[pyramation]

Summary

CI has never built v13. It downloaded the prebuilt @libpg-query/[email protected] WASM, which has Emscripten's default 64 KiB stack and traps on SELECT 1+1+… with 1000 terms. This PR makes v13 build from source like 14–18 do.

Root cause. PG13's src/postgres/include/storage/s_lock.h has no test-and-set (TAS) for wasm32, so the build stops at:

./src/postgres/include/storage/s_lock.h:946:2: error: PostgreSQL does not have native spinlock support on this platform.  To continue the compilation, rerun configure using --disable-spinlocks.

The old emscripten_disable_spinlocks.patch only added --disable-spinlocks and #undef HAVE_SPINLOCKS to libpg_query's configure and extract_source steps. make build never runs those steps; it uses the pg_config.h already checked in, which still has HAVE_SPINLOCKS 1.

PG14 added a generic fallback, which is why 14–18 build fine: if nothing else defined HAS_TEST_AND_SET, it uses __sync_lock_test_and_set() when HAVE_GCC__SYNC_INT32_TAS / HAVE_GCC__SYNC_CHAR_TAS is set. PG13's pg_config.h already defines both of those. This PR's patch backports that block into PG13's s_lock.h, word for word from PG14.

Changes

• versions/13/patches/emscripten_disable_spinlocks.patch: the content is replaced with the s_lock.h backport. The filename stays the same, so the template and Makefile need no changes and copy:templates produces no diff.
• .github/workflows/ci.yml: the v13 tarball download is removed; every version now runs pnpm run build.

Verified locally (Linux, Docker, cd versions/13 && pnpm run build from an empty .cache, about 35–40 s):

emsdk	wasm sha256	pnpm test
6.0.13 (untagged emscripten/emsdk, what main uses)	c568fa4c…d67a	41/41 pass
6.0.3 (#174's pin)	c81f444c…8708	41/41 pass, two clean builds gave identical bytes

• On 10 sample statements the parse output is identical to the published [email protected] (DDL, CTE, window, plpgsql function, GRANT, COPY…).
• Left-deep 1+1+… chains: 1000 and 5000 terms parse; 20000 terms gives a clean RangeError (published 13.7.3 behaves the same). The CI prebuilt 13.5.7 traps at 1000 (table index is out of bounds).

Conflicts with #173, which edits the same v13 CI step. Whichever merges second just deletes the if v13 branch.

Link to Devin session: https://app.devin.ai/sessions/46dadacb8a2045eab1b68b35e90ad8de
Open in Devin Desktop: https://app.devin.ai/desktop/session/46dadacb8a2045eab1b68b35e90ad8de?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 issues found — approved ✅.

---

This PR makes PostgreSQL 13 buildable under Emscripten by backporting the upstream __sync_lock_test_and_set-based TAS spinlock fallback into versions/13/patches/emscripten_disable_spinlocks.patch, removing the #error path when --disable-spinlocks is used on wasm32. The CI workflow is updated to build all supported versions from source, dropping the prebuilt v13 WASM download and eliminating that supply-chain fetch. | Files | Change | | --- | --- | | versions/13/patches/emscripten_disable_spinlocks.patch | Backports the wasm32 spinlock fallback (__sync_lock_test_and_set / __sync_lock_release) so PG13 builds with --disable-spinlocks. | | .github/workflows/ci.yml | Builds v13 WASM from source with the new patch instead of fetching prebuilt artifacts. | No blocking issues were identified; the verified scan found no defects in either change.

_{Reviewed commit: 555e166}