sqllens 1.1.0 → 1.2.0

package/dist/tsql/behavior.d.ts

@@ -0,0 +1,2 @@
+import type { DialectBehavior } from "../dialect-behavior/behavior.js";
+export declare const tsqlBehavior: DialectBehavior;

package/dist/tsql/behavior.js

@@ -0,0 +1,20 @@
+import { acceptsFor } from "../dialect-behavior/coerce-rules.js";
+import { likePatternToRegExp } from "../scope/like-pattern.js";
+import { SIGNATURES } from "../signature/signatures.js";
+import { displayName, fold, foldTableName, matchesSourceKey } from "./fold.js";
+import { tsqlLiteral, tsqlParseType, tsqlSpecial, TSQL_FUNCTION_RETURNS } from "./infer.js";
+export const tsqlBehavior = {
+    fold,
+    displayName,
+    foldTableName,
+    matchesSourceKey,
+    likeMatch: (pattern, value) => likePatternToRegExp(pattern).test(value),
+    literal: tsqlLiteral,
+    parseType: tsqlParseType,
+    functions: TSQL_FUNCTION_RETURNS,
+    division: "integer",
+    special: tsqlSpecial,
+    signatures: SIGNATURES.tsql,
+    // T-SQL implicit coercion: VARCHAR containing a number coerces to numeric (STR_TO_NUM=true), bool<->num coerces too (BOOL_NUM=true).
+    accepts: (argType, paramText) => acceptsFor(tsqlParseType, true, true, argType, paramText),
+};

package/dist/tsql/fold.d.ts

@@ -0,0 +1,8 @@
+import { type FoldRule, type IdentKind } from "../ident/fold.js";
+export declare const TSQL_FOLD_RULE: FoldRule;
+/** Fold an identifier to its T-SQL identity key. */
+export declare function fold(raw: string, kind?: IdentKind): string;
+/** Presentation twin: strip delimiters, no case change. */
+export declare function displayName(raw: string): string;
+export declare function foldTableName(parts: string[]): string[];
+export declare function matchesSourceKey(key: string, rawPart: string): boolean;

package/dist/tsql/fold.js

@@ -0,0 +1,34 @@
+// T-SQL identifier folding. The FoldRule plus its bound engine, colocated here because BOTH the
+// upstream lower() and the downstream DialectBehavior need it (the fold rule is the one dialect concern
+// used at two stages).
+// learn.microsoft.com/en-us/sql/relational-databases/databases/database-identifiers +
+// .../sql/t-sql/statements/collations — verified live: case folding is NOT a hardcoded T-SQL
+// rule, it's a function of the identifier's COLLATION: "you can create two tables with names
+// that differ only in case in a database that has case-sensitive collation, but you can't...
+// in a database that has case-insensitive collation." Most SQL Server / Fabric SQL database
+// defaults are case-insensitive (e.g. SQL_Latin1_General_CP1_CI_AS), but this is not universal
+// — Fabric Warehouse defaults to the case-SENSITIVE Latin1_General_100_BIN2_UTF8. This module
+// encodes the common default-collation (CI) behavior; a caller with a known case-sensitive
+// collation is a documented boundary this module does not cross. Delimiting with `[ ]` or
+// `" "` does not itself change case behavior — quoting only unlocks reserved words/specials.
+import { displayWith, foldWith } from "../ident/fold.js";
+const DOUBLE_QUOTE = ['"', '"'];
+export const TSQL_FOLD_RULE = {
+    delimiters: [["[", "]"], DOUBLE_QUOTE],
+    unquoted: "lower",
+    quoted: "lower",
+};
+/** Fold an identifier to its T-SQL identity key. */
+export function fold(raw, kind = "other") {
+    return foldWith(TSQL_FOLD_RULE, raw, kind);
+}
+/** Presentation twin: strip delimiters, no case change. */
+export function displayName(raw) {
+    return displayWith(TSQL_FOLD_RULE, raw);
+}
+export function foldTableName(parts) {
+    return parts.map((p) => fold(p, "table"));
+}
+export function matchesSourceKey(key, rawPart) {
+    return key === fold(rawPart) || key === fold(rawPart, "table");
+}

package/dist/tsql/index.d.ts

@@ -0,0 +1,7 @@
+import { lower } from "./lower.js";
+import { parseTSql } from "./parse.js";
+export declare const tsql: {
+    parse: typeof parseTSql;
+    lower: typeof lower;
+    behavior: import("../dialect-behavior/behavior.js").DialectBehavior;
+};

package/dist/tsql/index.js

@@ -0,0 +1,10 @@
+// The complete tsql dialect module: parse + lower (front end) and behavior (semantic knowledge).
+// The registry wires this; to understand everything sqllens does for tsql, read this folder.
+import { tsqlBehavior } from "./behavior.js";
+import { lower } from "./lower.js";
+import { parseTSql } from "./parse.js";
+export const tsql = {
+    parse: parseTSql,
+    lower,
+    behavior: tsqlBehavior,
+};

package/dist/tsql/infer.d.ts

@@ -0,0 +1,16 @@
+import { type FnRule } from "../infer/functions.js";
+import { type Type } from "../infer/types.js";
+import type { Expr } from "../ir/ir.js";
+export declare const TSQL_FUNCTION_RETURNS: Record<string, FnRule>;
+/** T-SQL pre-registry inference hook — the XML data type methods, which `lowerUdtElem` lowers to
+ *  `function` nodes with the receiver as arg 0. `value(xpath, 'sqltype')` is typed by its literal
+ *  sqltype argument (never-wrong: no XML shredding — the declared sqltype is the value's runtime
+ *  type; a non-literal sqltype falls through to unknown); `exist()` → boolean (bit); `query()` → xml.
+ *  https://learn.microsoft.com/en-us/sql/t-sql/xml/xml-data-type-methods */
+export declare function tsqlSpecial(fn: Extract<Expr, {
+    kind: "function";
+}>): Type | undefined;
+/** T-SQL literal forms: `'str'` / `N'unicode'` → string, `0x…` → binary, `1` → int, `1.5` → decimal
+ *  (numeric), `1e3` → float. */
+export declare function tsqlLiteral(text: string): Type;
+export declare function tsqlParseType(text: string): Type;

package/dist/tsql/infer.js

@@ -0,0 +1,289 @@
+import { commonType } from "../infer/coerce.js";
+import { B, BIG, BIN, D, DATE, I, S, TS, common, dateArg, fixed, firstArg, group, } from "../infer/functions.js";
+import { parseType, scalar, TSQL_ALIASES, UNKNOWN } from "../infer/types.js";
+import { fold } from "./fold.js";
+/** T-SQL SUM/AVG return type (per the MS reference): tinyint/smallint promote to int; int/bigint/
+ *  decimal/float keep their (canonical) type. Unlike Spark, SUM(int) is int, not bigint. */
+const tsqlNumericAgg = (args) => {
+    const t = args[0];
+    if (t?.kind !== "scalar")
+        return UNKNOWN;
+    return t.name === "tinyint" || t.name === "smallint" ? I : t;
+};
+// Function return-type registry for T-SQL (Transact-SQL), from Microsoft's built-in function
+// reference (verified against MS Learn). Same discipline as the Spark registry: a missing rule
+// yields `unknown`, never a wrong type. Where T-SQL and Spark share a name but differ in meaning,
+// the T-SQL rule wins here (e.g. `count` is int, not bigint; `sum(int)` is int, not bigint;
+// `isnull(check, repl)` returns check's type, not a boolean predicate). CAST/CONVERT/TRY_CAST/PARSE
+// lower to cast nodes, not functions, so they aren't here.
+export const TSQL_FUNCTION_RETURNS = {
+    // string → string
+    ...group(fixed(S), [
+        "left",
+        "right",
+        "substring",
+        "upper",
+        "lower",
+        "ltrim",
+        "rtrim",
+        "trim",
+        "replace",
+        "replicate",
+        "reverse",
+        "stuff",
+        "concat",
+        "concat_ws",
+        "format",
+        "str",
+        "quotename",
+        "space",
+        "soundex",
+        "translate",
+        "string_escape",
+        "nchar",
+        "char",
+        "parsename",
+        "string_agg",
+        "json_value",
+        "json_query",
+        "json_modify",
+    ]),
+    // string → int
+    ...group(fixed(I), ["len", "datalength", "charindex", "patindex", "ascii", "unicode", "difference"]),
+    // date/time → datetime (canonical timestamp)
+    ...group(fixed(TS), [
+        "getdate",
+        "getutcdate",
+        "sysdatetime",
+        "sysutcdatetime",
+        "sysdatetimeoffset",
+        "current_timestamp",
+        "eomonth",
+        "switchoffset",
+        "todatetimeoffset",
+        "datetimefromparts",
+        "datetime2fromparts",
+        "smalldatetimefromparts",
+    ]),
+    datefromparts: fixed(DATE),
+    timefromparts: fixed(scalar("time")),
+    // date → int / bigint / string
+    ...group(fixed(I), ["datediff", "datepart", "year", "month", "day"]),
+    datediff_big: fixed(BIG),
+    datename: fixed(S),
+    // DATEADD keeps the date argument's type (datetime by default)
+    dateadd: dateArg,
+    // predicates that return int 0/1
+    ...group(fixed(I), [
+        "isdate",
+        "isnumeric",
+        "isjson",
+        "checksum",
+        "binary_checksum",
+        "object_id",
+        "grouping",
+        "grouping_id",
+    ]),
+    // numeric → same type as input (MS: ABS, CEILING, FLOOR, ROUND, SIGN, POWER, DEGREES, RADIANS)
+    ...group(firstArg, ["abs", "ceiling", "floor", "round", "sign", "power", "degrees", "radians"]),
+    // numeric → float (MS: EXP, LOG, LOG10, SQUARE, SQRT and the trig fns cast to float → canonical double)
+    ...group(fixed(D), [
+        "sqrt",
+        "square",
+        "exp",
+        "log",
+        "log10",
+        "sin",
+        "cos",
+        "tan",
+        "atn2",
+        "acos",
+        "asin",
+        "atan",
+        "cot",
+        "pi",
+        "rand",
+    ]),
+    // aggregates — COUNT is int (COUNT_BIG bigint); SUM/AVG promote small ints to int; MIN/MAX keep type
+    count: fixed(I),
+    count_big: fixed(BIG),
+    sum: tsqlNumericAgg,
+    avg: tsqlNumericAgg,
+    ...group(firstArg, ["min", "max"]),
+    ...group(fixed(D), ["stdev", "stdevp", "var", "varp"]),
+    // window/ranking — ROW_NUMBER/RANK/DENSE_RANK/NTILE → bigint; PERCENT_RANK/CUME_DIST → float;
+    // the value-returning analytics keep their argument's type
+    ...group(fixed(BIG), ["row_number", "rank", "dense_rank", "ntile"]),
+    ...group(fixed(D), ["percent_rank", "cume_dist"]),
+    ...group(firstArg, ["lag", "lead", "first_value", "last_value"]),
+    // null / choice — ISNULL/NULLIF return the first argument's type; COALESCE/IIF/CHOOSE a common type
+    isnull: firstArg,
+    nullif: firstArg,
+    ...group(common, ["coalesce"]),
+    iif: (args) => commonType(args.slice(-2)),
+    choose: (args) => commonType(args.slice(1)),
+    // logical / choice (SQL Server 2022+)
+    ...group(common, ["greatest", "least"]),
+    // window/analytic value functions keep the argument's type; PERCENTILE_CONT → float
+    ...group(firstArg, ["first_value", "last_value"]),
+    percentile_cont: fixed(D),
+    // JSON builders → string; JSON_PATH_EXISTS → int (bit)
+    ...group(fixed(S), ["json_object", "json_array"]),
+    json_path_exists: fixed(I),
+    // date/time helpers — DATETRUNC/DATE_BUCKET keep the date argument's type
+    ...group(dateArg, ["datetrunc", "date_bucket"]),
+    datetimeoffsetfromparts: fixed(TS),
+    ...group(fixed(S), ["current_timezone", "current_timezone_id"]),
+    // string-returning system / user / metadata-name functions
+    ...group(fixed(S), [
+        "formatmessage",
+        "system_user",
+        "session_user",
+        "current_user",
+        "user_name",
+        "suser_name",
+        "suser_sname",
+        "host_name",
+        "host_id",
+        "app_name",
+        "original_login",
+        "error_message",
+        "error_procedure",
+        "db_name",
+        "object_name",
+        "object_schema_name",
+        "schema_name",
+        "col_name",
+    ]),
+    // int-returning error / metadata-id / property functions
+    ...group(fixed(I), [
+        "error_number",
+        "error_severity",
+        "error_state",
+        "error_line",
+        "db_id",
+        "schema_id",
+        "col_length",
+        "columnproperty",
+        "objectproperty",
+        "objectpropertyex",
+        "type_id",
+        "xact_state",
+        "checksum_agg",
+    ]),
+    approx_count_distinct: fixed(BIG),
+    rowcount_big: fixed(BIG),
+    current_transaction_id: fixed(BIG),
+    // binary-returning compression / context functions
+    ...group(fixed(BIN), ["compress", "decompress", "context_info"]),
+    // system / metadata
+    newid: fixed(S),
+    newsequentialid: fixed(S),
+    hashbytes: fixed(BIN),
+    ...group(fixed(scalar("decimal")), ["scope_identity", "ident_current"]),
+    // ------------------------------------------------------------------
+    // Entries below: return types fetched from MS Learn 2026-06-10
+    // (per-function "Return types" sections, ver17 — the regex / fuzzy /
+    // vector / JSON-agg families are SQL Server 2025+). Functions whose
+    // documented type depends on an argument value or is a bare
+    // sql_variant stay absent → unknown.
+    // ------------------------------------------------------------------
+    // bit manipulation (SQL Server 2022): shifts/set keep the input's type
+    bit_count: fixed(BIG),
+    get_bit: fixed(B),
+    ...group(firstArg, ["left_shift", "right_shift", "set_bit"]),
+    // aggregates / analytic
+    any_value: firstArg,
+    approx_percentile_cont: fixed(D), // float(53)
+    // regex (2025; REGEXP_LIKE is documented as boolean-valued)
+    regexp_like: fixed(B),
+    regexp_count: fixed(I),
+    regexp_instr: fixed(I), // doc states "Integer."
+    ...group(fixed(S), ["regexp_replace", "regexp_substr"]),
+    // fuzzy string match (2025)
+    ...group(fixed(I), ["edit_distance", "edit_distance_similarity", "jaro_winkler_similarity"]),
+    jaro_winkler_distance: fixed(D), // float
+    // JSON (2025): aggregates are nvarchar(max); JSON_CONTAINS is int (0/1/NULL), not bit
+    ...group(fixed(S), ["json_arrayagg", "json_objectagg"]),
+    json_contains: fixed(I),
+    // encoding (2025)
+    base64_encode: fixed(S),
+    base64_decode: fixed(BIN),
+    // vectors (2025)
+    ...group(fixed(D), ["vector_distance", "vector_norm"]),
+    vector_normalize: fixed(scalar("vector")),
+    // metadata / system. object_definition is nvarchar(max); original_db_name and
+    // trigger_nestlevel state no type on their pages — a name and a count, typed as such.
+    ...group(fixed(S), ["type_name", "object_definition", "original_db_name"]),
+    ...group(fixed(I), [
+        "has_dbaccess",
+        "is_rolemember",
+        "is_srvrolemember",
+        "suser_id",
+        "user_id",
+        "database_principal_id",
+        "textvalid",
+        "trigger_nestlevel",
+    ]),
+    ...group(fixed(scalar("decimal")), ["ident_incr", "ident_seed"]),
+    cursor_status: fixed(scalar("smallint")),
+    ...group(fixed(BIN), [
+        "suser_sid",
+        "textptr",
+        "columns_updated",
+        "encryptbypassphrase",
+        "decryptbypassphrase",
+        "encryptbykey",
+        "decryptbykey",
+        "crypt_gen_random",
+    ]),
+    eventdata: fixed(scalar("xml")),
+};
+/** T-SQL pre-registry inference hook — the XML data type methods, which `lowerUdtElem` lowers to
+ *  `function` nodes with the receiver as arg 0. `value(xpath, 'sqltype')` is typed by its literal
+ *  sqltype argument (never-wrong: no XML shredding — the declared sqltype is the value's runtime
+ *  type; a non-literal sqltype falls through to unknown); `exist()` → boolean (bit); `query()` → xml.
+ *  https://learn.microsoft.com/en-us/sql/t-sql/xml/xml-data-type-methods */
+export function tsqlSpecial(fn) {
+    if (fn.qualifier !== undefined)
+        return undefined;
+    switch (fn.name.toLowerCase()) {
+        case "value": {
+            // args = [receiver, xpath, sqltype]; the second method argument (arg 2) is the declared type.
+            const sqltype = fn.args[2];
+            return sqltype?.kind === "literal" ? parseType(unquoteLiteral(sqltype.text), TSQL_ALIASES) : undefined;
+        }
+        case "exist":
+            return scalar("boolean"); // .exist() returns bit
+        case "query":
+            return scalar("xml"); // .query() returns xml
+        default:
+            return undefined;
+    }
+}
+/** Strip a surrounding SQL string-literal quote (`'varchar(100)'` → `varchar(100)`). */
+function unquoteLiteral(text) {
+    const t = text.trim();
+    return t.length >= 2 && t.startsWith("'") && t.endsWith("'") ? t.slice(1, -1) : t;
+}
+/** T-SQL literal forms: `'str'` / `N'unicode'` → string, `0x…` → binary, `1` → int, `1.5` → decimal
+ *  (numeric), `1e3` → float. */
+export function tsqlLiteral(text) {
+    const t = text.trim();
+    if (/^n?['"]/i.test(t))
+        return scalar("string");
+    if (/^0x/i.test(t))
+        return scalar("binary");
+    if (/^null$/i.test(t))
+        return UNKNOWN;
+    if (/^[+-]?\d+$/.test(t))
+        return scalar("int");
+    if (/^[+-]?(\d+\.?\d*|\.\d+)e[+-]?\d+$/i.test(t))
+        return scalar("float");
+    if (/^[+-]?(\d+\.\d*|\.\d+)$/.test(t))
+        return scalar("decimal");
+    return UNKNOWN;
+}
+export function tsqlParseType(text) {
+    return parseType(text, TSQL_ALIASES, fold);
+}

package/dist/tsql/lower.js

@@ -3,7 +3,7 @@ import { TSqlParser as P } from "../generated/tsql/TSqlParser.js";
 import { keywordCategory, swallowedCategories, swallowedStatements } from "../ir/statement.js";
 import { partSpansOf } from "../ir/part-span.js";
 import { freezeIR } from "../ir/freeze.js";
-import { displayName, foldIdentifier } from "../ident/fold.js";
+import { displayName, fold } from "./fold.js";
 // ---------------------------------------------------------------------------
 // Lowering — T-SQL (grammars-v4 sql/tsql) CST -> the shared, dialect-neutral IR
 // (src/ir/ir.ts). The semantic layer (scope/qualify/infer/lineage/symbols) runs
@@ -447,7 +447,7 @@ function lowerUdtElem(udt) {
         : [];
     return {
         kind: "function",
-        name: method ? displayName(method.getText(), "tsql").toLowerCase() : "",
+        name: method ? displayName(method.getText()).toLowerCase() : "",
         args: [receiverExpr, ...methodArgs],
         aggregate: false,
         distinct: false,
@@ -924,7 +924,7 @@ function windowParts(node) {
 /** Resolve a window name against the enclosing query spec's WINDOW clause. A definition may
  *  itself start with another window's name (chained); `seen` guards against cycles. */
 function resolveNamedWindow(from, name, seen) {
-    const key = foldIdentifier(name, "tsql");
+    const key = fold(name);
     if (seen.has(key))
         return undefined;
     seen.add(key);
@@ -938,7 +938,7 @@ function resolveNamedWindow(from, name, seen) {
         return undefined;
     for (const def of directChildrenOfRule(clause, P.RULE_window_definition)) {
         const defName = directChildrenOfRule(def, P.RULE_id_)[0]?.getText();
-        if (!defName || foldIdentifier(defName, "tsql") !== key)
+        if (!defName || fold(defName) !== key)
             continue;
         const ws = directChildrenOfRule(def, P.RULE_window_specification)[0];
         if (!ws)

package/dist/tsql/signatures.generated.d.ts

@@ -0,0 +1,6 @@
+import type { FnSignature } from "../signature/signatures.js";
+/** The merged function-signature table for tsql: curated overrides folded over the harvested
+ *  doc-derived long tail (overrides win by key, replacing the whole overload set), keyed by
+ *  lowercased name. Each name maps to an ORDERED overload set - a name with one documented shape
+ *  is a one-element array. `origin` says which layer produced the set. */
+export declare const TSQL_SIGNATURES: Record<string, FnSignature[]>;