Skip to content

ci: deploy to production#803

Merged
fricklerhandwerk merged 18 commits intoproductionfrom
main
Feb 13, 2026
Merged

ci: deploy to production#803
fricklerhandwerk merged 18 commits intoproductionfrom
main

Conversation

@fricklerhandwerk
Copy link
Copy Markdown
Collaborator

No description provided.

fricklerhandwerk and others added 18 commits February 10, 2026 09:14
This reverts commit deaf331.

The calculation was too tight.
It both OOMs on the evaluation and runs out of tmpfs disk space.
Reverting so we can use staging for experimenting with evaluations again.
We never serialise the evaluated attribute, and the loop helper is not used anywhere.
This is a rather large change that does multiple things at once:
- Rename a few helpers
- Deduplicate derivations by degrees of freedom
- Improve accuracy of time measurement

  We now capture the entire processing time for each data type separately.

- Only add maintainers and licenses from the rolling release

  This is because we must assume that everything else is stale.
  There's a chance this can produce temporarily out-of-sync results:
  - A license is deprecated between commits A and B
  - `nixos-unstable-small` (B) evaluates before `nixos-unstable` (A)
  - State of A will be shown
  - Only when `nixos-unstable` (B) evaluates, state of B settles
  This can only be fixed by observing commit order, but this is far out of scope for this change.

- Remove maintainers' GitHub handle uniqueness requirement

  While since recently, any given Nixpkgs commit will have unique GitHub handles, this is not necessarily true across commits.
  In particular this allows handling the special case where a maintainer changes GitHub IDs but recovers the handle:
  - commit A: id X, handle Y
  - commit B: id X, handle Z
  - commit C: id W, handle Y

The change was primarily smoke-tested by running evaluations manually.
This means, we're essentially relying on correctness by self-consistency via types and the logic being simple.
It would be really good to have fine-grained tests of the desired behavior though (property testing would make most sense here).
This should have been part of the previous commit, but got lost in history rewrites.
We're not using that data for anything, but it takes up most of the space.
So much that by now it's too much to meaningfully migrate around.
If we ever need it, it can be recreated by evaluating the respective Nixpkgs commits.
Now that we're not tracking dependencies until further notice, there's no point in accessing the disk at all.
@fricklerhandwerk fricklerhandwerk merged commit 8463776 into production Feb 13, 2026
7 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants