@drzl/validation-core 3.22.2 → 3.22.4

package/dist/index.cjs

@@ -55,6 +55,7 @@ __export(index_exports, {
   canonicalMembers: () => canonicalMembers,
   canonicalNumericText: () => canonicalNumericText,
   classifyTableChecks: () => classifyTableChecks,
+  codePointCompare: () => codePointCompare,
   columnMetaFacts: () => columnMetaFacts,
   comparisonWire: () => comparisonWire,
   describeSet: () => describeSet,
@@ -67,6 +68,7 @@ __export(index_exports, {
   isProjectInstallPath: () => isProjectInstallPath,
   lengthCheckLabel: () => lengthCheckLabel,
   lengthMeasure: () => lengthMeasure,
+  measureCompare: () => measureCompare,
   measureExpression: () => measureExpression,
   moduleFileName: () => moduleFileName,
   moduleSpecifier: () => moduleSpecifier,
@@ -112,6 +114,34 @@ function lengthMeasure(column, check) {
   if (column.tsType !== "string") return void 0;
   return check.unit === "bytes" ? "utf8Bytes" : "codePoints";
 }
+function codePointCompare(variable, operator, bound) {
+  const units = `${variable}.length`;
+  const spread2 = `[...${variable}].length`;
+  switch (operator) {
+    case "<=":
+    case "<":
+      return `(${units} ${operator} ${bound} || ${spread2} ${operator} ${bound})`;
+    case ">=":
+    case ">":
+      return `(${units} ${operator} ${bound} && ${spread2} ${operator} ${bound})`;
+    case "===":
+      return `(${units} >= ${bound} && ${spread2} === ${bound})`;
+    case "!==":
+      return `(${units} < ${bound} || ${spread2} !== ${bound})`;
+  }
+}
+function measureCompare(measure, variable, operator, bound) {
+  const js = {
+    ">=": ">=",
+    ">": ">",
+    "<=": "<=",
+    "<": "<",
+    "=": "===",
+    "<>": "!=="
+  }[operator];
+  if (measure === "codePoints") return codePointCompare(variable, js, bound);
+  return `${measureExpression(measure, variable)} ${js} ${bound}`;
+}
 function measureExpression(measure, variable) {
   switch (measure) {
     case "codePoints":
@@ -123,7 +153,7 @@ function measureExpression(measure, variable) {
   }
 }
 function lengthCheckLabel(check) {
-  const fn = check.unit === "bytes" ? "octet_length" : "length";
+  const fn = check.fn ? check.fn === "char_length" ? "length" : check.fn : check.unit === "bytes" ? "octet_length" : "length";
   const rule = `${fn}(${check.column}) ${check.operator} ${check.value}`;
   return check.name ? `${check.name}: ${rule}` : rule;
 }
@@ -374,7 +404,13 @@ function foldDisjunctionToSet(branches, name) {
     sets: [{ column, values, kind, name }]
   };
 }
-function parseCheck(expression, name) {
+function lengthUnit(fn, dialect) {
+  const name = fn.toLowerCase();
+  if (name === "octet_length") return "bytes";
+  if (name === "char_length") return "characters";
+  return dialect === "mysql" ? "bytes" : "characters";
+}
+function parseCheck(expression, name, dialect) {
   const expr = unwrap((expression ?? "").trim());
   if (!expr) return { ok: false, reason: "empty expression" };
   if (expr.includes("?")) return { ok: false, reason: "expression contains an unresolved value" };
@@ -451,11 +487,12 @@ function parseCheck(expression, name) {
   const lengthOf = expr.match(LENGTH_OF);
   if (lengthOf) {
     const op2 = lengthOf[3] === "!=" ? "<>" : lengthOf[3];
-    const unit = lengthOf[1].toLowerCase() === "octet_length" ? "bytes" : "characters";
+    const fn = lengthOf[1].toLowerCase();
+    const unit = lengthUnit(fn, dialect);
     return {
       ok: true,
       checks: [],
-      lengths: [{ column: lengthOf[2], operator: op2, value: lengthOf[4], unit, name }]
+      lengths: [{ column: lengthOf[2], operator: op2, value: lengthOf[4], unit, fn, name }]
     };
   }
   const cardinalityOf = expr.match(CARDINALITY_OF);
@@ -962,7 +999,7 @@ function classifyTableChecks(table) {
   const out = [];
   for (const k of table.checks ?? []) {
     const expression = (k.expression ?? "").trim();
-    const parsed = parseCheck(k.expression, k.name);
+    const parsed = parseCheck(k.expression, k.name, table.dialect);
     if (!parsed.ok) {
       out.push({
         ...k.name ? { name: k.name } : {},
@@ -1944,6 +1981,7 @@ async function formatCode(code, filePath, fmt) {
   canonicalMembers,
   canonicalNumericText,
   classifyTableChecks,
+  codePointCompare,
   columnMetaFacts,
   comparisonWire,
   describeSet,
@@ -1956,6 +1994,7 @@ async function formatCode(code, filePath, fmt) {
   isProjectInstallPath,
   lengthCheckLabel,
   lengthMeasure,
+  measureCompare,
   measureExpression,
   moduleFileName,
   moduleSpecifier,

package/dist/index.d.cts

@@ -452,6 +452,15 @@ interface LengthCheck {
      * a given column.
      */
     unit?: 'characters' | 'bytes';
+    /**
+     * The function as written, before the engine decided what it counts.
+     *
+     * `unit` used to imply this, because the two were the same question: `octet_length` meant bytes
+     * and everything else meant characters. On MySQL `length()` also means bytes, so deriving the
+     * name back from the unit relabelled a user's `length(name) <= 5` as `octet_length(name) <= 5`,
+     * which is a constraint they did not write. The label the ledger matches on has to be theirs.
+     */
+    fn?: 'length' | 'char_length' | 'octet_length';
     name?: string;
 }
 /**
@@ -504,6 +513,35 @@ declare function lengthMeasure(column: {
  * TypeBox reach it off the row object as `o["blob"]`. `CODEPOINT_LENGTH` is the fixed-variable
  * spelling of the first arm and stays for the call sites that already use it.
  */
+/**
+ * A code-point count compared against a bound, without spreading the string when it cannot matter.
+ *
+ * `[...v].length` allocates an array of code points, and it sits on the path every ordinary row
+ * takes. A UTF-16 unit count is free and is never *smaller* than a code-point count, since only a
+ * surrogate pair costs two units for one code point. That one fact settles every operator here:
+ *
+ *   `<=`, `<`   units within the bound means code points are too      short-circuit to ACCEPT
+ *   `>=`, `>`   units under the bound means code points are too       short-circuit to REJECT
+ *   `===`       units under the bound means code points cannot equal it   short-circuit to REJECT
+ *   `!==`       units under the bound means code points cannot equal it   short-circuit to ACCEPT
+ *
+ * So the spread runs only for a string long enough in units to be in doubt, which for a cap is the
+ * only string that was ever in doubt. Measured over a 20000-string random pool plus the awkward
+ * cases (emoji, combining marks, a lone surrogate, empty, CJK at the bound): the two forms answer
+ * identically on every one. Throughput on the check alone, cap 64: ordinary ASCII rows 23171k/s
+ * against 215029k/s, rows at the cap 4250k/s against 235657k/s.
+ *
+ * Only the code-point measurement is rewritten. A `byteLength` reads `v.length` already and has
+ * nothing to avoid, and the UTF-8 arm has its own sound short-circuits in both directions but a
+ * different allocation to weigh, so it is left for its own measurement rather than folded in here.
+ */
+declare function codePointCompare(variable: string, operator: '>=' | '>' | '<=' | '<' | '===' | '!==', bound: string | number): string;
+/**
+ * A length comparison over whichever measurement the column answers, short-circuited where that is
+ * sound. The one place an operator and a count meet, so no generator has to re-derive which
+ * rewrites are safe; see `codePointCompare`.
+ */
+declare function measureCompare(measure: LengthMeasure, variable: string, operator: LengthCheck['operator'], bound: string | number): string;
 declare function measureExpression(measure: LengthMeasure, variable: string): string;
 /**
  * A count constraint as the sentence every emitted schema attaches and the ledger keys on.
@@ -558,21 +596,11 @@ type ParsedCheck = {
     reason: string;
 };
 /**
- * Parse one check expression.
- *
- * Deliberately narrow. `BETWEEN` is included because it is common and means exactly two
- * inclusive bounds; `AND` of arbitrary predicates is not, because getting its scope wrong would
- * silently change what is enforced.
- *
- * **The order below is load bearing, in both directions.** `BETWEEN` is matched before the `AND`
- * split because it *holds* an `AND`; splitting first turned every `BETWEEN` into an unparseable
- * pair and dropped a constraint that had been enforced. The unary `IS` predicates are matched
- * *after* the splits for the mirror-image reason: none of them holds an `AND` or an `OR`, and
- * their trailing operand is greedy, so matching first would let `a IS DISTINCT FROM 5 AND b > 0`
- * swallow the second predicate. `OR` is split before `AND` because SQL binds `AND` tighter, and
- * no SQL operator spells an `OR` inside itself the way `BETWEEN` spells an `AND`.
+ * `dialect` decides what `length()` counts, and nothing else in this parser. Optional, and absent
+ * means the Postgres reading, which is what every caller got before it existed: a caller that does
+ * not know its dialect keeps the answer it already had rather than being given a new one.
  */
-declare function parseCheck(expression: string | undefined, name?: string): ParsedCheck;
+declare function parseCheck(expression: string | undefined, name?: string, dialect?: string): ParsedCheck;
 /**
  * A number-kind literal, spelled for the wire type of the column it constrains.
  *
@@ -1668,4 +1696,4 @@ declare function isProjectInstallPath(manifestPath: string): boolean;
  */
 declare function formatCode(code: string, filePath: string, fmt?: FormatOptions): Promise<any>;
 
-export { AFFIX_PREFIX_PATTERN, AFFIX_PROBE_TABLE, AFFIX_SUFFIX_PATTERN, type AffixIssue, type AffixOptions, type AffixValue, type BrandAlias, type BrandPlan, type BrandingOption, type BrandingOptions, CODEPOINT_LENGTH, COERCIBLE_DATE_STRING, COLUMN_FORMATS, CONSTRAINTS_MODULE, type CardinalityCheck, type CheckPart, type CheckPartPlace, type ClassifiedCheck, type ColumnCheck, type ColumnMetaFacts, type ColumnSet, type ComparisonWire, type ConstraintFacts, type ConstraintKind, type ConstraintsModuleOptions, type ConstraintsOption, DEFAULT_IMPORT_EXTENSION, DEFAULT_MODE_PREFIX, DEFAULT_NESTED_DEPTH, DEFAULT_SCHEMA_SUFFIX, DEFAULT_TYPE_SUFFIX, type FileSink, type FormatOptions, type GeneratorFs, IMPORT_EXTENSIONS, type ImportExtension, type LengthCheck, type LengthMeasure, MAX_NESTED_DEPTH, type MetaFactOptions, NAME_MODES, NESTED_PREFIX, NUMERIC_CANON_NAME, NUMERIC_CANON_SOURCE, type NameMode, type NestedArm, type NestedMode, type NestedNode, type NullCheck, type ParsedCheck, type ResolvedAffix, type ResolvedBranding, type RowCheck, type Table, type TableCase, type TableConstraints, type TableMetaFacts, type TableMetaOptions, type UnenforcedLiteral, type UnenforcedPart, type ValidationGenerateOptions, type ValidationLibrary, type ValidationRenderer, type WireLiteralFit, type WireLiteralQuestion, applyTableCase, applyWirePolicy, buildBrandPlan, buildNestedPlan, canonicalMembers, canonicalNumericText, classifyTableChecks, columnMetaFacts, comparisonWire, describeSet, fileWriter, formatCode, importSpecifier, insertColumns, isGeneratedColumn, isIntegerColumn, isProjectInstallPath, lengthCheckLabel, lengthMeasure, measureExpression, moduleFileName, moduleSpecifier, needsNumericCanon, nestedArmNotes, nestedNodeColumns, nestedSchemaName, nestedTypeName, nonFiniteAccepted, nonFiniteRefused, parseCheck, parsesToADate, pascalCase, renderConstraintsModule, renderDuplicateFinder, resolveAffix, resolveBranding, resolveConfiguredImport, resolveConstraints, resolveNestedDepth, schemaName, selectColumns, tableConstraints, tableMetaFacts, typeName, updateColumns, validateAffix, wireLiteralFit, wireNumberLiteral };
+export { AFFIX_PREFIX_PATTERN, AFFIX_PROBE_TABLE, AFFIX_SUFFIX_PATTERN, type AffixIssue, type AffixOptions, type AffixValue, type BrandAlias, type BrandPlan, type BrandingOption, type BrandingOptions, CODEPOINT_LENGTH, COERCIBLE_DATE_STRING, COLUMN_FORMATS, CONSTRAINTS_MODULE, type CardinalityCheck, type CheckPart, type CheckPartPlace, type ClassifiedCheck, type ColumnCheck, type ColumnMetaFacts, type ColumnSet, type ComparisonWire, type ConstraintFacts, type ConstraintKind, type ConstraintsModuleOptions, type ConstraintsOption, DEFAULT_IMPORT_EXTENSION, DEFAULT_MODE_PREFIX, DEFAULT_NESTED_DEPTH, DEFAULT_SCHEMA_SUFFIX, DEFAULT_TYPE_SUFFIX, type FileSink, type FormatOptions, type GeneratorFs, IMPORT_EXTENSIONS, type ImportExtension, type LengthCheck, type LengthMeasure, MAX_NESTED_DEPTH, type MetaFactOptions, NAME_MODES, NESTED_PREFIX, NUMERIC_CANON_NAME, NUMERIC_CANON_SOURCE, type NameMode, type NestedArm, type NestedMode, type NestedNode, type NullCheck, type ParsedCheck, type ResolvedAffix, type ResolvedBranding, type RowCheck, type Table, type TableCase, type TableConstraints, type TableMetaFacts, type TableMetaOptions, type UnenforcedLiteral, type UnenforcedPart, type ValidationGenerateOptions, type ValidationLibrary, type ValidationRenderer, type WireLiteralFit, type WireLiteralQuestion, applyTableCase, applyWirePolicy, buildBrandPlan, buildNestedPlan, canonicalMembers, canonicalNumericText, classifyTableChecks, codePointCompare, columnMetaFacts, comparisonWire, describeSet, fileWriter, formatCode, importSpecifier, insertColumns, isGeneratedColumn, isIntegerColumn, isProjectInstallPath, lengthCheckLabel, lengthMeasure, measureCompare, measureExpression, moduleFileName, moduleSpecifier, needsNumericCanon, nestedArmNotes, nestedNodeColumns, nestedSchemaName, nestedTypeName, nonFiniteAccepted, nonFiniteRefused, parseCheck, parsesToADate, pascalCase, renderConstraintsModule, renderDuplicateFinder, resolveAffix, resolveBranding, resolveConfiguredImport, resolveConstraints, resolveNestedDepth, schemaName, selectColumns, tableConstraints, tableMetaFacts, typeName, updateColumns, validateAffix, wireLiteralFit, wireNumberLiteral };

package/dist/index.d.ts

@@ -452,6 +452,15 @@ interface LengthCheck {
      * a given column.
      */
     unit?: 'characters' | 'bytes';
+    /**
+     * The function as written, before the engine decided what it counts.
+     *
+     * `unit` used to imply this, because the two were the same question: `octet_length` meant bytes
+     * and everything else meant characters. On MySQL `length()` also means bytes, so deriving the
+     * name back from the unit relabelled a user's `length(name) <= 5` as `octet_length(name) <= 5`,
+     * which is a constraint they did not write. The label the ledger matches on has to be theirs.
+     */
+    fn?: 'length' | 'char_length' | 'octet_length';
     name?: string;
 }
 /**
@@ -504,6 +513,35 @@ declare function lengthMeasure(column: {
  * TypeBox reach it off the row object as `o["blob"]`. `CODEPOINT_LENGTH` is the fixed-variable
  * spelling of the first arm and stays for the call sites that already use it.
  */
+/**
+ * A code-point count compared against a bound, without spreading the string when it cannot matter.
+ *
+ * `[...v].length` allocates an array of code points, and it sits on the path every ordinary row
+ * takes. A UTF-16 unit count is free and is never *smaller* than a code-point count, since only a
+ * surrogate pair costs two units for one code point. That one fact settles every operator here:
+ *
+ *   `<=`, `<`   units within the bound means code points are too      short-circuit to ACCEPT
+ *   `>=`, `>`   units under the bound means code points are too       short-circuit to REJECT
+ *   `===`       units under the bound means code points cannot equal it   short-circuit to REJECT
+ *   `!==`       units under the bound means code points cannot equal it   short-circuit to ACCEPT
+ *
+ * So the spread runs only for a string long enough in units to be in doubt, which for a cap is the
+ * only string that was ever in doubt. Measured over a 20000-string random pool plus the awkward
+ * cases (emoji, combining marks, a lone surrogate, empty, CJK at the bound): the two forms answer
+ * identically on every one. Throughput on the check alone, cap 64: ordinary ASCII rows 23171k/s
+ * against 215029k/s, rows at the cap 4250k/s against 235657k/s.
+ *
+ * Only the code-point measurement is rewritten. A `byteLength` reads `v.length` already and has
+ * nothing to avoid, and the UTF-8 arm has its own sound short-circuits in both directions but a
+ * different allocation to weigh, so it is left for its own measurement rather than folded in here.
+ */
+declare function codePointCompare(variable: string, operator: '>=' | '>' | '<=' | '<' | '===' | '!==', bound: string | number): string;
+/**
+ * A length comparison over whichever measurement the column answers, short-circuited where that is
+ * sound. The one place an operator and a count meet, so no generator has to re-derive which
+ * rewrites are safe; see `codePointCompare`.
+ */
+declare function measureCompare(measure: LengthMeasure, variable: string, operator: LengthCheck['operator'], bound: string | number): string;
 declare function measureExpression(measure: LengthMeasure, variable: string): string;
 /**
  * A count constraint as the sentence every emitted schema attaches and the ledger keys on.
@@ -558,21 +596,11 @@ type ParsedCheck = {
     reason: string;
 };
 /**
- * Parse one check expression.
- *
- * Deliberately narrow. `BETWEEN` is included because it is common and means exactly two
- * inclusive bounds; `AND` of arbitrary predicates is not, because getting its scope wrong would
- * silently change what is enforced.
- *
- * **The order below is load bearing, in both directions.** `BETWEEN` is matched before the `AND`
- * split because it *holds* an `AND`; splitting first turned every `BETWEEN` into an unparseable
- * pair and dropped a constraint that had been enforced. The unary `IS` predicates are matched
- * *after* the splits for the mirror-image reason: none of them holds an `AND` or an `OR`, and
- * their trailing operand is greedy, so matching first would let `a IS DISTINCT FROM 5 AND b > 0`
- * swallow the second predicate. `OR` is split before `AND` because SQL binds `AND` tighter, and
- * no SQL operator spells an `OR` inside itself the way `BETWEEN` spells an `AND`.
+ * `dialect` decides what `length()` counts, and nothing else in this parser. Optional, and absent
+ * means the Postgres reading, which is what every caller got before it existed: a caller that does
+ * not know its dialect keeps the answer it already had rather than being given a new one.
  */
-declare function parseCheck(expression: string | undefined, name?: string): ParsedCheck;
+declare function parseCheck(expression: string | undefined, name?: string, dialect?: string): ParsedCheck;
 /**
  * A number-kind literal, spelled for the wire type of the column it constrains.
  *
@@ -1668,4 +1696,4 @@ declare function isProjectInstallPath(manifestPath: string): boolean;
  */
 declare function formatCode(code: string, filePath: string, fmt?: FormatOptions): Promise<any>;
 
-export { AFFIX_PREFIX_PATTERN, AFFIX_PROBE_TABLE, AFFIX_SUFFIX_PATTERN, type AffixIssue, type AffixOptions, type AffixValue, type BrandAlias, type BrandPlan, type BrandingOption, type BrandingOptions, CODEPOINT_LENGTH, COERCIBLE_DATE_STRING, COLUMN_FORMATS, CONSTRAINTS_MODULE, type CardinalityCheck, type CheckPart, type CheckPartPlace, type ClassifiedCheck, type ColumnCheck, type ColumnMetaFacts, type ColumnSet, type ComparisonWire, type ConstraintFacts, type ConstraintKind, type ConstraintsModuleOptions, type ConstraintsOption, DEFAULT_IMPORT_EXTENSION, DEFAULT_MODE_PREFIX, DEFAULT_NESTED_DEPTH, DEFAULT_SCHEMA_SUFFIX, DEFAULT_TYPE_SUFFIX, type FileSink, type FormatOptions, type GeneratorFs, IMPORT_EXTENSIONS, type ImportExtension, type LengthCheck, type LengthMeasure, MAX_NESTED_DEPTH, type MetaFactOptions, NAME_MODES, NESTED_PREFIX, NUMERIC_CANON_NAME, NUMERIC_CANON_SOURCE, type NameMode, type NestedArm, type NestedMode, type NestedNode, type NullCheck, type ParsedCheck, type ResolvedAffix, type ResolvedBranding, type RowCheck, type Table, type TableCase, type TableConstraints, type TableMetaFacts, type TableMetaOptions, type UnenforcedLiteral, type UnenforcedPart, type ValidationGenerateOptions, type ValidationLibrary, type ValidationRenderer, type WireLiteralFit, type WireLiteralQuestion, applyTableCase, applyWirePolicy, buildBrandPlan, buildNestedPlan, canonicalMembers, canonicalNumericText, classifyTableChecks, columnMetaFacts, comparisonWire, describeSet, fileWriter, formatCode, importSpecifier, insertColumns, isGeneratedColumn, isIntegerColumn, isProjectInstallPath, lengthCheckLabel, lengthMeasure, measureExpression, moduleFileName, moduleSpecifier, needsNumericCanon, nestedArmNotes, nestedNodeColumns, nestedSchemaName, nestedTypeName, nonFiniteAccepted, nonFiniteRefused, parseCheck, parsesToADate, pascalCase, renderConstraintsModule, renderDuplicateFinder, resolveAffix, resolveBranding, resolveConfiguredImport, resolveConstraints, resolveNestedDepth, schemaName, selectColumns, tableConstraints, tableMetaFacts, typeName, updateColumns, validateAffix, wireLiteralFit, wireNumberLiteral };
+export { AFFIX_PREFIX_PATTERN, AFFIX_PROBE_TABLE, AFFIX_SUFFIX_PATTERN, type AffixIssue, type AffixOptions, type AffixValue, type BrandAlias, type BrandPlan, type BrandingOption, type BrandingOptions, CODEPOINT_LENGTH, COERCIBLE_DATE_STRING, COLUMN_FORMATS, CONSTRAINTS_MODULE, type CardinalityCheck, type CheckPart, type CheckPartPlace, type ClassifiedCheck, type ColumnCheck, type ColumnMetaFacts, type ColumnSet, type ComparisonWire, type ConstraintFacts, type ConstraintKind, type ConstraintsModuleOptions, type ConstraintsOption, DEFAULT_IMPORT_EXTENSION, DEFAULT_MODE_PREFIX, DEFAULT_NESTED_DEPTH, DEFAULT_SCHEMA_SUFFIX, DEFAULT_TYPE_SUFFIX, type FileSink, type FormatOptions, type GeneratorFs, IMPORT_EXTENSIONS, type ImportExtension, type LengthCheck, type LengthMeasure, MAX_NESTED_DEPTH, type MetaFactOptions, NAME_MODES, NESTED_PREFIX, NUMERIC_CANON_NAME, NUMERIC_CANON_SOURCE, type NameMode, type NestedArm, type NestedMode, type NestedNode, type NullCheck, type ParsedCheck, type ResolvedAffix, type ResolvedBranding, type RowCheck, type Table, type TableCase, type TableConstraints, type TableMetaFacts, type TableMetaOptions, type UnenforcedLiteral, type UnenforcedPart, type ValidationGenerateOptions, type ValidationLibrary, type ValidationRenderer, type WireLiteralFit, type WireLiteralQuestion, applyTableCase, applyWirePolicy, buildBrandPlan, buildNestedPlan, canonicalMembers, canonicalNumericText, classifyTableChecks, codePointCompare, columnMetaFacts, comparisonWire, describeSet, fileWriter, formatCode, importSpecifier, insertColumns, isGeneratedColumn, isIntegerColumn, isProjectInstallPath, lengthCheckLabel, lengthMeasure, measureCompare, measureExpression, moduleFileName, moduleSpecifier, needsNumericCanon, nestedArmNotes, nestedNodeColumns, nestedSchemaName, nestedTypeName, nonFiniteAccepted, nonFiniteRefused, parseCheck, parsesToADate, pascalCase, renderConstraintsModule, renderDuplicateFinder, resolveAffix, resolveBranding, resolveConfiguredImport, resolveConstraints, resolveNestedDepth, schemaName, selectColumns, tableConstraints, tableMetaFacts, typeName, updateColumns, validateAffix, wireLiteralFit, wireNumberLiteral };

package/dist/index.js

@@ -13,6 +13,34 @@ function lengthMeasure(column, check) {
   if (column.tsType !== "string") return void 0;
   return check.unit === "bytes" ? "utf8Bytes" : "codePoints";
 }
+function codePointCompare(variable, operator, bound) {
+  const units = `${variable}.length`;
+  const spread2 = `[...${variable}].length`;
+  switch (operator) {
+    case "<=":
+    case "<":
+      return `(${units} ${operator} ${bound} || ${spread2} ${operator} ${bound})`;
+    case ">=":
+    case ">":
+      return `(${units} ${operator} ${bound} && ${spread2} ${operator} ${bound})`;
+    case "===":
+      return `(${units} >= ${bound} && ${spread2} === ${bound})`;
+    case "!==":
+      return `(${units} < ${bound} || ${spread2} !== ${bound})`;
+  }
+}
+function measureCompare(measure, variable, operator, bound) {
+  const js = {
+    ">=": ">=",
+    ">": ">",
+    "<=": "<=",
+    "<": "<",
+    "=": "===",
+    "<>": "!=="
+  }[operator];
+  if (measure === "codePoints") return codePointCompare(variable, js, bound);
+  return `${measureExpression(measure, variable)} ${js} ${bound}`;
+}
 function measureExpression(measure, variable) {
   switch (measure) {
     case "codePoints":
@@ -24,7 +52,7 @@ function measureExpression(measure, variable) {
   }
 }
 function lengthCheckLabel(check) {
-  const fn = check.unit === "bytes" ? "octet_length" : "length";
+  const fn = check.fn ? check.fn === "char_length" ? "length" : check.fn : check.unit === "bytes" ? "octet_length" : "length";
   const rule = `${fn}(${check.column}) ${check.operator} ${check.value}`;
   return check.name ? `${check.name}: ${rule}` : rule;
 }
@@ -275,7 +303,13 @@ function foldDisjunctionToSet(branches, name) {
     sets: [{ column, values, kind, name }]
   };
 }
-function parseCheck(expression, name) {
+function lengthUnit(fn, dialect) {
+  const name = fn.toLowerCase();
+  if (name === "octet_length") return "bytes";
+  if (name === "char_length") return "characters";
+  return dialect === "mysql" ? "bytes" : "characters";
+}
+function parseCheck(expression, name, dialect) {
   const expr = unwrap((expression ?? "").trim());
   if (!expr) return { ok: false, reason: "empty expression" };
   if (expr.includes("?")) return { ok: false, reason: "expression contains an unresolved value" };
@@ -352,11 +386,12 @@ function parseCheck(expression, name) {
   const lengthOf = expr.match(LENGTH_OF);
   if (lengthOf) {
     const op2 = lengthOf[3] === "!=" ? "<>" : lengthOf[3];
-    const unit = lengthOf[1].toLowerCase() === "octet_length" ? "bytes" : "characters";
+    const fn = lengthOf[1].toLowerCase();
+    const unit = lengthUnit(fn, dialect);
     return {
       ok: true,
       checks: [],
-      lengths: [{ column: lengthOf[2], operator: op2, value: lengthOf[4], unit, name }]
+      lengths: [{ column: lengthOf[2], operator: op2, value: lengthOf[4], unit, fn, name }]
     };
   }
   const cardinalityOf = expr.match(CARDINALITY_OF);
@@ -863,7 +898,7 @@ function classifyTableChecks(table) {
   const out = [];
   for (const k of table.checks ?? []) {
     const expression = (k.expression ?? "").trim();
-    const parsed = parseCheck(k.expression, k.name);
+    const parsed = parseCheck(k.expression, k.name, table.dialect);
     if (!parsed.ok) {
       out.push({
         ...k.name ? { name: k.name } : {},
@@ -1844,6 +1879,7 @@ export {
   canonicalMembers,
   canonicalNumericText,
   classifyTableChecks,
+  codePointCompare,
   columnMetaFacts,
   comparisonWire,
   describeSet,
@@ -1856,6 +1892,7 @@ export {
   isProjectInstallPath,
   lengthCheckLabel,
   lengthMeasure,
+  measureCompare,
   measureExpression,
   moduleFileName,
   moduleSpecifier,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@drzl/validation-core",
-  "version": "3.22.2",
+  "version": "3.22.4",
   "private": false,
   "license": "Apache-2.0",
   "type": "module",
@@ -24,7 +24,7 @@
   ],
   "sideEffects": false,
   "dependencies": {
-    "@drzl/analyzer": "^1.21.2"
+    "@drzl/analyzer": "^1.21.4"
   },
   "peerDependencies": {
     "prettier": ">=3"