Repository navigation
Replies: 1 comment
|
Hey Ghaith, thanks for asking before building, this saves us both a lot of back and forth :)
One timing thing: I'm landing a batch of agents view work and tagging v0.13.2 today, so please keep off internal/app/cockpit*, render_cockpit* and inbox_* until the tag is out, otherwise we'll fight over rebases. Since you offered, here are a few I'd love you to own, in any order:
Thanks a lot for all of this, really appreciate it! |
Uh oh!
There was an error while loading. Please reload this page.
We are starting on the scope list you gave for the next round of work: performance, memory, cpu, ram, state machines, dedup and a single source of truth, everything configurable, comments, and tests. The performance, memory and dedup parts go the usual way: one finding, one PR, measured before and after. Four of the lines are about the shape of the code, so a few questions first, so we build them the way you want:
State machines. Our plan: we list every implicit one we find (bool clusters, string or int modes) in a table with its states, events and the combinations it allows that should not happen. Then one machine per PR: a named state type, one transition function that refuses a move it does not allow, and the state by event table in the PR body. Is that the shape you want, and should we start where we can show a wrong combination is reachable, or go area by area?
Everything configurable. Should we first send you a table of the hardcoded values a person can feel (timeouts, limits, sizes, colors, keys) and you pick, or do you want all of them as options with today's value as the default? Internal tuning numbers (buffer sizes, poll intervals) we would leave out unless you say otherwise.
Comments. AGENTS.md asks for doc comments on packages and exported names, so we would keep those and remove only body comments that restate the next line. One PR per package, or one sweep?
Tests. We would remove only unit tests outside the six kinds AGENTS.md keeps, and tests whose assertion another test already makes, each with a run showing the case is still caught. One PR per package, or one sweep?
These can turn into many PRs. How many open at once suits you?
All reactions