WASM: contribute a quantecon recipe to emscripten-forge

[!NOTE]
Updated 2026-08-21. The original Goal treated this recipe as what makes quantecon installable in JupyterLite; that is already true without it — quantecon is noarch: python on conda-forge, JupyterLite deployments resolve conda-forge alongside emscripten-forge-4x by default, the kernel environment merged in PR #938 (ci/wasm/environment.yml) installs it that way, and import quantecon has run in the browser (#927, emscripten-forge/recipes#6309). The body below reframes the recipe's value as a trimmed browser payload plus channel co-location with numba, makes the acceptance criteria recipe-specific, and updates the sequencing (the gini_coefficient gate became the O(n log n) rewrite in PR #937, and out-of-the-box use also waits on two upstream Numba fixes). The original text is preserved in the edit history.

Part of #925 (Phase 2 — distribution).

Goal

Contribute recipes/recipes_emscripten/quantecon/recipe.yaml to emscripten-forge/recipes so that JupyterLite deployments get a WASM-trimmed quantecon build from emscripten-forge-4x instead of the conda-forge noarch package.

Installability itself is not the motivation. quantecon 0.11.4 ships on conda-forge as noarch/quantecon-0.11.4-pyhd8ed1ab_0.conda (267 KB); emscripten-forge installs all noarch packages from conda-forge, and the default channel pair (emscripten-forge-4x, then conda-forge) is consulted both at deployment build time and by the %mamba install magic (mambajs). The Phase 0 environment now on main (ci/wasm/environment.yml, PR #938) lists quantecon directly against that pair, and import quantecon has been run in the xeus-python kernel (#927 comments; the emscripten-forge/recipes#6309 traceback). %mamba install quantecon has not yet been exercised in a browser, but it solves over the same channels and is expected to resolve the conda-forge build too. No quantecon recipe exists in emscripten-forge/recipes today.

What the recipe adds:

• A smaller browser payload. Roughly a third of the installed package (about 400 KB of ~1.2 MB) is the 40 files under the subpackage tests/ directories (32 test_*.py, 3 .gam, 5 __init__.py). pyproject.toml [tool.flit.sdist] excludes only quantecon/tests/ and quantecon/util/tests, so the subpackage tests ride along in the sdist and into every downstream build. A recipe that excludes **/tests/** and sets skip_pyc_compilation — exactly what the sympy recipe does, and sympy is likewise noarch-only on conda-forge — removes that.
• Channel co-location. emscripten-forge-4x is listed before conda-forge by default, so an emscripten-forge build shadows the conda-forge package automatically and sits next to the emscripten-forge numba build it runs on, giving one place for any WASM-specific pins.

An alternative for the payload half is to trim at source — exclude the subpackage tests/ directories from the sdist in pyproject.toml — which shrinks the conda-forge noarch build without any emscripten-forge recipe. If payload size turns out not to matter, this issue can be deprioritised.

Approach

Model on the existing pure-Python recipes: pip-install from a source tarball (the sympy recipe uses the GitHub tag archive; pytensor-base and numba use the PyPI sdist — either works for quantecon), exclude tests from the payload, and set skip_pyc_compilation (payload size is a first-class concern in the browser; sympy and numba both do this). Run-dependencies: python, numba, numpy, scipy, sympy, requests — if #931 demotes requests to an optional extra, it drops out here too. The pytensor-base recipe is the precedent for a Numba-downstream package; quantecon's floor (numba>=0.56.0 on main, >=0.49.0 in the released 0.11.4) is far below the emscripten-forge numba (0.67.0), so no pin gymnastics are needed.

Recipes carry a tests: block: sympy's is an echo "FIXME:" placeholder pending emscripten-forge/recipes#4364, while pytensor-base uses pytester with a test_import_pytensor_base.py, so the quantecon recipe will need a test_import_quantecon.py. The merged ci/wasm/environment.yml is the ready-made environment to test the recipe against.

The conda recipe is the right primary vehicle: jupyterlite-xeus's experimental pip: section does not resolve dependencies, so the (already pure) PyPI wheel is only a fallback path. Deployments already resolve conda-forge (where quantecon is noarch) alongside emscripten-forge-4x, so the recipe's advantage is the trimmed payload and co-location, not resolvability.

Sequencing

Best contributed after the Phase 1 library fixes (at minimum the gini_coefficient O(n log n) rewrite, #926 / PR #937 — approved, awaiting merge) ship in a release after 0.11.4, so the first browser-installable quantecon works out of the box. Two further Phase 1 blockers are upstream rather than library-side and are addressed by emscripten-forge numba rebuilds, not by a quantecon release: #944 (a cache-miss cache=True caller linking a cache-restored callee fails with "no compiled object yet", so simplex_grid breaks on a warm cache; tracked as emscripten-forge/recipes#6309, decision 2026-08-21: wait for upstream) and #927 (jitted generators hang, so support_enumeration and vertex_enumeration never return; no upstream report yet). "Works out of the box" will lag until those land. Not hard-blocked on any of this — the recipe can track the current release and be version-bumped.

Acceptance criteria

• Recipe merged in emscripten-forge/recipes, with a pytester test_import_quantecon.py
• In a fresh JupyterLite xeus-python deployment on the default channel pair, quantecon resolves from emscripten-forge-4x rather than conda-forge noarch — both via the kernel environment.yml and via %mamba install quantecon — and import quantecon works there
• The installed payload omits the subpackage tests/ directories and skips pyc compilation; size recorded against the 267 KB conda-forge artifact

Related

• #928 / PR #938 (merged 2026-08-21) — ci/wasm/environment.yml installs the conda-forge noarch package on the default channel pair and seeds the recipe test; the harness's test_import_quantecon checks whichever build resolves
• #933 — the CI job tests the released conda-forge quantecon until the package lands on emscripten-forge-4x; switch it over once this lands
• #944 / emscripten-forge/recipes#6309 — cache=True chain failure (simplex_grid on a warm cache); blocks out-of-the-box use, not installation
• #927 — jitted-generator hang (support_enumeration, vertex_enumeration); likewise an upstream blocker for out-of-the-box use
• #926 / PR #937 — the library-side Phase 1 fix the first release needs
• #931 — requests as an optional extra changes the recipe's run-dependencies
• #934 — a live %mamba install quantecon in the demo notebook pulls the conda-forge noarch build until this lands

Start with ci/wasm/environment.yml and the existing sympy and pytensor-base recipes in emscripten-forge/recipes, then inspect the quantecon source layout and dependencies. The work is complete when recipes/recipes_emscripten/quantecon/recipe.yaml includes the trimmed payload and pytester import test, resolves from emscripten-forge-4x in a fresh JupyterLite deployment, and omits subpackage tests while recording the installed size.