| xek | Artem Goncharov (gtema) I'm in the heat session | 06:05 |
|---|---|---|
| gtema | Good, I will join in 30+ min | 06:05 |
| opendevreview | Grzegorz Grasza proposed openstack/python-keystoneclient master: Support explicit EC2 credential consumers https://review.opendev.org/c/openstack/python-keystoneclient/+/1010419 | 06:36 |
| opendevreview | Grzegorz Grasza proposed openstack/keystone master: Bind EC2 credentials to exchange consumers https://review.opendev.org/c/openstack/keystone/+/1010401 | 06:37 |
| opendevreview | Grzegorz Grasza proposed openstack/keystone-specs master: Propose consumer-bound EC2 credentials https://review.opendev.org/c/openstack/keystone-specs/+/1010380 | 06:37 |
| zigo | gtema: I asked GLM-5 (our best model at Infomaniak) to evaluate how many Rust package I'd have to do to get Keystone-RS in Debian. The result is: about 126 crates not in Debian, about 60 needing version bump, and maybe about 7 more NEW for unit testing. | 15:41 |
| zigo | It's not impossible to do, but that's a non-negligeable effort. | 15:41 |
| zigo | GLM-5 says: "130–190 package-level tasks ≈ 1–3 months of sustained work" | 15:42 |
| zigo | I would say a month at least doesn't seem unrealistic. | 15:42 |
| gtema | thanks for looking, zigo. I definitely get the point. If we have a go "decision" we can talk about this in detail | 15:42 |
| gtema | and there are things you might not really need (like optional features) | 15:42 |
| zigo | Well, my point is: it'd be nice if you could reduce the amount of (build-)dependencies. | 15:43 |
| gtema | if we identify packages with more effort we can talk about vendoring certain logic | 15:43 |
| zigo | Vendoring is a no-go for me, I would then need to un-vendor. | 15:43 |
| gtema | I am always considering this when adding any new dep, believe me. I will have another look whether something else can be stripped | 15:44 |
| gtema | under vendoring I meant just re-implementing necessary functionality directly instead of pulling dep if that is just some 100loc | 15:44 |
| gtema | unfortunately lot of those deps are transitive | 15:45 |
| gtema | till now I didn't want to avoid making our "dev" life unnecessary complex by considering next steps. Once we are at that point we can have a look at every individual dep and decide whether we can get rid of it | 15:47 |
| gtema | zigo, I opened https://github.com/openstack-experimental/keystone/issues/1534 for tracking. I can definitely make extism/wasmtime deps optional, not-default - disable wasm plugins. I also work on aws-lc from the other side - adding FIPS support requires aws-lc-rs, but I try to make the backend pluggable now. Feel free to comment there on which things would be the hardest or we should try to avoid from your pov | 16:04 |
| zigo | I have very few experience with Rust packaging. | 16:08 |
| zigo | The only thing I did was rabbitmqadmin-ng, plus its 6 dependencies. | 16:08 |
| zigo | I made tooling to automate new packages though... :P | 16:08 |
| zigo | No way I'll write 126 new packages all by hand. | 16:08 |
| gtema | yeah, AI says wasm family is pretty complex, so I will definitely hide it behind the option | 16:09 |
Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!