[pyramation]

Summary

Adds a Native parity 🔍 step to the existing build-wasm matrix job. No new job. Right after pnpm run build, scripts/native-parity.js versions/<n> does the following:

1. Exports the libpg_query checkout that the WASM build already cloned (.cache/*/libpg_query/<x-publish.libpgQueryCommit>) to a temp dir with git archive, then runs a native make build. No network is needed.
2. Compiles scripts/native-parity/driver.c, which wraps pg_query_split_with_scanner and pg_query_parse.
3. Splits libpg_query's test SQL (test/sql/postgres_regress, plpgsql_regress, deparse*, test/plpgsql_samples.sql) into statements.
4. Checks that the freshly built wasm/index.cjs parseSync gives byte-identical JSON or the same error (message + cursor) for every statement. It exits 1 and prints the first 10 differences otherwise.

Native and WASM have one known difference, which the script maps instead of reporting. FETCH_ALL is LONG_MAX, so howMany/how_many is 2^63−1 natively and 2147483647 on wasm32.

On glibc ≥ 2.38 (the ubuntu-24.04 runners), PG ≤ 14's vendored snprintf.c won't compile natively (conflicting types for 'strchrnul'). The native build therefore gets CPPFLAGS=-DHAVE_STRCHRNUL= on Linux. This only affects the parity build, not the WASM.

This catches WASM build regressions (emsdk bump, wrapper or Makefile changes, fork bumps) against the C library itself, using the full PostgreSQL regression suite as input.

Evidence: this PR's CI run

version	statements	files	differences	step time
v13 @ 1097b2c	34,670	204	0	26 s
v14 @ 397cbb9	37,532	223	0	19 s
v15 @ ae10504	40,450	225	0	40 s
v16 @ fce106a	42,445	229	0	19 s
v17 @ 1ade880	46,032	402	0	20 s
v18 @ b0f5f33	49,655	411	0	13 s

Negative check (local): changing the FETCH_ALL mapping makes it report 58 differences (combocid.sql#26, hash_index.sql#54, …) and exit 1.

Parse only for now. On main, v18 fingerprint/normalize/scan still return some errors as results (fixed in #183), so they would differ from native until that lands.

Link to Devin session: https://app.devin.ai/sessions/91c28df261ac4b318f39d1c7242d008c
Open in Devin Desktop: https://app.devin.ai/desktop/session/91c28df261ac4b318f39d1c7242d008c?variant=devin
Requested by: @pyramation

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[tenki-reviewer]

Review complete. No blocking issues — approved ✅; 1 nitpick below.

🧹 Nitpicks (1) — 🟢 1 low

• 🟢 Tempdir cleanup skipped on every failure path (native-parity.js:69) — The temporary workdir work is removed only at the very end of the happy path (scripts/native-parity.js:69); none of the execFileSync calls (make, cc, driver, git archive) are wrapped in try/finally, so any failure leaks the full libpg_query checkout, build objects, and statements.bin in os.tmpdir().

---

This PR introduces a CI parity harness that builds libpg_query natively, drives it with a small C program over the JSON API, and compares its results statement-by-statement against the WASM bindings to catch divergence between the two builds.

Files	Change
.github/workflows/ci.yml	Adds a parity job/step that invokes the new script during CI runs.
scripts/native-parity.js	New orchestration script: builds natively, runs the driver, loads the WASM module, and diffs parse/split results.
scripts/native-parity/driver.c	New C driver that feeds SQL to the native library over stdin and emits JSON results for comparison.

_{Reviewed commit: 4761bd1}

[devin-ai-integration]

Folded into #190