fix: sql_variant rows read by their own type, no crash on a failed cell (#497); numeric strings keep a double's value (#498)

cpp/include/js/columns/number_column.h

@@ -18,8 +18,10 @@ class NumberColumn : public Column {
   NumberColumn(int id, shared_ptr<DatumStorageLegacy> storage)
       : Column(id), value((*storage->doublevec_ptr)[0]) {}
 
+  // the same text as String(n) in JS: the shortest string that reads back as the same
+  // double. to_wstring is %f, six decimal places and no exponent (1e-10 gave "0.000000", #498)
   inline Napi::Object ToString(Napi::Env env) override {
-    return AsString<double>(env, value);
+    return Napi::Number::New(env, value).ToString().As<Napi::Object>();
   }
 
   inline Napi::Object ToNative(Napi::Env env) override {

cpp/include/js/columns/result_set.h

@@ -40,6 +40,12 @@ class ResultSet {
   void add_column(size_t row_id, const shared_ptr<Column>& column);
   Napi::Object get_entry(Napi::Env env, const ColumnDefinition& definition);
   shared_ptr<Column> get_column(size_t row_id, size_t id) const;
+  // forget rows from row_id on: a row whose read failed part way
+  void drop_rows_from(size_t row_id) {
+    if (_rows.size() > row_id) {
+      _rows.resize(row_id);
+    }
+  }
   size_t get_result_count() const {
     return _rows.size();
   }

cpp/src/js/columns/result_set.cpp

@@ -11,6 +11,9 @@ shared_ptr<Column> ResultSet::get_column(const size_t row_id, const size_t colum
     return nullptr;
   }
   const auto& row = _rows[row_id];
+  if (column_id >= row.size()) {
+    return nullptr;
+  }
   return row[column_id];
 }
 

cpp/src/js/js_object_mapper.cpp

@@ -89,7 +89,9 @@ Napi::Object JsObjectMapper::fromQueryResult(const Napi::Env& env,
     auto row_array = Napi::Array::New(env, column_count);
     results_array.Set(static_cast<uint32_t>(row_id), row_array);
     for (auto c = 0; c < column_count; ++c) {
-      row_array.Set(static_cast<uint32_t>(c), resultset->get_column(row_id, c)->ToValue(env));
+      // a cell never stored (its read failed) is null, not a dereference of nothing
+      const auto column = resultset->get_column(row_id, c);
+      row_array.Set(static_cast<uint32_t>(c), column ? column->ToValue(env) : env.Null());
     }
   }
 

cpp/src/odbc/odbc_statement_legacy.cpp

@@ -259,9 +259,21 @@ bool OdbcStatementLegacy::fetch_read(const size_t number_rows) {
     const auto column_count = static_cast<int>(_resultset->get_column_count());
     for (auto c = 0; c < column_count; ++c) {
       const auto& definition = _resultset->get_meta_data(c);
-      res = dispatch(definition.dataType, row_id, c);
-      if (!res) {
-        break;
+      const auto errors = _errors->size();
+      if (!dispatch(definition.dataType, row_id, c)) {
+        // a cell that cannot be read ends the fetch with its error: drop the part of the row
+        // already read, rather than go on and leave a row with holes in it (#497)
+        _resultset->drop_rows_from(row_id);
+        if (_errors->size() == errors) {
+          // the read failed with no diagnostic from the driver: the query must still end in
+          // an error, not look complete with its remaining rows missing
+          const auto msg = "[msnodesql] cannot read column " + std::to_string(c) + " of the result";
+          const auto error = make_shared<OdbcError>("HY000", msg.c_str(), 0, 0, "", "", 0);
+          _errors->push_back(error);
+          _errorHandler->AddError(error);
+          set_state(OdbcStatementState::STATEMENT_ERROR);
+        }
+        return false;
       }
     }
   }
@@ -1505,6 +1517,12 @@ bool OdbcStatementLegacy::d_variant(
     SQL_LOG_DEBUG_STREAM("[" << _handle.toString() << "] d_variant failed to get data");
     return false;
   }
+  // a NULL is known from this first read; a second SQLGetData on the column would return
+  // SQL_NO_DATA, and the row would be lost
+  if (iv == SQL_NULL_DATA) {
+    _resultset->add_column(row_id, make_shared<NullColumn>(column));
+    return true;
+  }
   // Figure out the type
   ret = _odbcApi->SQLColAttribute(statement.get_handle(),
                                   column + 1,
@@ -1517,11 +1535,9 @@ bool OdbcStatementLegacy::d_variant(
     SQL_LOG_DEBUG_STREAM("[" << _handle.toString() << "] d_variant failed to get variant type");
     return false;
   }
-  // set the definiton to actual data underlying data type.
-  auto& definition = _resultset->get_meta_data(static_cast<int>(column));
-  definition.dataType = static_cast<SQLSMALLINT>(variant_type);
-  const auto res = dispatch(definition.dataType, row_id, column);
-  return res;
+  // read this row by its own base type. the column stays SQL_SS_VARIANT, so every row is
+  // resolved again: rows of one variant column can each hold a different type (#497)
+  return dispatch(static_cast<SQLSMALLINT>(variant_type), row_id, column);
 }
 
 bool OdbcStatementLegacy::d_time(
@@ -1940,9 +1956,9 @@ bool OdbcStatementLegacy::get_data_decimal(const size_t row_id, const size_t col
   }
 
   const auto v2 = trunc(v);
-  if (v2 == v &&
-      v2 >= static_cast<long double>(numeric_limits<DatumStorageLegacy::bigint_t>::min()) &&
-      v2 <= static_cast<long double>(numeric_limits<DatumStorageLegacy::bigint_t>::max())) {
+  // the bounds are exact as doubles: INT64_MAX is not, and rounds up to 2^63, which does not
+  // fit in an int64 (bigint max read as a double came back as INT64_MIN)
+  if (v2 == v && v2 >= -9223372036854775808.0 && v2 < 9223372036854775808.0) {
     auto bi = static_cast<DatumStorageLegacy::bigint_t>(v);
     const auto col = make_shared<BigIntColumn>(column, bi);
     if (_numericStringEnabled) {

test/contract/HANDOVER.md

@@ -1,6 +1,6 @@
 # Handover: the v6 event contract (from the 5.6.0 regressions, #478 – #481)
 
-Started 2026-10-15 from dhensby's reports. Last updated 2026-10-18. The newest state is
+Started 2026-10-15 from dhensby's reports. Last updated 2026-10-24. The newest state is
 first; the dated Progress sections below have the detail, and the older sections are history.
 
 ## Where we are — end of 2026-10-18
@@ -32,6 +32,31 @@ the contract below. Workarounds for a missing or doubled `done` (node-mssql 12.7
 break. Then decide the node-mssql guard: an RC on `next` for dhensby, node-mssql-shaped
 scenarios, or their suite as a canary job.
 
+## Progress — 2026-10-24: #497 (sql_variant) and #498 (numeric strings)
+
+Both reported by dhensby against 6.0.0; both reproduced here (ODBC 18), the crash included.
+Repros in `repro/i497.js`, `repro/i498.js`; corpus `scripts/issue-497` (+ `-ns`, agreed) and
+`scripts/issue-498` (known-bug). Unit tests in `test/params-complete-refactored.test.js`.
+
+- **#497 fixed.** `d_variant` wrote the first row's base type into the column's definition, so
+  every later row was read as that type. It now dispatches each row on its own type, and stores
+  a NULL from its first `SQLGetData` (a second read of a NULL is `SQL_NO_DATA`, which lost the row).
+- **The 0xC0000005 crash fixed.** A failed cell left a hole in the row, and building the JS rows
+  dereferenced it. `get_column` and `fromQueryResult` now give null for a missing cell, and a
+  failed cell ends the fetch: the partial row is dropped and the query ends in an error - one of
+  ours (HY000 "cannot read column N") if the driver gave none, so a failed read is never silent.
+  Still open: the rows read before it in the same fetch batch are not delivered.
+- **bigint max read as a double** came back as INT64_MIN: the int64 range check used bounds
+  that round to 2^63. Bounds exact as doubles now.
+- **#498 part 1 fixed:** a double under numeric strings is `String(n)`'s text (`"1e-10"`), not `%f`.
+- **#498 part 2 open, the owner's call:** `decimal`, `money` and `smallmoney` are still read as a
+  double, so under numeric strings digits beyond a double are lost (`decimal(38,2)` with 20
+  integer digits, large `money`); a prepared statement binds `numeric` as a double too. The fix,
+  reading `SQL_DECIMAL` as text like `SQL_NUMERIC`, changes what users get: the driver's text
+  keeps the scale (`"5.00"`) and has no leading zero (`".0000001"`, as `numeric` already does).
+- The replay keeps whole rows for scripts now (`run-one.js`), so a script's column values replay.
+  A transcript stops at 400 lines: split a script that needs more (as `issue-497-ns`).
+
 ## Progress — 2026-10-22: node-mssql against v6
 
 node-mssql's own msnodesqlv8 suite (`npx mocha --exit -t 30000 test/msnodesqlv8` in a clone,

test/contract/corpus/scripts/issue-497-ns.txt

@@ -0,0 +1,236 @@
+# scenario: scripts/issue-497-ns
+# status: agreed
+# script: test/contract/scripts/issue-497-ns.js
+== int first, numeric strings on
+int-ns: submitted
+int-ns: meta i,v
+int-ns: row
+int-ns: column 0 "1"
+int-ns: column 1 "7"
+int-ns: row
+int-ns: column 0 "2"
+int-ns: column 1 "9223372036854775807"
+int-ns: row
+int-ns: column 0 "3"
+int-ns: column 1 "1.50"
+int-ns: row
+int-ns: column 0 "4"
+int-ns: column 1 "2.25e-9"
+int-ns: row
+int-ns: column 0 "5"
+int-ns: column 1 "abc"
+int-ns: row
+int-ns: column 0 "6"
+int-ns: column 1 2020-01-02T00:00:00.000Z
+int-ns: row
+int-ns: column 0 "7"
+int-ns: column 1 true
+int-ns: row
+int-ns: column 0 "8"
+int-ns: column 1 null
+int-ns: done
+int-ns: free
+== bigint first, numeric strings on
+bigint-ns: submitted
+bigint-ns: meta i,v
+bigint-ns: row
+bigint-ns: column 0 "1"
+bigint-ns: column 1 "9223372036854775807"
+bigint-ns: row
+bigint-ns: column 0 "2"
+bigint-ns: column 1 "7"
+bigint-ns: row
+bigint-ns: column 0 "3"
+bigint-ns: column 1 "1.50"
+bigint-ns: row
+bigint-ns: column 0 "4"
+bigint-ns: column 1 "2.25e-9"
+bigint-ns: row
+bigint-ns: column 0 "5"
+bigint-ns: column 1 "abc"
+bigint-ns: row
+bigint-ns: column 0 "6"
+bigint-ns: column 1 2020-01-02T00:00:00.000Z
+bigint-ns: row
+bigint-ns: column 0 "7"
+bigint-ns: column 1 true
+bigint-ns: row
+bigint-ns: column 0 "8"
+bigint-ns: column 1 null
+bigint-ns: done
+bigint-ns: free
+== decimal first, numeric strings on
+decimal-ns: submitted
+decimal-ns: meta i,v
+decimal-ns: row
+decimal-ns: column 0 "1"
+decimal-ns: column 1 "1.50"
+decimal-ns: row
+decimal-ns: column 0 "2"
+decimal-ns: column 1 "7"
+decimal-ns: row
+decimal-ns: column 0 "3"
+decimal-ns: column 1 "9223372036854775807"
+decimal-ns: row
+decimal-ns: column 0 "4"
+decimal-ns: column 1 "2.25e-9"
+decimal-ns: row
+decimal-ns: column 0 "5"
+decimal-ns: column 1 "abc"
+decimal-ns: row
+decimal-ns: column 0 "6"
+decimal-ns: column 1 2020-01-02T00:00:00.000Z
+decimal-ns: row
+decimal-ns: column 0 "7"
+decimal-ns: column 1 true
+decimal-ns: row
+decimal-ns: column 0 "8"
+decimal-ns: column 1 null
+decimal-ns: done
+decimal-ns: free
+== float first, numeric strings on
+float-ns: submitted
+float-ns: meta i,v
+float-ns: row
+float-ns: column 0 "1"
+float-ns: column 1 "2.25e-9"
+float-ns: row
+float-ns: column 0 "2"
+float-ns: column 1 "7"
+float-ns: row
+float-ns: column 0 "3"
+float-ns: column 1 "9223372036854775807"
+float-ns: row
+float-ns: column 0 "4"
+float-ns: column 1 "1.50"
+float-ns: row
+float-ns: column 0 "5"
+float-ns: column 1 "abc"
+float-ns: row
+float-ns: column 0 "6"
+float-ns: column 1 2020-01-02T00:00:00.000Z
+float-ns: row
+float-ns: column 0 "7"
+float-ns: column 1 true
+float-ns: row
+float-ns: column 0 "8"
+float-ns: column 1 null
+float-ns: done
+float-ns: free
+== nvarchar first, numeric strings on
+nvarchar-ns: submitted
+nvarchar-ns: meta i,v
+nvarchar-ns: row
+nvarchar-ns: column 0 "1"
+nvarchar-ns: column 1 "abc"
+nvarchar-ns: row
+nvarchar-ns: column 0 "2"
+nvarchar-ns: column 1 "7"
+nvarchar-ns: row
+nvarchar-ns: column 0 "3"
+nvarchar-ns: column 1 "9223372036854775807"
+nvarchar-ns: row
+nvarchar-ns: column 0 "4"
+nvarchar-ns: column 1 "1.50"
+nvarchar-ns: row
+nvarchar-ns: column 0 "5"
+nvarchar-ns: column 1 "2.25e-9"
+nvarchar-ns: row
+nvarchar-ns: column 0 "6"
+nvarchar-ns: column 1 2020-01-02T00:00:00.000Z
+nvarchar-ns: row
+nvarchar-ns: column 0 "7"
+nvarchar-ns: column 1 true
+nvarchar-ns: row
+nvarchar-ns: column 0 "8"
+nvarchar-ns: column 1 null
+nvarchar-ns: done
+nvarchar-ns: free
+== datetime first, numeric strings on
+datetime-ns: submitted
+datetime-ns: meta i,v
+datetime-ns: row
+datetime-ns: column 0 "1"
+datetime-ns: column 1 2020-01-02T00:00:00.000Z
+datetime-ns: row
+datetime-ns: column 0 "2"
+datetime-ns: column 1 "7"
+datetime-ns: row
+datetime-ns: column 0 "3"
+datetime-ns: column 1 "9223372036854775807"
+datetime-ns: row
+datetime-ns: column 0 "4"
+datetime-ns: column 1 "1.50"
+datetime-ns: row
+datetime-ns: column 0 "5"
+datetime-ns: column 1 "2.25e-9"
+datetime-ns: row
+datetime-ns: column 0 "6"
+datetime-ns: column 1 "abc"
+datetime-ns: row
+datetime-ns: column 0 "7"
+datetime-ns: column 1 true
+datetime-ns: row
+datetime-ns: column 0 "8"
+datetime-ns: column 1 null
+datetime-ns: done
+datetime-ns: free
+== bit first, numeric strings on
+bit-ns: submitted
+bit-ns: meta i,v
+bit-ns: row
+bit-ns: column 0 "1"
+bit-ns: column 1 true
+bit-ns: row
+bit-ns: column 0 "2"
+bit-ns: column 1 "7"
+bit-ns: row
+bit-ns: column 0 "3"
+bit-ns: column 1 "9223372036854775807"
+bit-ns: row
+bit-ns: column 0 "4"
+bit-ns: column 1 "1.50"
+bit-ns: row
+bit-ns: column 0 "5"
+bit-ns: column 1 "2.25e-9"
+bit-ns: row
+bit-ns: column 0 "6"
+bit-ns: column 1 "abc"
+bit-ns: row
+bit-ns: column 0 "7"
+bit-ns: column 1 2020-01-02T00:00:00.000Z
+bit-ns: row
+bit-ns: column 0 "8"
+bit-ns: column 1 null
+bit-ns: done
+bit-ns: free
+== null first, numeric strings on
+null-ns: submitted
+null-ns: meta i,v
+null-ns: row
+null-ns: column 0 "1"
+null-ns: column 1 null
+null-ns: row
+null-ns: column 0 "2"
+null-ns: column 1 "7"
+null-ns: row
+null-ns: column 0 "3"
+null-ns: column 1 "9223372036854775807"
+null-ns: row
+null-ns: column 0 "4"
+null-ns: column 1 "1.50"
+null-ns: row
+null-ns: column 0 "5"
+null-ns: column 1 "2.25e-9"
+null-ns: row
+null-ns: column 0 "6"
+null-ns: column 1 "abc"
+null-ns: row
+null-ns: column 0 "7"
+null-ns: column 1 2020-01-02T00:00:00.000Z
+null-ns: row
+null-ns: column 0 "8"
+null-ns: column 1 true
+null-ns: done
+null-ns: free
+check ok