@mason-data/cli 0.8.0 → 0.9.0

package/dist/src/duck/token.js

@@ -14,11 +14,14 @@ export function mintToken(apiKey, seconds, now = Date.now()) {
     const expiry = Math.floor(now / 1000) + seconds;
     return { token: `${PREFIX}_${expiry}_${mac(apiKey, expiry)}`, expiresAt: new Date(expiry * 1000).toISOString() };
 }
+const TOKEN_RE = /^mason_(\d{1,12})_([A-Za-z0-9_-]{43})$/;
+/** Whether a password is a token by its shape, good or not. */
+export const isToken = (password) => TOKEN_RE.test(password);
 /** null when the token is good; else why not, in a sentence a person reads after "password authentication failed" */
 export function tokenProblem(apiKey, token, now = Date.now()) {
     if (token === apiKey)
         return "the API key is not a password here: mason user add <name> makes a user whose password does not expire";
-    const m = /^mason_(\d{1,12})_([A-Za-z0-9_-]{43})$/.exec(token);
+    const m = TOKEN_RE.exec(token);
     if (m === null)
         return "not a Mason user's password or token: mason user add <name> makes a user whose password does not expire";
     const expiry = Number(m[1]);

package/dist/src/duck/types.js

@@ -11,9 +11,11 @@
  * UNION as its text. Each column that diverges carries the reason, and the runner logs it once per statement.
  */
 import { Buffer } from "node:buffer";
-import { DuckDBTypeId, DuckDBDateValue, DuckDBTimestampValue, DuckDBTimestampTZValue, DuckDBDecimalValue, DuckDBUUIDValue, DuckDBListValue, DuckDBArrayValue, DuckDBIntervalValue, DuckDBBlobValue, DuckDBTimeValue, DuckDBListType, DuckDBArrayType, DuckDBTimestampSecondsValue, DuckDBTimestampMillisecondsValue, DuckDBTimestampNanosecondsValue, } from "@duckdb/node-api";
+import { DuckDBTypeId, DuckDBDateValue, DuckDBTimestampValue, DuckDBTimestampTZValue, DuckDBDecimalValue, DuckDBUUIDValue, DuckDBListValue, DuckDBArrayValue, DuckDBIntervalValue, DuckDBBlobValue, DuckDBTimeValue, DuckDBListType, DuckDBArrayType, DuckDBTimeTZValue, DuckDBTimestampSecondsValue, DuckDBTimestampMillisecondsValue, DuckDBTimestampNanosecondsValue, } from "@duckdb/node-api";
 import { jsonbText } from "./jsonb.js";
 import { gapMark } from "./gaps.js";
+import { heldType, ZONED_HELD } from "./held-types.js";
+import { pgEvalSync } from "./pg-eval.js";
 import { EncodeError, encodeValue, OID, TIMESTAMP } from "../api/postgres/encode.js";
 const MICROS_PER_DAY = 86400000000n;
 const DATE_INFINITY = 2_147_483_647, DATE_NEG_INFINITY = -2_147_483_647;
@@ -137,6 +139,12 @@ function timeText(micros) {
     const secs = micros / 1000000n, frac = micros % 1000000n;
     return `${pad(secs / 3600n)}:${pad((secs % 3600n) / 60n)}:${pad(secs % 60n)}${frac === 0n ? "" : `.${pad(frac, 6).replace(/0+$/, "")}`}`;
 }
+/** A timetz as timetz_out writes it: the time, then its offset as `+HH`, `+HH:MM` or `+HH:MM:SS` (DuckDB's offset, as
+ *  Postgres writes it, is east of UTC; its stored zone is west) */
+export function timetzText(v) {
+    const abs = Math.abs(v.offset), s = abs % 60, m = Math.trunc(abs / 60) % 60;
+    return `${timeText(v.micros)}${v.offset < 0 ? "-" : "+"}${pad(Math.trunc(abs / 3600))}${m !== 0 || s !== 0 ? `:${pad(m)}` : ""}${s !== 0 ? `:${pad(s)}` : ""}`;
+}
 /** An interval as Postgres's `interval_out` writes it under IntervalStyle postgres: `1 year 2 mons 3 days 04:05:06.7` */
 export function intervalText(months, days, micros) {
     const parts = [];
@@ -213,7 +221,7 @@ function arrayElement(text) {
 /** Postgres's array OIDs of the element OIDs this file sends */
 const ARRAY_OID = {
     16: 1000, 17: 1001, 20: 1016, 21: 1005, 23: 1007, 25: 1009, 700: 1021, 701: 1022, 1082: 1182, 1083: 1183,
-    1114: 1115, 1184: 1185, 1186: 1187, 1700: 1231, 2950: 2951, 3802: 3807,
+    1114: 1115, 1184: 1185, 1186: 1187, 1266: 1270, 1700: 1231, 2950: 2951, 3802: 3807,
 };
 function custom(name, oid, size, text, binary, divergence = null) {
     return {
@@ -285,6 +293,8 @@ export function wireColumn(name, t) {
         case DuckDBTypeId.TIMESTAMP_TZ: return timestampColumn(name, true);
         case DuckDBTypeId.TIME:
             return custom(name, 1083, 8, (v) => timeText(v.micros), (v) => { const b = Buffer.alloc(8); b.writeBigInt64BE(v.micros); return b; });
+        case DuckDBTypeId.TIME_TZ:
+            return custom(name, 1266, 12, (v) => timetzText(v), (v) => { const t = v; const b = Buffer.alloc(12); b.writeBigInt64BE(t.micros, 0); b.writeInt32BE(-t.offset, 8); return b; });
         case DuckDBTypeId.INTERVAL:
             return custom(name, 1186, 16, (v) => { const i = v; return intervalText(i.months, i.days, i.micros); }, (v) => { const i = v; const b = Buffer.alloc(16); b.writeBigInt64BE(i.micros, 0); b.writeInt32BE(i.days, 8); b.writeInt32BE(i.months, 12); return b; });
         case DuckDBTypeId.BLOB:
@@ -313,7 +323,7 @@ export function wireColumn(name, t) {
 // int4, its DECIMAL as a numeric, its JSON as jsonb's canonical text. Where DuckDB's value cannot be written as
 // Postgres's type (a timestamp where Postgres says date), DuckDB's own encoding stands and the column says why.
 /** pg_type.typlen of the OIDs a result column may have */
-const TYPLEN = { 16: 1, 18: 1, 19: 64, 20: 8, 21: 2, 23: 4, 26: 4, 700: 4, 701: 8, 1082: 4, 1083: 8, 1114: 8, 1184: 8, 1186: 16, 2950: 16 };
+const TYPLEN = { 16: 1, 18: 1, 19: 64, 20: 8, 21: 2, 23: 4, 26: 4, 700: 4, 701: 8, 1082: 4, 1083: 8, 1114: 8, 1184: 8, 1186: 16, 1266: 12, 2950: 16 };
 const PG_TEXT = new Set([25, 1043, 1042, 19, 18]);
 const PG_INT = { 21: "int2", 23: "int4", 20: "int8" };
 const DUCK_EXACT = new Set([DuckDBTypeId.TINYINT, DuckDBTypeId.SMALLINT, DuckDBTypeId.INTEGER, DuckDBTypeId.BIGINT, DuckDBTypeId.HUGEINT,
@@ -349,6 +359,45 @@ export function floatAsNumeric(x, digits = 15) {
     }
     return s === "-0" ? "0" : s;
 }
+/** the binary forms of the text-held types a driver asks for by default (pgjdbc: point, box): each a float8's bytes */
+const HELD_BINARY = {
+    600: floats, 603: floats,
+};
+function floats(text) {
+    const parts = text.match(/-?(?:Infinity|NaN|[\d.]+(?:e[-+]?\d+)?)/g) ?? [];
+    const b = Buffer.alloc(parts.length * 8);
+    parts.forEach((p, i) => b.writeDoubleBE(Number(p), i * 8));
+    return b;
+}
+/** a time range's text in the session's zone: Postgres's own output, a value at a time, the last few remembered */
+const zonedTexts = new Map();
+function inZone(type, text, zone) {
+    const key = `${zone}\u0000${type}\u0000${text}`;
+    let written = zonedTexts.get(key);
+    if (written === undefined) {
+        written = pgEvalSync(`SELECT $1::${type}::text`, [text], zone)[0]?.[0] ?? text;
+        if (zonedTexts.size >= 1024)
+            zonedTexts.clear();
+        zonedTexts.set(key, written);
+    }
+    return written;
+}
+/** a column of a type held as its text. A value's text only: Mason writes the binary form of few of them */
+function heldColumn(name, oid, own, size) {
+    const binary = HELD_BINARY[oid], zoned = ZONED_HELD.get(oid);
+    return {
+        name, oid, size, divergence: null,
+        encode: (v, format, s) => {
+            if (v === null)
+                return null;
+            if (format === 0)
+                return zoned !== undefined ? Buffer.from(inZone(zoned, str(v), s.timeZone), "utf8") : own.encode(v, 0, s);
+            if (binary !== undefined)
+                return binary(str(v));
+            throw new EncodeError(`column ${JSON.stringify(name)} (type OID ${oid}) is held as its text and cannot be sent in binary by Mason: ask for text (${gapMark("binary-held-type")})`, "0A000", "binary-held-type");
+        },
+    };
+}
 /** The wire column of a result column whose Postgres type the analysis gave, holding DuckDB values of type `t`. */
 export function pgTypedColumn(name, oid, t, typmod = -1) {
     const own = wireColumn(name, t);
@@ -360,6 +409,16 @@ export function pgTypedColumn(name, oid, t, typmod = -1) {
         return retyped(viaApi(name, "numeric", (v) => padDecimals(exactText(v), (typmod - 4) & 0xffff)));
     if (own.oid === oid)
         return retyped(own);
+    // a type the copy holds as its text, or an array of one (src/duck/held-types.ts): the text, under Postgres's OID
+    const held = heldType(oid);
+    if (held !== undefined && (own.oid === 25 || own.oid === 1009))
+        return heldColumn(name, oid, own, held.oid === oid ? held.size : -1);
+    // an oid element of an oid[], held as its digits
+    if (oid === 26 && own.oid === 25)
+        return custom(name, 26, 4, str, (v) => { const b = Buffer.alloc(4); b.writeUInt32BE(Number(str(v))); return b; });
+    // an oid: DuckDB's UINTEGER
+    if (oid === 26 && DUCK_EXACT.has(t.typeId))
+        return custom(name, 26, 4, (v) => exactText(v), (v) => { const b = Buffer.alloc(4); b.writeUInt32BE(Number(exactText(v))); return b; });
     // an inet: stored as Postgres's text with its mask (`10.3.6.45/32`), sent as inet_out writes it (a full mask dropped)
     if (oid === 869 && own.oid === 25)
         return custom(name, 869, -1, (v) => inetOut(str(v)), (v) => inetBytes(str(v)));
@@ -404,12 +463,9 @@ export function pgTypedColumn(name, oid, t, typmod = -1) {
     }
     return { ...own, divergence: `Postgres says type OID ${oid}; the engine computed ${t.toString()}, sent as its own type` };
 }
-/** How a cell of an exact numeric (src/duck/rewrite.ts ExactColumn) reads as a decimal text, by DuckDB's type: an exact
- *  number's digits, a float as Postgres's cast to numeric writes it — or null where the cell needs digits DuckDB did not
- *  keep (a double where the program wants exact ones). */
 export function cellText(t, conv) {
     // a numeric's text as Postgres's numeric_out wrote it (a text read by numeric_in, src/duck/pg-eval.ts)
-    if (conv === "text")
+    if (conv === "text" || conv === "parsed")
         return t.typeId === DuckDBTypeId.VARCHAR ? (v) => String(v) : null;
     if (DUCK_EXACT.has(t.typeId))
         return exactText;

package/dist/src/duck/usage.js

@@ -1,5 +1,5 @@
 import { MASON_SCHEMA } from "./store.js";
-import { History, Streak } from "./history.js";
+import { History, Streak, cutoff } from "./history.js";
 import { outcomeOf } from "../api/request-log.js";
 export const USAGE_TABLE = `${MASON_SCHEMA}.usage`;
 export const SIZES_TABLE = `${MASON_SCHEMA}.copy_sizes`;
@@ -265,10 +265,12 @@ export class UsageSampler {
             this.log(`usage: measured ${rows.length} synced tables' own size in ${Math.round(performance.now() - t0)} ms (every ${Math.round((this.options.sizeEveryMs ?? SIZE_EVERY_MS) / 60_000)} min)`);
         }
     }
-    /** Deletes the samples and sizes older than the retention. */
+    /** Deletes the samples and sizes older than the retention, counted back from the module's ONE clock (`now`, which
+     *  stamps each sample), never the engine's `now()` (src/duck/events.ts `trim` says why). */
     async trim() {
-        await this.history.writer.run(`DELETE FROM ${USAGE_TABLE} WHERE sampled_at < now() - INTERVAL (${this.days}) DAY`);
-        await this.history.writer.run(`DELETE FROM ${SIZES_TABLE} WHERE sampled_at < now() - INTERVAL (${this.days}) DAY`);
+        const before = cutoff(this.now(), this.days);
+        await this.history.writer.run(`DELETE FROM ${USAGE_TABLE} WHERE sampled_at < $1::TIMESTAMPTZ`, [before]);
+        await this.history.writer.run(`DELETE FROM ${SIZES_TABLE} WHERE sampled_at < $1::TIMESTAMPTZ`, [before]);
     }
     /**
      * The samples from `since` until `until`, oldest first; with `stepSeconds`, re-bucketed in ONE query: buckets of that

package/dist/src/duck/users.js

@@ -59,6 +59,8 @@ export const newUserId = () => crypto.randomBytes(8).toString("hex");
 export const userPassword = (apiKey, name, id) => `mason_${id}_${mac(apiKey, name, id)}`;
 /** Whether a password is a user's by its shape (else it is read as a token, src/duck/token.ts). */
 export const isUserPassword = (password) => PASSWORD_RE.test(password);
+/** What every `mason share` password begins with (src/cli/share-run.ts `newSharePassword`): no user's has it, `share` being no hex. */
+export const SHARE_PASSWORD_PREFIX = "mason_share_";
 /** Whether a user was added under this API key (one rotated since is no longer live). */
 export const underKey = (u, apiKey) => u.key === keyFingerprint(apiKey);
 /** The users the file lists; none when there is no file; why not, when it cannot be read — never an empty list then. */
@@ -263,12 +265,15 @@ export const disabledMessage = (name) => `role "${name}" is not permitted to log
  * HOW THE DUCKDB RUN'S LISTENER TAKES A PASSWORD (src/cli/dev.ts; the tests wire the same): a user's password (its shape)
  * is the users file's to judge, and its session is asked again before every statement (a stat); anything else is read as a
  * token, the kind `mason connect` minted before `mason user` (src/duck/token.ts: the API key used as a password is refused by name).
+ * A `mason share` password is refused by name too: it was never this port's, and making a user is not the way out.
  */
 export function listenerLogins(users, apiKey) {
     return {
         authenticate: (password, user) => {
             if (isUserPassword(password))
                 return users.login(user, password);
+            if (password.startsWith(SHARE_PASSWORD_PREFIX))
+                throw new Error("a mason share password logs in only at the address mason share printed, never on this port");
             const problem = tokenProblem(apiKey, password);
             if (problem !== null)
                 throw new Error(problem);