cultcache-ts: a load and an index registration name the least violation of the resulting state #103

Merged
libby merged 6 commits from hands/variants-pair-order-ts into main 2026-11-11 12:00:00 +00:00
2 changed files with 117 additions and 47 deletions

View file

@ -47,6 +47,7 @@ type EntryLookups = {
};
type HydratedEntry = CultCacheEnvelope & {
catalogEntry: NonNullable<CultCacheEnvelope["catalogEntry"]>;
value: unknown;
lookups: EntryLookups;
};
@ -316,6 +317,16 @@ export class CultCache {
}
}
// Judged whole, once: the refusal names the least violation of the state this load would leave, whatever the
// order of the stores and of the entries in them.
refuseLeastIndexSharing(function* () {
for (const entry of next.entries.values()) {
for (const [alias, value] of entry.lookups.indexes) {
yield { type: entry.type, schemaName: entry.catalogEntry.schemaName, alias, value, key: entry.key };
}
}
}());
this.#hydrated = next;
});
}
@ -652,6 +663,7 @@ export class CultCache {
async #writeNow(registered: RegisteredDefinition, entry: CultCacheEnvelope, value: unknown): Promise<void> {
const home = this.#homeStore(entry.type);
const hydrated = this.#admitEntry(this.#hydrated, registered, entry, value);
this.#refuseIndexSharing(this.#hydrated, entry.type, registered.catalogEntry.schemaName, entry.key, hydrated.lookups);
const legacy = this.#hydrated.legacyGlobals.get(entry.type);
await home?.push(entry);
if (home && legacy) {
@ -903,7 +915,6 @@ export class CultCache {
}
const lookups = this.#lookupsFor(registered.nameAccessor, registered.indexAccessors, value);
this.#refuseIndexSharing(state, entry.type, registered.catalogEntry.schemaName, entry.key, lookups);
return {
...entry,
payload: this.#cloneBytes(entry.payload),
@ -995,7 +1006,6 @@ export class CultCache {
const entryLookups: Array<[HydratedEntry, EntryLookups]> = [];
const nameMap = nameAccessor ? new Map<string, string>() : undefined;
const indexMaps = new Map<string, Map<string, string>>();
const sharing = new Map<string, string[]>();
for (const indexName of indexAccessors.keys()) {
indexMaps.set(indexName, new Map());
}
@ -1008,21 +1018,16 @@ export class CultCache {
}
for (const [indexName, indexValue] of lookups.indexes) {
const map = indexMaps.get(indexName)!;
const holder = map.get(indexValue);
if (holder === undefined) {
map.set(indexValue, entry.key);
} else {
sharing.set(`${indexName}\u0000${indexValue}`, [...(sharing.get(`${indexName}\u0000${indexValue}`) ?? [holder]), entry.key]);
}
if (!map.has(indexValue)) map.set(indexValue, entry.key);
}
}
// Judged whole: any value held by two records refuses, naming the two smallest of its keys. Held records already
// passed the judge for every alias installed before, so only the alias being registered can be shared here.
for (const [id, keys] of sharing) {
keys.sort(compareOrdinal);
throw indexSharingRefusal(keys[0], keys[1], schemaName, id.slice(0, id.indexOf("\u0000")));
}
// Judged whole, by the same judge as a load: the least violation of the state the registration would leave.
refuseLeastIndexSharing(function* () {
for (const [entry, lookups] of entryLookups) {
for (const [alias, value] of lookups.indexes) yield { type, schemaName, alias, value, key: entry.key };
}
}());
return { entryLookups, nameMap, indexMaps };
}
@ -1099,6 +1104,34 @@ function compareOrdinal(a: string, b: string): number {
return a < b ? -1 : a > b ? 1 : 0;
}
// The whole-state judge (variants:question:unique-index-refusal-pair-rule): over every (type, alias, value) held at two
// or more distinct keys, throws for the one least by (alias, least key, second-least key, schema name), each compared by
// UTF-16 code unit. A function of the state, never of the order the rows arrive in.
function refuseLeastIndexSharing(rows: Iterable<{ type: string; schemaName: string; alias: string; value: string; key: string }>): void {
const groups = new Map<string, { schemaName: string; alias: string; keys: string[] }>();
for (const { type, schemaName, alias, value, key } of rows) {
const id = JSON.stringify([type, alias, value]);
const group = groups.get(id);
if (group) group.keys.push(key);
else groups.set(id, { schemaName, alias, keys: [key] });
}
let least: { schemaName: string; alias: string; first: string; second: string } | undefined;
for (const { schemaName, alias, keys } of groups.values()) {
if (keys.length < 2) continue;
keys.sort(compareOrdinal);
const candidate = { schemaName, alias, first: keys[0], second: keys[1] };
if (!least || compareLeast(candidate, least) < 0) least = candidate;
}
if (least) throw indexSharingRefusal(least.first, least.second, least.schemaName, least.alias);
}
function compareLeast(
a: { schemaName: string; alias: string; first: string; second: string },
b: { schemaName: string; alias: string; first: string; second: string },
): number {
return compareOrdinal(a.alias, b.alias) || compareOrdinal(a.first, b.first) || compareOrdinal(a.second, b.second) || compareOrdinal(a.schemaName, b.schemaName);
}
function indexSharingRefusal(keyA: string, keyB: string, schemaName: string, alias: string): Error {
const [first, second] = compareOrdinal(keyA, keyB) <= 0 ? [keyA, keyB] : [keyB, keyA];
return new Error(`Records ${first} and ${second} (${schemaName}) both hold unique index '${alias}'; an index value names one record.`);

View file

@ -119,14 +119,6 @@ test("a removed or changed holder releases its value", async () => {
assert.equal(cache.getKeyByIndex(document, "sku", "T"), "a");
});
test("the keys in a refusal are in ordinal order whichever one lands second", async () => {
const document = item("item");
const cache = memory(document);
await cache.put(document, "m", { sku: "S" });
await assert.rejects(cache.put(document, "z", { sku: "S" }), { message: refusal("m", "z", "tests.item", "sku") });
await assert.rejects(cache.put(document, "B", { sku: "S" }), { message: refusal("B", "m", "tests.item", "sku") });
});
// The C# White_Space set, and nothing else, is blank. Every UTF-16 code unit is judged, so a rule written for a few
// characters (or for JavaScript's \s, which counts U+FEFF and omits U+0085) cannot pass.
const CSHARP_WHITE_SPACE = (code: number) =>
@ -365,6 +357,76 @@ test("vectors duplicate-store.cc and blank-store.cc hold their verdicts", async
}
});
// refusals.msgpack: rows [label, ops, refusal] written by the C# reference. TypeScript has no batch commit
// (variants:idea:unique-index-runtimes.ts-batch-commit), so it performs every row whose commits each hold one record:
// 5, 8-13, 19-22, 24, 25, 27, 29, 31, 32 and 33 of the README (18).
const REFUSAL_SCHEMAS: [string, string, string, Record<string, "sku" | "second">][] = [
["unique_item", "tests.unique_item", "sha256:93bb890de83069061af696b23f94fd32fb3c2dad3c91ee645e8add56085f1882", { Sku: "sku" }],
["uq_pair", "tests.uq_pair", "sha256:2538cceafb6dcbce22ef11b932ecb804af5f96bc3aaedff692a52d3640433e60", { Sku: "sku", Alt: "second" }],
["uq_alias", "tests.uq_alias", "sha256:77a77a394ebd8bf71e14b9b6f2cea4978c25d4e705fd12364f660edbf6c63b1f", { Sku: "sku", alt: "second" }],
["uq_alpha", "tests.uq_alpha", "sha256:53aeaf2ea9b25e85ade6b15fc04c7653e7b206c87e258f1aa74dd32a840a1389", { Sku: "sku" }],
["uq_beta", "tests.uq_beta", "sha256:45ed62a64c91f6d815d3ec975c4a4a1b2ce07b6e7262a32d828673cf10899586", { Sku: "sku" }],
];
type RefusalOp = ["open", string | null] | ["pull", string] | ["commit", [string, string, string, string][]];
test("vectors refusals.msgpack: every row TS can perform", async () => {
const rows = decode(await readFile(join(vectors, "refusals.msgpack"))) as [string, RefusalOp[], [string, string, string, string] | null][];
const definitions = new Map(
REFUSAL_SCHEMAS.map(([type, schemaName, schemaId, indexes]) => [
schemaName,
defineDocumentType({
type,
schemaId,
schemaName,
schemaVersion: `${schemaName}.v1`,
schema: z.object({ sku: z.string(), second: z.string() }),
codec: codecs.slots<{ sku: string; second: string }>({ 0: ["sku", codecs.string], 1: ["second", codecs.string] }, `Uq${type}`),
indexes,
}),
]),
);
const dir = await mkdtemp(join(tmpdir(), "uidx-refusals-"));
let walked = 0;
try {
for (const [label, ops, refused] of rows) {
if (ops.some((op) => op[0] === "commit" && op[1].length > 1)) continue;
const file = join(dir, `${walked}.cc`);
let cache: CultCache | undefined;
const perform = async (op: RefusalOp) => {
if (op[0] === "open") {
if (op[1] !== null) await copyFile(join(vectors, op[1]), file);
let builder = CultCache.builder().withGenericStore(new SingleFileMessagePackBackingStore(file));
for (const definition of definitions.values()) builder = builder.withDocumentType(definition);
cache = builder.build();
await cache.pullAllBackingStores();
} else if (op[0] === "pull") {
await copyFile(join(vectors, op[1]), file);
await cache!.pullAllBackingStores();
} else {
for (const [key, schema, sku, second] of op[1]) await cache!.put(definitions.get(schema)!, key, { sku, second });
}
};
for (const op of ops.slice(0, -1)) await perform(op);
const last = ops[ops.length - 1];
if (refused === null) {
await perform(last);
} else {
const [k1, k2, schema, alias] = refused;
const before = last[0] === "commit" || (last[0] === "open" && last[1] === null)
? (existsSync(file) ? await readFile(file) : Buffer.alloc(0))
: await readFile(join(vectors, last[1]!));
await assert.rejects(perform(last), { message: refusal(k1, k2, schema, alias) }, label);
assert.deepEqual(existsSync(file) ? await readFile(file) : Buffer.alloc(0), before, `${label}: the file is untouched`);
}
walked++;
}
} finally {
await rm(dir, { recursive: true, force: true });
}
assert.equal(walked, 18);
});
test("a refused load leaves the cache's prior state as it was, and a later good load still works", async () => {
const dir = await mkdtemp(join(tmpdir(), "uidx-load-"));
try {
@ -487,28 +549,3 @@ test("registerIndex of a second unique alias judges it, naming the two smallest
assert.equal(cache.getKeyByIndex(document, "a", "1"), "z", "the alias already installed is intact");
assert.equal(cache.getKeyByIndex(document, "b", "s"), undefined);
});
test("verdict parity: the same stores are refused whatever the load order, and the first collision refuses (text is informational)", async () => {
const dir = await mkdtemp(join(tmpdir(), "uidx-verdict-"));
try {
// Three records, one pair colliding; whichever position the pair has in the file, the store is refused and a clean store opens.
const roles = pair({ roles: { a: "a" } });
for (const [name, values, refused] of [
["pair-first", ["s", "s", "t"], true],
["pair-last", ["t", "s", "s"], true],
["pair-split", ["s", "t", "s"], true],
["distinct", ["r", "s", "t"], false],
] as const) {
const file = join(dir, `${name}.cc`);
const writer = CultCache.builder().withDocumentType(roles).withGenericStore(new SingleFileMessagePackBackingStore(file)).build();
for (const [i, a] of values.entries()) await writer.put(roles, `r${i}`, { a });
const unique = pair({ indexes: { a: "a" } });
const reader = CultCache.builder().withDocumentType(unique).withGenericStore(new SingleFileMessagePackBackingStore(file)).build();
if (refused) await assert.rejects(reader.pullAllBackingStores(), /both hold unique index 'a'/u, name);
else await reader.pullAllBackingStores();
assert.equal(reader.snapshot().length, refused ? 0 : 3, name);
}
} finally {
await rm(dir, { recursive: true, force: true });
}
});