Capturing logger output from Python

[tchittesh]

I'm using the python symforce.opt.Optimizer which is printing optimization info

[2025-01-10 01:44:21.428] [info] LM<sym::Optimize> [iter   21] lambda: 4.768e-03, error prev/linear/new: 0.000/0.000/0.000, rel reduction: 0.15934
[2025-01-10 01:44:21.447] [info] LM<sym::Optimize> [iter   22] lambda: 2.384e-03, error prev/linear/new: 0.000/0.000/0.000, rel reduction: 0.19686
[2025-01-10 01:44:21.467] [info] LM<sym::Optimize> [iter   23] lambda: 1.192e-03, error prev/linear/new: 0.000/0.000/0.000, rel reduction: 0.25366

Is there a way to capture this output (generated by spdlog in C++) or redirect it to stdout? Some naive approaches like

from symforce import logger as symforce_logger, LOGGING_FORMAT as SYMFORCE_LOGGING_FORMAT
symforce_logger.handlers.clear()
handler = logging.StreamHandler(sys.stdout)
handler.setLevel(logging.WARNING)
formatter = logging.Formatter(SYMFORCE_LOGGING_FORMAT)
handler.setFormatter(formatter)
symforce_logger.addHandler(handler)

don't seem to work

[aaron-skydio]

Hmm. Indeed the loggers in python and C++ are separate, and we don't currently have a way to manage the C++ logger from Python other than the functions wrapped in cc_logger.cc (which end up accessible on cc_sym, e.g. symforce.cc_sym.set_log_level). spdlog has lots of options to add sinks and modify them, if you're writing C++ you can call those spdlog functions directly. We could wrap more things in cc_logger if that would be helpful?

[aaron-skydio]

One thing we could maybe do is add a function there to change spdlog to only sink to the Python logger, and then you could do whatever you want to the Python logger without needing to wrap all the functionality of spdlog to modify things there?

[tchittesh]

Indeed the loggers in python and C++ are separate

gotcha, thanks for confirming!

add a function there to change spdlog to only sink to the Python logger

I think for the Python symforce interface, this may be ideal (closer to expected behavior at least).

[aaron-skydio]

Cool, created an issue for that here: #422