Locale issue in profiler/trace logger

[mvieth]

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): yes, but bug is not in custom code
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 20.04.2
• Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device: no
• TensorFlow installed from (source or binary): pip3
• TensorFlow version (use command below): v2.4.0-49-g85c8b2a817f 2.4.1
• Python version: 3.8.5
• Bazel version (if compiling from source): no
• GCC/Compiler version (if compiling from source): no
• CUDA/cuDNN version: -
• GPU model and memory: -

Describe the current behavior
The trace.json written by the TensorBoard callback is locale dependent. Example that was written while a German locale was set:

"ts":45,805999999999997,"dur":318,03199999999998

Describe the expected behavior
The decimal commas should be decimal points to comply with the JSON syntax.
I believe that snprintf here should write a decimal point, no matter which locale is set.
Also for reference: MaxPrecision is called from here, and PicosToMicrosString is called from here.

As a side note, it is unnecessary to write so many digits/decimal places. Looking at PicosToMicros, there can be no more than six non-zero digits after the decimal point, in my example above seemingly only three.

Contributing

• Do you want to contribute a PR? (yes/no): inclined to contribute, but don't know enough about tensorflow to know what solution would be best
• Briefly describe your candidate solution (if contributing): not sure

Standalone code to reproduce the issue
This example should reproduce the issue if a locale without decimal point is set.

Thank you!

[stephanwlee]

Thanks a lot for pointer to the code, repro steps, and very clear description. Assigning it to @yisitu, owner of https://github.com/tensorflow/profiler.