How to handle failed write and how should terminate be used?

[yun-li]

I have an C++ MCAP writer who writes mcap message. After a message is written, it needs to check if the write is successful. If not, we need to call terminate, can we write more messages after the terminate call?

For example, can we do think like

for (int i = 0; i < 10; i++) {
    const auto res = writer.write(message[i])
    if (!res.ok()) {
      std::cerr << "Failed to write message: " << res.message << "\n";
      writer.terminate(); // do not close the writer, and continue writing the file in the next for-loop
    }
}

Does that work? Or do we need to call writer.close() after writer.terminate()?
Can we continue writing message to the same file by reopen the file after it terminates? Or do we need to open a new file?

I know according to this example https://github.com/foxglove/mcap/blob/main/cpp/examples/protobuf/writer.cpp#L155
When the write failed,

    const auto res = writer.write(msg);
    if (!res.ok()) {
      std::cerr << "Failed to write message: " << res.message << "\n";
      writer.terminate();
      writer.close();
      std::ignore = std::remove(outputFilename);
      return 1;
    }

It closes every thing and terminate the program, which is not what we want.

Thanks!

[james-rms]

The IWritable interface does not communicate issues that happen in the filesystem write() call back to the caller. The default FileWriter implementation that comes with the MCAP C++ library simply asserts that the write succeeded, which will cause your program to terminate early in debug mode, or fail silently in release mode if there is a write error.

The merits of this approach are debatable, but the bottom line today is that if you want to handle file write errors, you need to implement your own IWritable which communicates write errors to the application through a side effect or a callback.

The purpose of the Writer::write() return value Status is to communicate errors around the way the Writer is being used - for example, if you write a message before calling open() or before writing the schema for that message. Errors like this are not recoverable - the right thing is usually to close the file, terminate the application, then later change the code that used the Writer API in a way which caused an error.

[yun-li]

OK. Thanks!

Do you recommend using try-catch block to catch the error, and then reopen the file for usecases where application failure is not acceptable?

With the write error, would that MCAP file still decodable after calling terminate and close?

Thanks!