bazel build failed on nl_classifier_demo

[howl-anderson]

source: commit 1202da8

OS: Ubuntu 18.04.5

bazel: 3.7.0

c++ toolchain: gcc (Ubuntu 7.5.0-3ubuntu1~18.04) 7.5.0

comand:

bazel build tensorflow_lite_support/examples/task/text/desktop:nl_classifier_demo

error message:

tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.cc: In function 'tflite::support::StatusOr<std::unique_ptr<tflite::support::text::tokenizer::Tokenizer> > tflite::task::text::nlclassifier::{anonymous}::CreateRegexTokenizerFromProcessUnit(const tflite::ProcessUnit*, const tflite::metadata::ModelMetadataExtractor*)':
tensorflow_lite_support/cc/task/text/nlclassifier/nl_classifier.cc:136:10: error: could not convert 'regex_tokenizer' from 'std::unique_ptr<tflite::support::text::tokenizer::RegexTokenizer>' to 'tflite::support::StatusOr<std::unique_ptr<tflite::support::text::tokenizer::Tokenizer> >'
   return regex_tokenizer;
          ^~~~~~~~~~~~~~~

[SunAriesCN]

Maybe commit bdaa876a8e2e955f8cec34ab970a2bd6abd6f61c has resolve this problem.

[howl-anderson]

Hi @SunAriesCN, thank you for your suggestion, however, I just tried that commit, It doesn't work. Thank you anyway!

[SunAriesCN]

Well, I think it maybe a runtime problem, a docker image for compilation provided by TFers may help problems like this.

[howl-anderson]

Well, I compile the same target on Mac OS 10.15.7 with bazelisk instead of bazel, it works! Unfortunately, I try the same thing the same way on Ubuntu, it still not works.

[xunkai55]

Hi, could you also provide bazel version and C++ toolchain version? Thanks!

[howl-anderson]

Hi @xunkai55 I already added the bazel version and C++ toolchain version to the issue description, please take a look.

[xunkai55]

I can reproduce that bug right now. Looking into it.

[lu-wang-g]

@xunkai55 any update?

[lu-wang-g]

This issue should have been fixed by explicitly std::move regex_tokenizer to StatusOr<std::unique_ptr>. See the code.

Feel free to reopen the issue if you have further questions.