WASM: JupyterLite environment config and browser smoke suite

[kp992]

Part of #928 and #925 Phase 0.

Adds under ci/wasm/:

• environment.yml — xeus-python kernel config for a minimal JupyterLite deployment (emscripten-forge + conda-forge channels). Seed for the CI job in WASM: add a WebAssembly/JupyterLite smoke-test job to CI #933.
• smoke_test.py — pytest suite covering every Numba feature class in the library: lazy @njit, NRT-allocated arrays, parallel guvectorize (cpu fallback), eager guvectorize, nelder_mead/brent_max/newton, lemke_howson, vertex_enumeration (numba.typed.Dict), bare np.linalg.solve in @njit (isolates the _LAPACK mechanism from WASM: verify _numba_linalg_solve (numba.np.linalg._LAPACK) in the browser; add fallback solver if needed #927), support_enumeration (end-to-end WASM: verify _numba_linalg_solve (numba.np.linalg._LAPACK) in the browser; add fallback solver if needed #927 test), gini_coefficient (marked xfail on Emscripten per WASM: remove @njit(parallel=True) from gini_coefficient via an O(n log n) rewrite #926), simplex_grid (feeds WASM: audit 32-bit intp behaviour on wasm32 (overflow guards, simplex_index wrapping, dtypes) #929), and the objmode shim in searchsorted.
• test_jupyterlite.py — Playwright harness that drives the suite's key items inside a locally-built JupyterLite site. Manual for now; the CI job that builds the site and runs it lands with WASM: add a WebAssembly/JupyterLite smoke-test job to CI #933.

Plus a native CI gate (ci_wasm_smoke.yml) that runs the suite against this repo's source on every PR, and testpaths = quantecon in pytest.ini so bare pytest keeps collecting only the package tests.

All 16 tests pass natively (Python 3.11, Numba 0.61).

---

Maintainer edit (2026-08-19): scoped per review — the Emscripten/JupyterLite runner job moved out to a follow-up PR under #933 (it exercises the released conda-forge package rather than this PR's source, and needs its own red/green iteration; the salvaged job with review fixes applied is parked in #933). This PR now references #928 as Part-of, since the results table there can only be filled after the first successful WASM run.

[coveralls]

coverage: 90.78%. remained the same — kp992:wasm-phase0-smoke-suite into QuantEcon:main

[mmcky]

Nice scoping on the smoke suite — one test per Numba feature class, matching #928's checklist exactly, and the native job mirrors ci.yml's conventions cleanly. All 16 tests passing natively is confirmed by the green native check.

Requesting changes on the CI wiring, which is currently red in three independent ways (details inline):

1. ci_np2.yml runs bare pytest from the repo root and now fails collecting ci/wasm/test_jupyterlite.py — needs testpaths = quantecon in pytest.ini.
2. The Emscripten job hung at the Playwright browser install for the full 6-hour job limit — drop --with-deps and add timeout-minutes: to both jobs.
3. --timeout=660 needs pytest-timeout, which isn't installed.
4. The wasm job lacks needs: native, so the fast gate doesn't actually gate.

Also worth noting: since quantecon isn't packaged on emscripten-forge-4x, the wasm job installs the released conda-forge noarch package — it tests the stack, not this repo's source. That's exactly what Phase 0 needs, but it argues for schedule/workflow_dispatch triggers rather than every PR, and it means the Playwright helpers in test_jupyterlite.py have never actually executed — I've flagged a few likely flakes inline.

On scope: the workflow itself is really #933's deliverable, and #928's remaining deliverable is the results table in the issue — which can't be filled in until the wasm job completes once. Two options: split the wasm job into a follow-up PR under #933 and merge the environment + suite + native gate now, or keep it together and switch "Closes #928" to "Part of #928" until the results are recorded. Happy either way.

[mmcky]

Thanks @kp992 — really nice scoping on the suite itself: every API call checks out and the native run is green, so all the changes below are CI wiring, not test content. Per the review I've pushed the fixes directly to your branch (hope that's OK!):

• a18c110 — testpaths = quantecon in pytest.ini, which un-breaks the numba-rc workflow's bare pytest (it was dying collecting test_jupyterlite.py without playwright). Bare pytest collects exactly what it did before this PR; pytest ci/wasm/smoke_test.py still works.
• 3baf546 — the Emscripten/JupyterLite job is split out to a follow-up PR under WASM: add a WebAssembly/JupyterLite smoke-test job to CI #933: it tests the released conda-forge package rather than this PR's source, and its first run hung at the Playwright browser install for the full 6-hour limit, so it needs its own red/green iteration without holding this PR hostage. The job — with the review fixes folded in — is parked in WASM: add a WebAssembly/JupyterLite smoke-test job to CI #933 ready to pick up. The native gate stays on every PR and now has timeout-minutes: 30.
• 564dc8d — test_jupyterlite.py stays here, reworked to be sentinel-driven: each cell is wrapped in try/finally printing a unique token and _run waits for that token in the output area. This removes the kernel-status selector, the stale-idle race, and the traceback-prone substring assertions in one stroke, and switches to insert_text so multi-line cells aren't mangled by the console's run-on-Enter binding.
• 1e0c598 — dropped the unused pytest from the kernel env and softened the smoke_test docstring (in-kernel pytest execution is WASM: add a WebAssembly/JupyterLite smoke-test job to CI #933 wiring).

PR body updated to "Part of #928" since the results table there can only be filled after the first successful WASM run.

[kp992]

Thanks @mmcky for pushing the fixes. This looks a good start to me. Please feel free to update or merge this. I will move to the next item in the TODO list.

[mmcky]

Thanks @kp992 for the quick turnaround! Nothing further to apply on top of a18c110…1e0c598 — CI is fully green (the numba-rc tests job is back to passing, and the native smoke gate runs in about a minute), so this is ready to merge once I refresh my review.

Follow-ups queued after merge: the #933 PR seeds from the runner job parked there, and the first green WASM run supplies the results table for #928. Enjoy the next TODO item!

[mmcky]

@oyamad are you happy for us to work through the WASM work package and make staged updates. I have documented the plan in #941 re: interaction with releases. I have gated checks to make sure any merges don't hold up any release milestones.

[oyamad]

@mmcky Yes; now the blocker is not LAPACK, but the jitted generator issue (#927 (comment)).