← all records

State and Component

pitch-d81872 · in warm_bubble

  • its tasks need 4.6 weeks with the people on them, more than the 4.0 the bet buys over 2 people, 1 full-time between them — cut scope, re-bet it, or put more people on it
  • No No-gos section. The pitch template asks for one — it is what keeps the appetite honest.

Progress 3/5 wk

Problem

Components and state are being brainstormed at https://c2sm.github.io/icon4py-knowledge/tags/components . We need to start converging on a design, so that interfaces stop diverging and scientists are helped rather than hindered in separating components cleanly. What holds now is an informal contract.

One concrete instance of the problem: the physics EntryState of the two-layer coupling (#1436) mixes two categories as flat attributes -- six pointers for prognostic fields (exner, theta_v, rho, vn, w and the tracers) and six owned diagnostic buffers (ta, tv, pressure, pressure_ifc, u, v, diagnosed once per step and then frozen). The distinction exists only in comments.

Appetite

Full cycle, four person-weeks over Mikael and Jacopo.

Solution

Agree the hard requirements for the interface across the team and set those as the MVP. Build as many proofs of concept as possible, so that real designs can be compared rather than argued about.

Rabbit holes

Leaving the design open for extension, for eDSLs among other things, is welcome; letting it grow very large is not. The design has to be usable soon rather than in two years.

No-gos

For later