Data-bound chart components 8/8: full mark-kind coverage - #469

_{Comment @coderabbitai help to get the list of available commands.}

Greptile Summary

The PR expands data-bound plans to every standalone mark kind through empty structural probes and adds callable-aware plan addressing. It also adds flat Reflex factories, mixed-shape data support, and an all-kinds demo with browser-smoke coverage.

• Adds structural-probe behavior for aggregating marks without fabricating validation data.
• Adds data-bound factories and exports for box, violin, ECDF, hexbin, contour, heatmap, stairs, and triangle mesh.
• Allows mixed-length and multidimensional columns while leaving coupled-shape checks to mark binding.
• Adds callable serialization and last-write-wins plan registration, though pure-Python fingerprints still omit referenced global state.

Confidence Score: 4/5

The PR should not merge until callable digests account for behavior-affecting global state or reject reducers whose behavior is not fully addressable.

A supported module-level reducer can read mutable or worker-specific global state that is absent from its fingerprint, allowing one plan digest to execute different aggregation behavior across reloads or workers.

Files Needing Attention: python/reflex_xy/plan.py

_{Reviews (2): Last reviewed commit: "fix(reflex): zero-row structural plan pr..." | Re-trigger Greptile}

Merging this PR will not alter performance

✅ 109 untouched benchmarks
⏩ 2 skipped benchmarks¹

_{Comparing stack/8-full-kind-coverage (ce04ec4) with stack/7-demos-and-gallery (7c555e4)}

Review addressed in b0437af + 362394c — both hard blocks are redesigned, not patched:

Synthetic validation is gone entirely. The fix is the core structural-validation path the review asked for: xy.structural_probe() (new in python/xy, spec'd in spec/api/chart-kind-contract.md "Structural probe"). Under it, a mark whose channels are all empty validates its configuration — enums, bounds, colormaps, range/level shapes — and skips aggregation; non-empty data behaves identically in and out of the mode, and outside it the aggregating validators keep their at-least-one-value contracts. build_plan compiles every kind zero-row; _SYNTHETIC_CHANNELS and the shape table are deleted. Your repros are pinned as tests: the stairs+scatter shared-edges composition builds a plan and binds real mixed-length data (test_plan.py::test_shared_columns_between_aggregating_and_zero_row_marks_probe), and hexbin range=((0,.5),(0,.5))/mincnt/gridsize=2048 probe without failing or allocating (::test_hexbin_value_dependent_configs_probe_without_aggregating). Config errors still fail reflex run for every aggregating kind (tests/test_validation_timing.py, three-way pins: compiles empty under probe / still refuses empty normally / still raises config errors under probe).

Callables are content-addressed, not name-addressed. Bound methods are refused outright (your two-reducers repro is pinned: test_bound_methods_are_refused_as_plan_callables). Pure-Python functions serialize as import path plus a code fingerprint (bytecode, names, nested code objects, defaults; frozenset consts sorted for cross-process determinism) — editing a reducer's body changes the digest, so a rolling deployment resyncs instead of executing two behaviors behind one address (test_callable_digest_follows_the_body_not_only_the_name builds two same-qualname bodies and asserts distinct digests). C-level callables (ufuncs/builtins) must resolve back to the same object by qualified name and carry their distribution's version. Closures and partials are refused.

Also: the @data docstring drops the stale equal-length claim, /kinds now carries browser render coverage (smoke step 7, all 26 cells pixel-probed; the full smoke passes against the running app), and the plan-tier benchmark from 5/8 covers the page-time/startup/republish costs. Specs revised across reflex-integration.md §3.6, chart-kind-contract.md, and the implementation/options decision records — the completion record now describes exactly what is implemented.

Rebased onto the current stack/7-demos-and-gallery (ce04ec4). The five commits replayed without conflicts and their content is unchanged — I diffed the pre- and post-rebase ranges and the only differences are hunk offsets.

The rebase did surface one real failure, so there is one new commit on top that was not in the reviewed history:

test_public_surface.py::test_every_public_export_is_statically_typed was added by af9f99c on stack/5-chart-factories, after this branch was originally cut. It requires every name in reflex_xy.__all__ to carry a static declaration next to its __getattr__ route, so consumers see real signatures instead of Any. This layer's eight new flat factories (box_chart, contour_chart, ecdf_chart, heatmap_chart, hexbin_chart, stairs_chart, triangle_mesh_chart, violin_chart) and the newly curated triangle_mesh node re-export went into _EXPORTS/_XY_REEXPORTS only, so they typed as missing. ce04ec4 restates them in the TYPE_CHECKING block — declarations only, no runtime change.

Worth noting because it means the old green CI here was green against a lineage that did not yet contain that pin, not because the surface was complete.

Local verification on the rebased branch: tests/reflex_adapter tests/test_validation_timing.py tests/test_example_apps.py tests/test_check_typing.py — 312 passed, 1 skipped (311 + 1 failure before the fix). pre-commit run --all-files, ruff check, ruff format --check clean; ty check shows the same 7 pre-existing pyplot diagnostics that main has.

Generated by Claude Code