Upstream port candidates — week of 2026-06-08

[richardwooding]

Summary

• Since-SHA: ef505ab (cel2sql Go PR #117, Apache Spark dialect — 2026-04-27), the most recent upstream commit ported in pycel2sql PR Add Apache Spark dialect, observe-fork options, and BigQuery COALESCE fix #8.
• Upstream cel2sql Go scan window: 2026-04-27 → 2026-06-08 (11 commits, all chores — no feat/fix commits).
• cel2sql4j cross-reference window: same period (1 port candidate found).

---

Port candidates

1835215 — fix: pass elem writer to JSON array membership dialect methods (cel2sql4j PR #20)

Source: cel2sql4j (Java cross-reference port), 2026-05-23. No corresponding upstream Go commit yet — the Go implementation has the same latent bug but hasn't addressed it.

Diff scope: Widened the writeJSONArrayMembership and writeNestedJSONArrayMembership dialect method signatures to accept a writeElem writer alongside the existing writeArray writer. The caller (Converter.visitInJSONArray) no longer writes elem = inline before delegating to the dialect; each dialect now owns the full predicate.

Why it matters — per dialect before vs after:

Dialect	Old (broken) output	Correct output
PostgreSQL	elem = ANY(ARRAY(SELECT jsonFunc(arr)))	unchanged semantics, elem moves inside
BigQuery	elem = UNNEST(JSON_VALUE_ARRAY(arr)) ❌ invalid	elem IN UNNEST(JSON_VALUE_ARRAY(arr)) ✅
DuckDB	elem = (SELECT value FROM json_each(arr)) ❌ scalar subquery returns last row	EXISTS (SELECT 1 FROM json_each(arr) WHERE value = elem) ✅
SQLite	same as DuckDB	same fix
MySQL	elem = JSON_CONTAINS(arr, CAST(? AS JSON)) ❌ compares to 0/1	JSON_OVERLAPS(JSON_ARRAY(elem), arr) ✅
Spark	throws UnsupportedDialectFeatureError ❌	array_contains(from_json(arr, 'ARRAY<STRING>'), elem) ✅

pycel2sql current state (grep probes):

$ grep -nA4 "write_json_array_membership\|write_nested_json_array_membership" \
    src/pycel2sql/dialect/_base.py
131:    def write_json_array_membership(
132:        self, w: StringIO, json_func: str, write_expr: WriteFunc   # ← old single-writer
133:    ) -> None: ...
135:    def write_nested_json_array_membership(
136:        self, w: StringIO, write_expr: WriteFunc                   # ← old single-writer
137:    ) -> None: ...

All six dialect files have the old single-writer signatures. In addition, _converter.py::_visit_in never calls these methods — write_json_array_membership / write_nested_json_array_membership are dead code; the converter always falls through to write_array_membership regardless of whether the RHS is a JSON array field. This means:

1. The method signatures need widening (add write_elem: WriteFunc parameter to both, remove json_func: str from the ABC now that each dialect owns the full predicate).
2. _visit_in needs to detect JSON array RHS (using _is_json_variable_root / _is_field_json / _is_nested_json_field already present in the converter) and route to write_json_array_membership / write_nested_json_array_membership instead of plain write_array_membership.
3. The test_spark.py::test_json_array_membership_dialect_method_raises test asserts the old throwing behaviour and will need to be replaced with a positive assertion (mirroring the cel2sql4j PR Upstream port candidates — week of 2026-06-01 #20 test update).

cel2sql4j cross-reference: cel2sql4j PR #20 diff is the closest template — the Java SqlWriter functional interface maps directly to pycel2sql's WriteFunc = Callable[[], None].

Portability: Medium. The Dialect ABC signature change + 6 dialect implementations + converter routing + test updates. Each dialect change is small and mechanical; the converter routing is the interesting piece (it already has the JSON-field detection helpers).

---

Already done

• cel2sql4j 5e8eaa6 — fix missed Spark writeFormat impl (cel2sql4j PR Add add-sql-dialect agent skill #11). pycel2sql already has SparkDialect.write_format emitting format_string(...) — added in pycel2sql PR Add Apache Spark dialect, observe-fork options, and BigQuery COALESCE fix #8. Grep: src/pycel2sql/dialect/spark.py:521.

---

Filtered (upstream Go chore commits — no feat/fix changes to port)

All 11 upstream cel2sql Go commits since ef505ab are chores:

• 65f4a6c chore: prepare CHANGELOG for v3.8.2
• 53b5faf chore(deps): bump google.golang.org/api 0.276→0.277
• 79d81ed chore(deps): bump go-sql-driver/mysql 1.9.3→1.10.0
• 651bc04 chore: prepare CHANGELOG for v3.8.1
• df88b1e chore(deps): bump lib/pq 1.11.2→1.12.3
• b0d57a3 chore(deps): bump cloud.google.com/go/bigquery 1.74→1.76
• fac750a chore(deps): bump modernc.org/sqlite 1.46.1→1.50.0
• fd4f0f8 chore(deps): bump github.com/google/cel-go 0.27→0.28
• c10a520 chore: prepare CHANGELOG for v3.8.0
• 93843bf chore: add three development skills (add-sql-dialect, add-cel-feature, release-cel2sql)
• abbbfdd chore: add skill-authoring skill

None of these carry feature or behaviour changes that need porting. The cel-go bump (0.27→0.28) is a Go dependency; it does not affect pycel2sql's Python cel-python parsing layer.

---

Generated by the weekly upstream-scan routine. https://claude.ai/code/routines/weekly-upstream-port-scan