From 45a929fb85c1b9436955ead3b96d22f10af333d9 Mon Sep 17 00:00:00 2001
From: Nghia Truong <nghiat@nvidia.com>
Date: Mon, 30 Sep 2024 15:45:35 -0700
Subject: [PATCH] Fix start character

Signed-off-by: Nghia Truong <nghiat@nvidia.com>
---
 src/main/cpp/src/json_utils.cu | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/src/main/cpp/src/json_utils.cu b/src/main/cpp/src/json_utils.cu
index 06865ec81..0edebc42c 100644
--- a/src/main/cpp/src/json_utils.cu
+++ b/src/main/cpp/src/json_utils.cu
@@ -88,6 +88,14 @@ std::tuple<std::unique_ptr<cudf::column>, std::unique_ptr<rmm::device_buffer>, c
       if (is_null_literal) { return {false, true}; }
 
       auto const not_eol = i < d_str.size_bytes();
+
+      // If the first row is not null or empty, it should start with `{`.
+      // Otherwise, we need to replace it by a null.
+      // This is necessary for libcudf's JSON reader to work.
+      // Note that if we want to support ARRAY schema, we need to check for either `{` or `[`.
+      auto constexpr start_character = '{';
+      if (not_eol && d_str[i] != start_character) { return {false, true}; }
+
       return {not_eol, not_eol};
     });