release: v0.9.0

[avrabe]

Summary

Bump workspace version 0.8.0 → 0.9.0.

Measurement and harvest release. First objective measurements of LOOM against wasm-opt -O3 across multiple workloads, plus harvesting of v0.8.0's infrastructure into concrete wins on component-shaped fixtures.

Merged PRs

PR	Title
#110	feat(measurement): corpus-wide LOOM vs wasm-opt harness + measured wins on components (PR-P)
#111	feat(opt): span-based CSE replacement for pure Call exprs (PR-K2)
#112	feat(opt): grow Souper rule set to 12 identities, wire into pipeline (PR-L2)

Real measured wins (first time)

Workload	Baseline	LOOM	wasm-opt -O3	LOOM Δ%	wasm-opt Δ%
gale	1,941	1,846	1,925	−4.9%	−0.8%
calculator_root (2.3MB)	2,337,724	2,327,794	(errors)	−0.4%	n/a
simple_component	261	212	(errors)	−18.8%	n/a
calc_component	442	392	(errors)	−11.3%	n/a

Two strategic facts established:

1. LOOM beats wasm-opt -O3 on gale by 4.1 points at total-file level. First measured workload where LOOM dominates.
2. PR-M (v0.8.0) delivers −11% to −19% on small adapter-heavy components. The percentage dilutes on large components. The strategic moat is real: wasm-opt cannot process Component-Model components at all (errors out on every one).

Two surprises discovered during measurement

1. PR-L (v0.8.0) was never wired into the optimizer pipeline. The module existed but loom optimize never called it. PR-L2 adds the missing registration. Discovered via --stats per-pass breakdown.
2. The Z3 verifier models every Call as a fresh symbolic constant, so PR-K2's cross-call CSE dedup is rejected by the validator with a counterexample. The replacement infrastructure ships, but the tests stay #[ignore]'d until PR-K3 (verifier-side change to model pure calls as uninterpreted functions f(args)).

Deferred to v1.0.0

• PR-K3: verifier-side Call modeling (unblocks the entire cross-call dedup feature).
• PR-L3: power-of-2 mul/div → shift rules.
• PR-Q: land a real corpus (httparse, nom_numbers, etc.) so harness has more rows.

🤖 Generated with Claude Code