Apply suggestions from reviews

cpp/src/io/parquet/page_decode.cuh

@@ -149,13 +149,15 @@ inline __device__ bool is_bounds_page(page_state_s* const s,
   size_t const begin      = start_row;
   size_t const end        = start_row + num_rows;
 
-  // For non-nested schemas, rows cannot span pages, so use a more restrictive test except for
-  // the page_end. This is because we may have adjusted the `num_rows` for the last page in
-  // `generate_list_column_row_count_estimates()` to compensate for the list row size estimates
-  // in case of chunked read mode.
+  // For non-nested schemas, rows cannot span pages, so use a more restrictive test. Make sure to
+  // relax the predicate for page_end if we adjusted the `num_rows` for the last page to compensate
+  // for list row size estimates in `generate_list_column_row_count_estimates()` when chunked read
+  // mode.
   return has_repetition
            ? ((page_begin <= begin && page_end >= begin) || (page_begin <= end && page_end >= end))
-           : ((page_begin < begin && page_end > begin) || (page_begin < end && page_end >= end));
+           : ((page_begin < begin && page_end > begin) ||
+              (page_begin < end &&
+               (s->page.is_num_rows_adjusted ? (page_end >= end) : (page_end > end))));
 }
 
 /**

cpp/src/io/parquet/parquet_gpu.hpp

@@ -310,8 +310,10 @@ struct PageInfo {
   // - In the case of a nested schema, you have to decode the repetition and definition
   //   levels to extract actual column values
   int32_t num_input_values;
-  int32_t chunk_row;  // starting row of this page relative to the start of the chunk
-  int32_t num_rows;   // number of rows in this page
+  int32_t chunk_row;          // starting row of this page relative to the start of the chunk
+  int32_t num_rows;           // number of rows in this page
+  bool is_num_rows_adjusted;  // Flag to indicate if the number of rows of this page have been
+                              // adjusted to compensate for the list row size estimates.
   // the next four are calculated in gpuComputePageStringSizes
   int32_t num_nulls;       // number of null values (V2 header), but recalculated for string cols
   int32_t num_valids;      // number of non-null values, taking into account skip_rows/num_rows

cpp/src/io/parquet/reader_impl_preprocess.cu

@@ -730,6 +730,7 @@ struct set_final_row_count {
     size_t const page_start_row = chunk.start_row + page.chunk_row;
     size_t const chunk_last_row = chunk.start_row + chunk.num_rows;
     page.num_rows               = chunk_last_row - page_start_row;
+    page.is_num_rows_adjusted   = true;
   }
 };