Thank you for the refresh. I ran a final reproducibility check after MobilityDB/MEOS-API#15 landed, and the committed IDL doesn't match what regenerates from the current canonical state. Could you regenerate against master once more before we merge?

Pulled MEOS-API master (now with #15 merged), ran python run.py against a fresh setup.py clone of MobilityDB master, and diffed output/meos-idl.json against codegen/res/meos-idl.json from this branch:

Esteban's IDL: 2830 functions, 47 structs, 6 enums
Fresh IDL:     2672 functions, 47 structs, 6 enums

In Esteban but not Fresh: 160
In Fresh but not Esteban: 2
Common: 2670
Signature drift in common funcs: 512

tsc --noEmit still passes on this branch (I re-confirmed), because the drifts are in functions no core/types/**/*.ts wrapper currently consumes, but the IDL is the binding-generation source of truth, so the next wrapper that touches one of these will get an incorrect signature.

3 causes:

1. Older typerecover.py (the largest of the three)

Timestamps:

• 23722ba (this PR's refresh commit): 2026-05-22T21:30:29Z
• 9ef867a (the typerecover.py commit that landed in MEOS-API#15): 2026-05-24T19:18:42Z

The refresh was generated ~46h before _TYPE_MAP reached its final form. The merged version recovers GSERIALIZED, Datum, DateADT, GBOX, BOX3D, AFFINE; the IDL in this PR was generated before those entries existed. Concrete examples:

512 functions are affected: every GSERIALIZED * / Datum / DateADT / GBOX * / BOX3D * slot in scope is still spelled int/int * here.

2. MobilityDB headers were not master

The IDL contains 160 functions that don't exist in MobilityDB master:

• 60 tbigint_*
• 63 th3index_*
• 121 tcbuffer_* (vs ~70 on master)
• 54 tpose_*

These look like a feature branch with the extended type families. Could you confirm which MobilityDB commit/branch was used? If it's a not-yet-merged feature branch, the binding generator's source of truth would point at code that doesn't exist upstream, I'd prefer the canonical MEOS.js to track a state of MobilityDB that's already public.

3. Two functions present on master but missing here

mindistance_tgeo_tgeo and tgeoarr_tgeoarr_mindist — confirms the headers snapshot used was slightly behind current master.

What I'm asking for

1. Rebase on current main (now with Initial commit #1 merged) and regenerate codegen/res/meos-idl.json via python run.py against:
   • MEOS-API at current master (after the #15 merge)
   • MobilityDB headers at a stable state: either master, or a specific feature branch you can document in the PR body
2. Re-run npm run generate against the new IDL, commit the updated bindings.c + functions.generated.ts
3. Force-push the amended 23722ba

The reproducibility check I ran is:

# in a fresh MEOS-API checkout, master at or after the #15 merge
python setup.py                         # or --branch <feature> if you used a branch
python run.py
diff <(jq -S . output/meos-idl.json) \
     <(jq -S . path/to/MEOS.js/codegen/res/meos-idl.json)

Once that diff is empty, I'll re-verify and approve.

Done — rebased onto main (now with #1) and force-pushed bbf7831. Regenerated against MEOS-API master (post-#15) and MobilityDB master 2c4243a, and your reproducibility diff is now empty:

$ diff <(jq -S . output/meos-idl.json) <(jq -S . codegen/res/meos-idl.json)
$       # (empty)
Done: 2672 functions, 47 structs, 6 enums

Point by point:

1. typerecover (512 drifts) — fixed at the root by regenerating with #15. The GSERIALIZED / Datum / DateADT / GBOX / BOX3D / AFFINE slots are recovered; your cited examples now read const GSERIALIZED *, Datum, DateADT.

2. Source state — I took your preferred option and pointed the regen at public master, so the 160 extended-type functions (tbigint / th3index / extended tcbuffer / tpose) are gone. They sit on a not-yet-merged feature branch; I'd rather refresh them in once they land upstream than have the canonical binding track non-public code.

3. Missing functions — mindistance_tgeo_tgeo and tgeoarr_tgeoarr_mindist are present.

One thing your Datum examples surfaced: temporal_start_value (and 170 others) now correctly type as Datum — but Datum is internal (meos_internal.h) and can't be marshalled to a JS scalar (it's a pointer for Float8/text/geometry, and only 32-bit under WASM32). So the generator now skips every Datum-signature function, and the user-facing path stays the typed wrappers (tint_start_value / tfloat_start_value / …). functions.generated.ts carries no Datum in any signature.

I also bumped the Dockerfile MOBILITYDB_COMMIT pin to 2c4243a and tracked the upstream mult_* → mul_* rename in the TFloat/TInt/TNumber wrappers (the only core/types change).

npm run generate → 2465 generated / 172 skipped / 35 manual; npx tsc --noEmit clean.

Pushed a refinement to 395ffa5 (force-push) — heads-up on what changed since the last review:

The generated binding now mirrors the PUBLIC MEOS API only. The generator excludes, in addition to Datum-signature functions:

• the internal headers (meos_internal.h / meos_internal_geo.h) — not user-facing, unused by core/types;
• the bundled liblwgeom glue header (postgis_ext_defs.in.h) — PostGIS internals.

The C preamble now also #includes the public extended-type headers (meos_cbuffer/npoint/pose/rgeo.h) so those wrappers compile. npm run generate → 2124 generated / 35 manual; tsc --noEmit clean. The committed meos-idl.json is unchanged, so the reproducibility diff against MobilityDB master is still empty — this refinement is generator-side only.

On the WASM build: I verified make release end-to-end — meos.js + meos.wasm build and link cleanly with the extended-type modules (the Dockerfile now enables CBUFFER/NPOINT/POSE/RGEO and adds the json-c/gsl include paths). It does require one upstream fix: five functions are declared in meos_rgeo.h on master but never implemented (trgeo_points, trgeo_rotation, trgeo_segments, trgeo_traversed_area, nad_stbox_trgeo), which I've fixed in MobilityDB/MobilityDB#1124. Once that lands on master, the pinned-master WASM build links with no further changes here.

Re-verified after the rebase and the v3 force-push (395ffa5). Everything checks out, approving.

Reproducibility check (the key one this round), synced my MobilityDB clone to your pin 2c4243a (after the mult_* -> mul_* rename in 4b70d8a), regenerated, and:

Bit-for-bit match. The first round's drift was entirely explained by your IDL being generated against a typerecover state that hadn't yet landed in #15 plus an unpinned MobilityDB; this round, with everything pinned and aligned, the chain is fully reproducible from canonical inputs.

Other checks:

• npx tsc --noEmit -> 0 errors
• No Datum in any TypeScript signature, only in comments of the manual overrides (tstzspan_lower, span_eq, adjacent_span_span, tstzspanset_lower, the // SKIP trgeo_restrict_value line). The "no Datum in any signature" claim holds.
• 2150 exported wrappers in functions.generated.ts, consistent with 2124 generated + 35 manual

LGTM, merging.