[C++][Parquet] Forced UTF8 encoding of BYTE_ARRAY on stream::read/write

[asfimport]

StreamReader& StreamReader::operator>>(optional<std::string>& v) {
 CheckColumn(Type::BYTE_ARRAY, ConvertedType::UTF8);
 ByteArray ba;

 

StreamWriter& StreamWriter::WriteVariableLength(const char* data_ptr,
 std::size_t data_len) {
 CheckColumn(Type::BYTE_ARRAY, ConvertedType::UTF8);

 

Though the C++ Parquet::Schema::Node allows physical type of BYTE_ARRAY with ConvertedType=NONE, the stream reader/writer classes throw when ConvertedType != UTF8.

std::string is, unfortunately, the canonical byte buffer class in C++.

A simple approach might be to create >>parquet::ByteArray.. with columnCheck(BYTE_ARRAY, NONE), and let the user take it from there.  that would use the existing methods that >>std::string uses.. just an idea.

I am new to this forum, and have assigned MAJOR to this bug, but gladly defer to those who have a better grasp of classification.

Reporter: ian

_{Note: This issue was originally created as PARQUET-1958. Please see the migration documentation for further details.}

[asfimport]

Micah Kornfield / @emkornfield:
I actually am not sure that the check is needed at all.  It seems any byte array should be readable at this point.  My memory of this code is a little bit hazy, but do you think there is harm in removing the check?

[asfimport]

ian:
hi micah.  no.. it should be removed to align with schema.node checks and to support unencoded byte arrays.  thank you for following up.  best, ian.Sent from my Verizon, Samsung Galaxy smartphone
-------- Original message --------From: "Micah Kornfield (Jira)" jira@apache.org Date: 1/22/21 22:27 (GMT-06:00) To: com@ianlavie.com Subject: [jira] [Commented] (PARQUET-1958) Forced UTF8 encoding of
  BYTE_ARRAY on stream::read/write     [ https://issues.apache.org/jira/browse/PARQUET-1958?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17270548#comment-17270548 ] Micah Kornfield commented on PARQUET-1958:-----------------------------------------I actually am not sure that the check is needed at all.  It seems any byte array should be readable at this point.  My memory of this code is a little bit hazy, but do you think there is harm in removing the check?> Forced UTF8 encoding of BYTE_ARRAY on stream::read/write> ------------------------------------------------------->>                 Key: PARQUET-1958>                 URL: https://issues.apache.org/jira/browse/PARQUET-1958>             Project: Parquet>          Issue Type: Bug>          Components: parquet-cpp>    Affects Versions: cpp-1.5.0>            Reporter: ian>            Priority: Major>> ```java

StreamReader& StreamReader::operator>>(optionalstd::string& v) {>  CheckColumn(Type::BYTE_ARRAY, ConvertedType::UTF8);>  ByteArray ba;
>  > java
StreamWriter& StreamWriter::WriteVariableLength(const char* data_ptr,>  std::size_t data_len) {>  CheckColumn(Type::BYTE_ARRAY, ConvertedType::UTF8);

[asfimport]

Antoine Pitrou / @pitrou:
Indeed the UTF8 check doesn't seem necessary.