ORC-1669: [C++] Deprecate HDFS support

[wgtmac]

What changes were proposed in this pull request?

Mark readHdfsFile as deprecated.

Why are the changes needed?

Reading ORC on HDFS was introduced in #134 without any test. It has not been updated for 7 years and updating libhdfspp will result in extra dependency like boost library. Staying at an old version of libhdfspp will also prohibit us from updating other libraries like protobuf.

How was this patch tested?

It does not need test.

Was this patch authored or co-authored using generative AI tooling?

No.

[wgtmac]

cc @dongjoon-hyun @stiga-huang @ffacs

[wgtmac]

Also cc @omalley in case there is any blocker to this.

[dongjoon-hyun]

As I mentioned here, Apache ORC follows Semantic Versioning, @wgtmac .

• #1864 (comment)

It's too bad. Apache ORC community cannot remove it at ORC 2.x because ORC 2.0.0 is released already and we follow Semantic Versioning policy. The best thing we can do is to deprecate it at Apache ORC 2.0.1.

Since we cannot remove it in any ORC 2.x releases, please don't claim any removal. Shall we change like the following?

- [[deprecated("readHdfsFile is deprecated in 2.0.1 and will be removed in 2.1.0")]] std::
+ [[deprecated("readHdfsFile is deprecated in 2.0.1")]] std::

[dongjoon-hyun]

cc @williamhyun , @pavibhai , @deshanxiao , too.

[wgtmac]

Does it mean that we cannot upgrade libprotobuf like #1857 in 2.x? Or at least we can choose appropriate protobuf version according to the BUILD_LIBHDFSPP option. @dongjoon-hyun

[stiga-huang]

+1 for choosing the protobuf version based on the BUILD_LIBHDFSPP option. It might have less complaints.

[dongjoon-hyun]

I'm not sure what you mean by that. May I ask what is broken in libprotobuf PR, #1857, specifically , @wgtmac ?

Does it mean that we cannot upgrade libprotobuf like #1857 in 2.x? Or at least we can choose appropriate protobuf version according to the BUILD_LIBHDFSPP option.

Apache ORC 2 has two-layer protections.

• One is orc-format which exists independently from this repo. apache/orc repository and PRs cannot break any ORC file format structure because orc_proto.proto doesn't belong to this repo. It exists in apache/orc-format without any change.

  • https://github.com/apache/orc-format/blob/main/src/main/proto/orc/proto/orc_proto.proto

• The other is that ProtocolBuffer. ProtocolBuffer is a layer to provide a language and platform neutral binary compatibility, isn't it? ORC-1662: [C++] Upgrade protobuf to 3.21.12 #1857 only changes from 3.5.1 to 3.21.12. The major version is the same. Do you mean they break anything there?

FYI, Semantic Versioning is defined here. You can do whatever you want as long as you follow the policy. In this PR, my concern is the part of a backward compatible manner of this policy. In general, Removal is not an option for backward compatibility in general. That's the reason why I told you that we can do only Deprecatation. Deprecation doesn't mean Removal.

• https://semver.org

Given a version number MAJOR.MINOR.PATCH, increment the:

MAJOR version when you make incompatible API changes
MINOR version when you add functionality in a backward compatible manner
PATCH version when you make backward compatible bug fixes

[wgtmac]

#1857 only changes from 3.5.1 to 3.21.12. The major version is the same. Do you mean they break anything there?

Yes, libhdfspp does not link with 3.21.12. For example:

[ 84%] Linking CXX executable orc-test
/usr/bin/ld: ../libs/thirdparty/libhdfspp_ep-install/lib/libhdfspp_static.a(namenode_operations.cc.o): in function `hadoop::hdfs::GetFsStatusRequestProto::~GetFsStatusRequestProto()':
/root/build/libhdfspp_ep-prefix/src/libhdfspp_ep-build/lib/proto/ClientNamenodeProtocol.pb.h:9833: undefined reference to `google::protobuf::internal::ZeroFieldsBase::~ZeroFieldsBase()'
/usr/bin/ld: ../libs/thirdparty/libhdfspp_ep-install/lib/libhdfspp_static.a(namenode_operations.cc.o): in function `hadoop::hdfs::DisallowSnapshotResponseProto::~DisallowSnapshotResponseProto()':
/root/build/libhdfspp_ep-prefix/src/libhdfspp_ep-build/lib/proto/ClientNamenodeProtocol.pb.h:23777: undefined reference to `google::protobuf::internal::ZeroFieldsBase::~ZeroFieldsBase()'
/usr/bin/ld: ../libs/thirdparty/libhdfspp_ep-install/lib/libhdfspp_static.a(namenode_operations.cc.o): in function `hadoop::hdfs::AllowSnapshotResponseProto::~AllowSnapshotResponseProto()':
/root/build/libhdfspp_ep-prefix/src/libhdfspp_ep-build/lib/proto/ClientNamenodeProtocol.pb.h:23487: undefined reference to `google::protobuf::internal::ZeroFieldsBase::~ZeroFieldsBase()'
/usr/bin/ld: ../libs/thirdparty/libhdfspp_ep-install/lib/libhdfspp_static.a(namenode_operations.cc.o): in function `hadoop::hdfs::RenameSnapshotResponseProto::~RenameSnapshotResponseProto()':
/root/build/libhdfspp_ep-prefix/src/libhdfspp_ep-build/lib/proto/ClientNamenodeProtocol.pb.h:23197: undefined reference to `google::protobuf::internal::ZeroFieldsBase::~ZeroFieldsBase()'
/usr/bin/ld: ../libs/thirdparty/libhdfspp_ep-install/lib/libhdfspp_static.a(namenode_operations.cc.o): in function `hadoop::hdfs::DeleteSnapshotResponseProto::~DeleteSnapshotResponseProto()':
/root/build/libhdfspp_ep-prefix/src/libhdfspp_ep-build/lib/proto/ClientNamenodeProtocol.pb.h:24090: undefined reference to `google::protobuf::internal::ZeroFieldsBase::~ZeroFieldsBase()'
/usr/bin/ld: ../libs/thirdparty/libhdfspp_ep-install/lib/libhdfspp_static.a(namenode_operations.cc.o):/root/build/libhdfspp_ep-prefix/src/libhdfspp_ep-build/lib/proto/ClientNamenodeProtocol.pb.h:4055: more undefined references to `google::protobuf::internal::ZeroFieldsBase::~ZeroFieldsBase()' follow
/usr/bin/ld: ../libs/thirdparty/libhdfspp_ep-install/lib/libhdfspp_static.a(ClientNamenodeProtocol.pb.cc.o):

[dongjoon-hyun]

Although we don't want to touch that file, I guess we are able to update the source code like we did in the following PR. WDTY?

• ORC-1549: [C++] Update libhdfspp.tar.gz by adding #include <cstdint> #1685

[wgtmac]

Right, I feel very reluctant to touch that file. I'll look into it later.

[dongjoon-hyun]

+1, LGTM.

[wgtmac]

Thanks @dongjoon-hyun!