laplace_sample service console output has too many newlines

[mitzimorris]

Summary:

The console output of the laplace_sample method contains hundreds of blank lines.

Description:

This arises from two things:

• messages sent to the logger include newlines which aren't needed as the callbacks writer add these, e.g.: https://github.com/stan-dev/stan/blob/e043da75516985fa8e5ed32c826e19dc9f9a0730/src/stan/services/optimize/laplace_sample.hpp#L64
• the logger is called on empty stringstreams

To fix the former, remove newlines. To fix the latter, check the stringstream before calling the logger.

 model.write_array(rng, unc_draw, draw_vec, include_tp, include_gq, &write_array_msgs);
    if (refresh > 0) {
      if (write_array_msgs.peek() != std::char_traits<char>::eof())
        logger.info(write_array_msgs);

Reproducible Steps:

Use CmdStan branch https://github.com/stan-dev/cmdstan/tree/feature/1123-laplace-approx to invoke the laplace_sample method, model and input files are in the src/test/test-models/ directory.

>  src/test/test-models/multi_normal_model laplace mode=src/test/test-models/multi_normal_mode.csv

Current Output:

the beginning of the console ouput from call to laplace_sample - refresh is default 100, so lots of whitespace between iteration messages.

Calculating Hessian

Calculating inverse of Cholesky factor

Generating draws

iteration: 
0

Expected Output:

Calculating Hessian
Calculating inverse of Cholesky factor
Generating draws
iteration: 0
iteration: 100
iteration: 200
iteration: 300
iteration: 400
iteration: 500
iteration: 600
iteration: 700
iteration: 800
iteration: 900

Additional Information:

Provide any additional information here.

Current Version:

v2.30.0

[bob-carpenter]

Thanks for figuring this out. This is why I was asking if there was a newline appended to all of the callbacks. I had thought the answer was "no", but it's apparently "yes"!

[WardBrian]

There is a longstanding issue to make the logger ignore blank lines: #2577. Partial fix was started in #3098

[bob-carpenter]

The issue arose because I assumed there wasn't a newline inserted at the end of each message. If we are going to add newlines, then we probably do not want to ignore blank lines, as they correspond to newlines in the output.

[WardBrian]

In that case we can probably close those two

[WardBrian]

Closed by https://github.com/stan-dev/stan/pull/3153