fix(driver-sql): make the SQLite Field.json codec injective — one encoding across all three dialects

[os-warren]

Fixes#12380

Implements route A of the 2026-08-26 maintainer ruling: JSON.stringify on write, JSON.parse on read, unconditionally, matching what Postgres and MySQL already do. This deletes a dialect branch rather than adding one. B (declare the current behaviour the contract) and C (narrow what can be written) are not taken.

What was wrong, measured live

Measured through the driver boundary on live SQLite (better-sqlite3), live Postgres 16.13 (server TZ Asia/Shanghai) and live MySQL 8.0.46 (server TZ +08:00), process TZ America/New_York, with every stored cell read back through a separate raw catalog query (typeof() / json_typeof() / json_type()) — never the emitted DDL.

Three independent mechanisms, only two of them reversible. Each was verified against origin/main before being built on:

#	mechanism	reversible
A	Read-side — formatOutputJSON.parses every string in a json column	yes
B	Write-side — the column is declared type json, which under SQLite's affinity rules resolves to NUMERIC affinity, converting number-like strings to INTEGER/REAL before storage	⛔ no
C	Native booleans — true read back as number 1; formatOutput's booleanFields pass is keyed to declared Field.boolean columns, not to booleans inside a json payload	yes

The contract settles which dialect is right, not strictness. json's stored contract is z.unknown() because "openness is now an explicit decision, not an accident of nobody checking" (packages/spec/src/data/field-value.zod.ts). An open contract admits both 123 and '123' as legal values of one field ⇒ no driver may collapse them.

Mechanism B decided the design, and it is pinned live rather than reasoned

No DDL change was needed. The column stays declared json, so NUMERIC affinity is still in force — and the encoded form defeats it, because a string's encoding carries its quotes ('123' → "123", which is not a well-formed numeric literal).

That is not an argument in the PR body; it is a test. sql-driver-12380-json-roundtrip.test.ts binds the pre-fix value and the encoded value into the same column in the same INSERT statement, so nothing but SQLite's own affinity rule can account for the difference:

raw '0123' → integer (the affinity, caught in the act)
JSON.stringify('0123') → text "0123" (the encoding defeating it)

…for each of '123', ' 123 ', '0123', '1e5', '1.0', '-0'.

What changes on disk

For new writes the delta is exactly two classes: strings (now quoted JSON text) and booleans (now TEXT true/false instead of INTEGER 1/0). Objects, arrays, nulland numbers are byte-identical to before — 123 bound as a number and "123" bound as text both land as INTEGER 123.

⚠️An out-of-band reader of a SQLite file — anything reading the table with its own SQL rather than through this driver — now sees quoted JSON text where it saw a bare value.

The migration, and the limits it states plainly

backfillCanonicalJsonEncoding runs on syncSchema/initObjects for existing tables, in the same shape and posture as the backfillCanonicalDatetimes / backfillCanonicalTimes storage-format migrations beside it: one UPDATE per column, failures logged and swallowed, correctness never contingent on it having run.

It converts the one on-disk class the pre-fix encoding left unambiguous — a TEXT cell that is not valid JSON, which nothing but a stored plain string could have produced.

⛔ It does not guess, and the ruling's unrecoverable class is named rather than discovered:

• INTEGER/REAL cells — a number, a boolean, and a number-like string eaten by NUMERIC affinity are the same bytes. 123 the number and '123' the string are one INTEGER 123; boolean true is one INTEGER 1 alongside the number 1. Cannot be restored, only stopped from growing.
• TEXT cells that already parse — a stored object {"a":1} and a stored string'{"a":1}' were byte-identical before this change. Re-quoting would turn every legacy object and array into a string, corrupting the common case to guess at the rare one.

⇒ Those rows read after this change exactly as they read before it. The class stops growing; it is not retroactively repaired.

The migration changes no read. A legacy plain string reads back as that string both before it runs (via formatOutput's parse fallback, kept deliberately and now documented as the pre-#12380 read-side repair) and after. It is a canonicalisation that makes storage injective going forward, not a repair of something that reads wrong today.

Idempotence is by construction, and pinned. The WHERE is the exact complement of the SET's output: json_quote(X) of a TEXT value is a quoted JSON string, for which json_valid() is 1, so a converted row cannot match again. The suite runs the migration three times and asserts the disk bytes are identical after runs 2 and 3 — specifically that "America/New_York" never becomes "\"America/New_York\"".

Blast radius, read from source

sys_setting.value is Field.json and the settings service persists verbatim with no re-coercion by declared type, so the driver's answer is what the caller gets — on the dialect tenant environments actually run. driver-turso and driver-sqlite-wasm extend SqlDriver and override neither half of the codec (checked); both inherit the fix and their suites run green against the built dist.

Filters are unaffected: every scalar comparison operator on a json column is already refused by the driver (JSON_COLUMN_INCOMPATIBLE_OPERATORS — $eq/$ne/$gt/$in/$nin/$between and their spellings), so no predicate could have been keyed to the old stored text.

Verification

Gate union derived, not recalled, at 56c22c0 — node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack (re-derived after merging origin/main, because the first run reported a STALE TREE). Every matched family run in the foreground; every exit code captured before any pipe. All green. Each gate's own verdict line:

• check-adr-0087-registration: 1 declared-breaking changeset(s), each carrying an ADR-0087 disposition. — the gate accepted not-required (no-migration-prescription). This is a declared-breaking change; the disposition is that no authorable metadata key is removed, renamed or re-shaped, so there is nothing for objectstack migrate meta to rewrite — the migration is over data rows and runs automatically.
• check-driver-conformance: OK — 45 covered cell(s), 0 in the DEBT ledger, 0 exempt. / 9 of 9 dialect-scored cell(s) have a matrix-routed suite.
• check-nul-bytes: OK (scanned 6859 text file(s) … no raw ASCII control bytes).
• check-type-check-coverage --re-measure: OK — 32 ledger entr(ies) re-measured in 247.2s, 1843 raw tsc error(s) total, none above its recorded number. (needed the full workspace closure built; it refused outright before that rather than measuring a different world)
• ✓ This diff introduces no 'major' bump. · ✓ No empty-frontmatter changeset introduced by this diff (1 declaring changeset(s) added).
• check-engine-double-contract: OK — 415 pinned · where-matcher conformance holds: 302 matcher(s) · check-test-source-alias OK · OK: 18 package(s) read outside themselves, all declared — plus check:objectui-changeset, check:page-declaration-shape, check:published-files, check:slot-lookup, check:type-source-resolution, check:query-options-erasure, check:ci-filter-parity, check:comment-mask-adoption, check:plugin-teardown-shape, both docs-audit gates and release-rehearsal-clone --self-test, all exit 0.

Suites, all foreground:

suite	result
new sql-driver-12380-json-roundtrip.test.ts	103 passed (24 cases × 3 dialects + disk/read collision + catalog + affinity + migration)
driver-sql full	145 files, 3020 passed, 1 skipped
driver-turso	1006 passed
driver-sqlite-wasm	395 passed
service-settings	519 passed
platform-objects	496 passed
driver-sql typecheck	exit 0

The four consumer suites ran against a built driver-sql/dist carrying this change. Their first run failed to resolve@objectstack/driver-sql on an unbuilt worktree — reported here because that shape reads exactly like a regression and was not one.

Ablation — direction predicted in writing first, then measured

Predicted: restoring the deleted pre-fix branch goes RED, asymmetrically — SQLite only: 14 of the 24 round-trip cases on the SQLite cell (the 12 JSON-looking/number-like strings + n_true/n_false), the controls s_empty/s_nan/s_tz/s_bad/n_str and all numbers/objects/arrays/null staying green, plus SQLite's whole-matrix, disk-collision and read-collision tests and the migration suite's fresh-table test — 18 failures, with PG and MySQL 100% green.

Measured: 18 failed | 85 passed (103), and 0 FAIL lines on the postgres or mysql cells — all 18 in (sqlite) or the SQLite-only migration suite. Prediction matched exactly, and the PG/MySQL regression control fires: those dialects were 17/17 faithful before this change and are unaffected by both the fix and its removal.

Mutation proven on disk before any result was read, by single-line anchored grep counts: ABLATION-12380-MUTANT0 → 1, ONE encoding, every dialect (#12380)1 → 0, git diff --numstat10 42. No rebuild was involved or needed — the suite imports ./sql-driver.js, resolved to source, so no dist was in the loop and none could be poisoned; the result changing at all is itself the proof that the subject is the source file. The script carried a trap … EXIT INT TERM restore; the restore is verified byte-identical (git status empty, fix marker back to 1, mutant marker 0).

Out of scope

The VALUE_ROUNDTRIP census case-set (option D) is ruled in but sequenced second as a separate card — it is cross-driver by design and would ship red on SQLite by construction if it landed before this. Not added here; this fix carries its own pin instead.

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

This PR changes 1 package(s): @objectstack/driver-sql, touching 5 documentable anchor(s).

6 hand-written doc(s) NAME something this change touched and may need an implementation-accuracy re-verification:

• content/docs/data-modeling/drivers.mdx(via SqlDriver (symbol))
• content/docs/data-modeling/index.mdx(via SqlDriver (symbol))
• content/docs/plugins/packages.mdx(via SqlDriver (symbol))
• content/docs/protocol/kernel/index.mdx(via SqlDriver (symbol))
• content/docs/protocol/kernel/lifecycle.mdx(via SqlDriver (symbol))
• content/docs/protocol/objectql/query-syntax.mdx(via SqlDriver (symbol))

⛔ 1 release-owned page(s) also name something this change touched. These are read-only:

• content/docs/releases/v17.mdx(via SqlDriver (symbol))

content/docs/releases/ is RELEASE-OWNED (AGENTS.md "Documentation Guardrails"): release
notes are written centrally at release time, and a code PR that edits them is the exact PR
that guardrail exists to stop. They are still audited — read-only. If one of them is actually
wrong, file an issue or open a dedicated docs-only PR; do not edit it here.

What this run could not see

• the SDK route bridge reached 47 of 219 client-bound route-ledger rows — the other 172 have no registrar path: tail to select them, so pages documenting THEIR client methods cannot appear above, on this or any run. Of those 172: 14 are remediable by widening that discovery convention (an in-repo file declares the path; the convention did not scan it); 56 are structural — on a ledger where NOT ONE row is declared in-repo, so no discovery change reaches them at any price; 102 are undecided (no in-repo declaration, on a ledger that has other in-repo registrars — absence and an unreadable spelling are not distinguishable here). The rows themselves: node scripts/docs-audit/affected-docs.mjs --bridge-coverage
• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.

Coarse fallback — 9 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json cdbd9204b65ae4ebec63f90f61669fb498efaa55 → packageMentionDocs.

Which tree this was computed on

This run read content/docs from 8fd0062ab7ff05fdd7e6aa8d377552ca5be564ba — the merge of head 56c22c04534c7183f2357c0ae9236f610103d029 into base cdbd9204b65ae4ebec63f90f61669fb498efaa55, which is what actions/checkout gives a pull_request run. Not the PR head.

A worktree cut from an older main holds a different content/docs, so re-deriving there can legitimately return a different list — that is a different tree, not a wrong row. To answer on the same tree:

# while this PR is open — GitHub drops the merge commit once it closes
git fetch origin 8fd0062ab7ff05fdd7e6aa8d377552ca5be564ba && git checkout 8fd0062ab7ff05fdd7e6aa8d377552ca5be564ba
# afterwards, rebuild it from the two parents, which stay fetchable
git fetch origin cdbd9204b65ae4ebec63f90f61669fb498efaa55 56c22c04534c7183f2357c0ae9236f610103d029 && git checkout -B drift-repro cdbd9204b65ae4ebec63f90f61669fb498efaa55 && git merge --no-ff 56c22c04534c7183f2357c0ae9236f610103d029
node scripts/docs-audit/affected-docs.mjs --json cdbd9204b65ae4ebec63f90f61669fb498efaa55

⚠️ That checkout carried uncommitted changes, so the commit above does not fully identify what was read.

Advisory only, and a precision-first one (#9192): a page is listed because it names a
symbol, wire route or SDK method this diff touched — not because it mentions a changed
package. Each row says which anchor put it there, so a wrong row is reportable rather than
merely annoying. To re-verify, run the docs-accuracy-audit workflow scoped to these files:
node scripts/docs-audit/affected-docs.mjs cdbd9204b65ae4ebec63f90f61669fb498efaa55 → pass the list as
args.docs, on the commit named under Which tree this was computed on.

[os-steve]

Contract review (clause-② chain): PASS at head 56c22c04534c7183f2357c0ae9236f610103d029 — verdict and evidence recorded on #12380. needs:contract-review cleared on the card; the enqueue gate is satisfied for this head. If the head moves after this note, the gate re-arms. Landing remains with the domain:engine seat's normal review.

---

Generated by Claude Code