Re-implement Nix CI workflow on GitHub instead of CircleCI #10
Conversation
btlogy
force-pushed
the
4134.nix-on-github
branch
15 times, most recently
from
7fe97b1 to
6cf6bf7
Compare
Signed-off-by: Benoit Donneaux <[email protected]>
- ipython-8.24.0 not supported for interpreter python3.9 - magic-wormhole-transit-relay-0.2.1 not supported for interpreter python3.12 Signed-off-by: Benoit Donneaux <[email protected]>
btlogy
force-pushed
the
4134.nix-on-github
branch
4 times, most recently
from
4f429fd to
58a18cf
Compare
Signed-off-by: Benoit Donneaux <[email protected]>
btlogy
force-pushed
the
4134.nix-on-github
branch
from
58a18cf to
c13e42f
Compare
Signed-off-by: Benoit Donneaux <[email protected]>
- tkinter-7.3.12 not supported for interpreter pypy3.9 Signed-off-by: Benoit Donneaux <[email protected]>
|
I have some updates for tahoe-lafs on-nix dependencies in here: PrivateStorageio/ZKAPAuthorizer#481 |
Courtesy of the ZKAPAuthorizer nix expressions
|
Much better than packaging things IRL ;) |
... it's old (3.12 is the default), it's not in the NixOS binary caches, some dependency fails building and fixing it feels like picking over bones
... it has served its purpose (development) and is using runner time
|
I saw that, contrary to what I had written in my commit message 68f706e, also Python 3.10 succeeds, for which a bunch of dependencies have to be built because they are not in the NixOS cache anymore. But really IDK what we learn from making old Python versions work on NixOS - I think only testing the current stable channel's default (here: 3.12) is fine. @ben do you see a reason why we shouldn't remove Python 3.11 as well? |
Why is this so complicated? I thought the nixpkgs were defined in flake.lock?
... so we have less places where we have to change versions
This reverts commit 0c0b835. Not setting cores by hand made the build take 5m31s instead of 2m46s. Undoing that change.
... they run much better in GitHub Actions
My understanding of the implications is too weak I'm afraid. Though, his is sometime problematic for LTS distros which keeps using stable version longer than what upstream keeps supporting... But I've not yet a clear idea what this means for NixOS honestly. |
For reference:
|
So we have time. But what I meant is that some distro support things like Python longer than the upstream developers.
😱 |
Thanks @meejah. I still don't know whether we learn enough from running our tests with non-standard Python versions on fast-moving platforms like NixOS. I propose to check all Pythons on one platform (say, Debian or some docker container that is useful for this) and then check all the different OSes we want to support with whatever is a sensible default for them in ticket: 4145. |
No description provided.