MINOR: [C++][CSV] reject blocks whose value count overflows int32

Author: metsw24-max

cpp/src/arrow/csv/parser.cc

@@ -605,6 +605,17 @@ class BlockParserImpl {
           rows_in_chunk = std::min(kTargetChunkSize, max_num_rows_ - batch_.num_rows_);
         }
 
+        // The values array holds one ParsedValueDesc per cell and those offsets
+        // are 31-bit, so the number of values in a chunk must fit in an int32.
+        // A first line with millions of fields can drive `num_cols_` high enough
+        // to overflow that, so error out rather than presize past the limit.
+        if (static_cast<int64_t>(rows_in_chunk) * batch_.num_cols_ >
+            std::numeric_limits<int32_t>::max()) {
+          return Status::Invalid("CSV parser: row group of ", rows_in_chunk, " rows x ",
+                                 batch_.num_cols_,
+                                 " columns exceeds the maximum number of values");
+        }
+
         ARROW_ASSIGN_OR_RAISE(
             auto values_writer,
             PresizedValueDescWriter::Make(pool_, rows_in_chunk, batch_.num_cols_));

cpp/src/arrow/csv/parser_test.cc

@@ -666,6 +666,17 @@ TEST(BlockParser, MismatchingNumColumns) {
   }
 }
 
+TEST(BlockParser, TooManyValues) {
+  // A first line carrying millions of fields drives num_cols high enough that
+  // the per-chunk value count (rows x columns) would overflow the 31-bit value
+  // offset, so the parser errors out instead of overflowing.
+  uint32_t out_size;
+  BlockParser parser(ParseOptions::Defaults(), /*num_cols=*/5000000);
+  Status st = Parse(parser, MakeCSVData({"a,b\n"}), &out_size);
+  EXPECT_RAISES_WITH_MESSAGE_THAT(
+      Invalid, testing::HasSubstr("exceeds the maximum number of values"), st);
+}
+
 TEST(BlockParser, MismatchingNumColumnsHandler) {
   struct CustomHandler {
     operator InvalidRowHandler() {