| gouthamr | clarkb: this was a great read: https://superuser.openinfra.org/articles/been-there-gated-that-what-ai-builders-can-learn-from-a-decade-of-open-source-ci/ | 17:30 |
|---|---|---|
| clarkb | awesome | 17:35 |
| gouthamr | i completely resonated with the sentiment.. | 17:38 |
| gouthamr | the case for zuul is always relevant - while being born in OpenStack, it continues to be the reason OpenStack remains viable.. standing up for quality against both humans and AI agents.. :D we could however bend over backwards and change the very tests that will gate the future.. :( but those fears are only more pronounced if we give up on the human judgement and the human centered aspects of our community | 17:38 |
| clarkb | an unfortunate reality for me right now is I've had to upgrade software we don't maintain multiple times to patch up some pretty glaring errors in code that was written by LLMs. It really does seem like we need to reorientate how we approach software development to avoid these issues (and openstack is already pointed in largely the right direction due to its history) | 17:43 |
| clarkb | everyone is full steam ahead on adding features because LLMs are making that easy when in reality maybe we'd be better served by first writing gates and comprehensive test cases then working on features with those safety nets in place | 17:44 |
| gouthamr | (sean-k-mooney was saying something similar in #openstack-agentic-workflows) | 17:44 |
| sean-k-mooney | for what it worth the last 2 feature i have been working on in cybrog have been writeen by ai with alot of my gudice | 17:46 |
| sean-k-mooney | but i created them by deploying multi node devstack including with tempest | 17:46 |
| sean-k-mooney | and not only having the agent run basic unit tests | 17:46 |
| sean-k-mooney | but write the temepst test and run/validate those | 17:46 |
| sean-k-mooney | before pushing a single patch | 17:46 |
| sean-k-mooney | so when i did push the patches | 17:47 |
| sean-k-mooney | i had fully working end to end ci on the first attempt | 17:47 |
| clarkb | and you already have that framework and expectation beacuse openstack has operated that way for over a decade | 17:47 |
| sean-k-mooney | yep | 17:47 |
| clarkb | the problem is a lot of software is getting written without even the concept that you can/should do this | 17:47 |
| clarkb | and it was maybe ok when everything was hand curated and looked over | 17:47 |
| sean-k-mooney | ya there are too parts to that | 17:48 |
| sean-k-mooney | one is the idea that just becasue changing the code is easy/cheap that ci is not needed which is wrong | 17:48 |
| sean-k-mooney | but the other part of it is that investing in local easy to run validation is not reiqured | 17:48 |
| sean-k-mooney | for me tiging the agent into the testing loop is basiclly a prequist for having useful output | 17:49 |
| sean-k-mooney | you dont have to go full TDD | 17:49 |
| sean-k-mooney | but you shoudl make the testing a core part of the generation loop and coevovle both toghether | 17:49 |
| gouthamr | maybe it is that getting a non-local LLM to work seems super easy/lightweight and developers are relying on the CI to do _any_ testing | 17:49 |
| sean-k-mooney | clarkb: but as you said that is a core part of the openstack culture. "if its not tested its broken" was one of the first tings i was told when starting with openstack | 17:50 |
| sean-k-mooney | gouthamr: perhaps, but if you want an llm to create a dev env for you it can. i tend to create new dev vms for a featre and let the llm work in that vm. | 17:51 |
| sean-k-mooney | for openstack that perhaps more work then a webste but devstack is not that hard to use adn claude can deploy it for you if you ask it | 17:52 |
| gouthamr | not at all yeah.. i'm thinking why someone wouldn't | 17:57 |
| gouthamr | i think this may be something we reinforce to new and relatively new contributors... testing was/is a baseline expectation when landing a feature or a bug fix.. and that's the way you earned trust in the community.. if you abuse the CI, that's one way you're not going to get review attention | 17:58 |
| sean-k-mooney | i mean its easy for me to say today that "devstack is simple" whewn it took me 6-8 weeks to relibely get it workign when i first started (pro tip dont try to use python 2.6 in centos 6.4 when master had just move to python 2.7) | 18:37 |
| gouthamr | that tip will be super helpful as soon as i find a way to install that deadsnake | 18:38 |
| sean-k-mooney | first rule of devstack use a supproted release and dont install things like eclipse or a gui before hand | 18:38 |
| sean-k-mooney | gouthamr: 2013 me would have loved ot hear it. | 18:38 |
| gouthamr | :D | 18:39 |
| sean-k-mooney | honestly when i finally stop usign our lab images and just did a clean install like devstack expected almost al my issues went away | 18:39 |
| opendevreview | Ivan Anfimov proposed openstack/openstack-manuals master: Add OpenStack-Ansible to deployment guide page for 2026.1 https://review.opendev.org/c/openstack/openstack-manuals/+/995424 | 23:36 |
| opendevreview | Ivan Anfimov proposed openstack/openstack-manuals master: Add OpenStack-Ansible to deployment guide page for 2026.1 https://review.opendev.org/c/openstack/openstack-manuals/+/995424 | 23:39 |
Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!