feat(stdlib): LinearAlgebra integration + ≥95% coverage for LinearAlgebra/Statistics/Dates/Random

[Dale-Black]

Summary

Comprehensive LinearAlgebra stdlib integration for WasmTarget.jl, plus a ≥95%-coverage campaign that brings every currently-supported stdlib to a provable, differentially-grounded support level:

stdlib	% in-scope supported	supported	boundary	out-of-scope
LinearAlgebra	97%	62	2	42
Statistics	100%	13	0	0
Dates	96%	45	2	2
Random	100%	11	0	5

% = supported / (supported + boundary) — of the in-scope surface, with out-of-scope (genuine non-wasm) excluded. Every "supported" name is backed by a real differential test (catalogue compose+diff or a test/fuzz/*_diff.jl sweep), using the same bit-exact / tolerance oracle (ORACLE_RTOL=1e-9) as core. See test/fuzz/STDLIB_COVERAGE.md for the full per-function matrix.

What's in here

LinearAlgebra — matrix + factorization + structured-type surface, verified by test/fuzz/linalg_diff.jl (90+ sweeps, in CI):

• 2-D Matrix bridge support (return-side, mirroring arg-side).
• Values: det/logdet/inv/\/svdvals/eigvals/eigmax/eigmin/cond/rank/opnorm/pinv/…
• Factorization objects: lu/cholesky/eigen/svd via hand-rolled LU / cyclic-Jacobi / one-sided-Jacobi (reconstruction-verified, sign/order-invariant) — the LAPACK-backed paths don't lower to WasmGC.
• Structured types (Diagonal/Symmetric/Hermitian/Upper/LowerTriangular): construction, dense conversion, matvec.
• In-place: mul!/triu!/tril!/lmul!/rmul!/axpy!/axpby!/normalize! + (this campaign) ldiv!/rdiv!/hermitianpart!/copyto!/copytrito!/fillstored!/copy_transpose!/copy_adjoint!/rotate!/reflect!/kron!.

Statistics / Dates / Random — value-layer + in-place surfaces fully fuzzed (test/fuzz/{stats,dates,random}_diff.jl):

• Statistics: mean!/median!/quantile! (mean! via a row-means overlay).
• Dates: epoch/calendar conversions (both directions), tuple extractors, micro/nanosecond, dayname/dayabbr/monthname/monthabbr (ENGLISH-table overlays), tofirst/tolast/tonext/toprev (modular-arithmetic overlays).
• Random: seeded-Xoshiro streams + randperm!/randcycle!/shuffle!/seed!/randsubseq/randsubseq! + randstring (charset-sampler overlay).

All new bit-exact equivalences live in the weakdep exts (ext/WasmTarget{LinearAlgebra,Dates,Random,Statistics}Ext.jl) via @overlay WASM_METHOD_TABLE; each is proven semantically identical in its commit message and in test/fuzz/FINDINGS.md.

Honest-denominator discipline

out-of-scope is only genuine non-wasm surface, never a tractable item reclassified to inflate the %. Notably:

• Random rand!/randn!/randexp! (Float64-array fills) → out-of-scope: native dispatches to an 8-lane SIMD xoshiro_bulk_simd (llvmcall intrinsics) whose stream provably differs from the scalar generator at n≥8. A scalar overlay was built, measured to diverge, and removed rather than ship a wrong-value path.
• Remaining honest boundary (in-scope, deferred — not gamed away): LA / + convert; Dates format + canonicalize. Quick-win next increments noted in FINDINGS.md.

Testing

Each stdlib was committed only after a full Pkg.test() gate passed. The suite is green end-to-end; the differential testsets ("Differential fuzz: LinearAlgebra matrix / Dates value layer / Random seeded streams / Statistics in-place") run in CI and skip cleanly without Node.js.

🤖 Generated with Claude Code

[huangyxi]

Thank you for the quick response.

My point is not just to report that the README is outdated, but also to raise the underlying issue: should bindings for other Node.js libraries/functions be provided as well?

The missing Math.pow import seems like a symptom of a broader problem rather than just a documentation issue. Perhaps the LLM agent should be guided to address the root cause instead of only updating the README.

---

I'm reposting this here in case your LLM agent ignores closed issues. More importantly, I think the agent should pass this information along to the user, since simply fixing the documentation may leave them encountering the same underlying problem.