Track CI duration wins from slim shells + committed artifacts

[thedavidmeister]

Goal

Track the CI duration wins from the slim-shell / committed-artifacts work, so we can see what's actually moved on main vs what's still pending in PRs.

Baseline on main (latest run per workflow as of 2026-05-30)

Workflow	Baseline	Notes
rainix-sol	2m	already fast
copy-artifacts	5m	was 2m pre-#2603; build-meta hook added 4m, rain.cli Cachix hit saves 1m
Subgraph unit tests	16m	full default devshell, runs forge build + npm ci + graph codegen + matchstick
Rainix CI (standard-tests matrix)	~25m	full default devshell, ./pointers.sh (dead forge build) + cargo + wasm + js bindings
Test webapp / Test ui-components	20–22m	webapp + ui-components vitest + svelte-check
Vercel Production	11–22m	variable, not in our scope

Wins already landed

• rain.cli remove input and output from take orders #19 + raindex Pin rain.cli via flake input for build-meta (deterministic Cachix hit) #2604: rain.cli is Cachix-published, raindex's script/build-meta.sh uses the flake-locked rain via .#rain-cli. copy-artifacts 6m → 5m.

Pending PRs and their expected gains

• raindex Commit subgraph/abis + subgraph/generated; use rainix-subgraph-test reusable #2606 (subgraph committed artifacts + thin rainix-subgraph-test wrapper, slim #subgraph-shell): drops forge build + graph codegen from the test path; matchstick runs over committed subgraph/abis/ + subgraph/generated/. Expected: Subgraph unit tests 16m -> low single-digit minutes.
• rainix Token detail page #212 (pkgs.sqlite in rust-build-inputs + subgraph-test task npm ci): unblocks raindex's rust matrix on #wasm-shell (rusqlite needs system sqlite), and makes subgraph-test self-contained over committed bindings.
• Phase 2 — rainix.yaml slim-shell refactor (not yet opened, depends on Token detail page #212): drop ./pointers.sh, switch each matrix task (rs-test, cargo build --release, rs-static, wasm-artifacts/test/browser-test, test-js-bindings) to nix develop github:rainlanguage/rainix#wasm-shell -c <inlined cmd>. Expected: Rainix CI 25m -> meaningfully less; biggest unknown, will be measured post-merge.

Other observations

• forge build in ./pointers.sh is dead weight for the rust matrix — none of the crates compile-time or runtime depend on out/ (no sol! macros pointing at it, no include_bytes!, no shell-outs to forge/cast/anvil). Phase 2 above drops it.
• Test webapp / Test ui-components and Rainix CI's js bindings are next-tier targets but not currently scoped.

[linear]

RAI-727

[thedavidmeister]

Closing — the in-scope CI-duration work has landed:

• Subgraph unit tests ~16m → ~3m: subgraph/abis/ + subgraph/generated/ are committed on main and test-subgraph.yml runs the rainix-subgraph-test reusable. (Commit subgraph/abis + subgraph/generated; use rainix-subgraph-test reusable #2606 was closed, but its substance shipped another way.)
• Slim-shell refactor done across test-subgraph, test-js-bindings, test-ui-components, test-webapp, and the vercel-* workflows; the monolithic Rainix CI matrix was split into per-task workflows (rs-test, wasm-test, wasm-artifacts, …).
• Enablers Pin rain.cli via flake input for build-meta (deterministic Cachix hit) #2604, Use rainix copy-artifacts meta hook; drop git-clean #2603, rainix#212 merged.

The only remaining items are the next-tier targets this issue itself flagged as out of scope (Test webapp / Test ui-components vitest durations) — worth a fresh, narrowly-scoped issue if/when picked up. The approach is now encoded in AGENTS.md.