively standard now.
The article clearly states otherwise
Flakes also are a symptom or cause of much intra-community strife between “pro-flakes” and “anti-flakes” factions
Oh, that’s easy – that’s utter bullshit. Article is generous, this is a blog post.
😂 why do you believe that?
Because I’m pretty active in the Nix community and have never seen a single anti-flakes comment anywhere. Plenty of people advocating for flakes, and those who don’t use or understand flakes. But never anyone who actively dislikes them.
OK, I see where you’re coming from. You’re interpreting “anti” as “dislike”. Have you considered interpreting it as those who don’t use and don’t want to use flakes?
Just because there is disagreement, doesn’t mean that it has to be tied to negative emotions like hate or dislike.
Well no, because that’s not what “anti” means.
And one can absolutely use NixOS without using flakes, whether they’re marked experimental or not. They’re simply a way to make inputs declarative, rather than the nix-channel method which is imperative.
So there’s no “disagreement” here either.
It must be nice to have never productionized a prototype; meanwhile, in the land of startups, putting experiments into production is standard operating procedure, despite being a bad idea.
The more important part is that flakes are unstable. The concept is solid and clearly works at scale, but the precise API available to users is not yet finalized or versioned.
It must be nice to have never productionized a prototype; meanwhile, in the land of startups, putting experiments into production is standard operating procedure, despite being a bad idea.
You’re making assumptions. I worked in startups that did took great care not to beta software or libs. We were in production and changing stuff all the time - multiple times a day, but that was one of the rules. When you work with time-sensitive, always on environments with world-wide customers, bringing down production for multiple hours could mean the end your company in those early stages.
Maybe I was extra-ordinarily lucky with the startups I was in.Just because other startups do it, doesn’t mean that it has to be done.
I need to read it, and do everything the other way
It took me months to understand that i don’t need to nix channel update if i’m on flakes
Or do i need? 😂
