Using logging statements for development

[carterbox]

We want to set up a system for logging information about the progress of reconstruction algorithms such as convergence criteria and which number of iterations have been completed. This system should probably use the standard logger of python and print statements for storing information optionally.

Instead of changing the API of tike to accommodate development, we can use the following tools to capture data in text files and display messages to stout.

Example

Here's an example that demonstrates these concepts.

dev_logging.py:

#!/usr/bin/env python
# -*- coding: utf-8 -*-

"""Define an example of the logging system to be used in tike."""

import numpy as np
import logging
# First set the logging level to logging.INFO.
logging.basicConfig(level=logging.INFO)
# Then we need to grab a reference to the existing logger or make a new one.
logger = logging.getLogger(__name__)

if __name__ == '__main__':
    # Go to https://pyformat.info/ for a summary of how to use format()
    print("{} is a random number.\nThis string will go to the standard output "
          "stream".format(np.random.randint(0, 100)))
    logger.info("This is an info level statement.\nLogging statements go to "
                "standard error.")
    logger.debug("This is a debugging level statement.\nIt appears when the "
                 "logging level is det to DEBUG or higher.")
    if __debug__:
        print("This statement appears if Python was unoptimized.\n"
              "i.e. if it was started without the -O option.")
    logger.info("Good bye!")

Run the script in a bash terminal:

$ python -O dev_logging.py > some_data.txt
INFO:__main__:This is an info level statement.
INFO:__main__:Good bye!

some_data.txt:

46 is a random number.
This string will go to the standard outputstream

[carterbox]

@dgursoy, This is what we talked about today. Any more thoughts?

[dgursoy]

Looks good. One concern is we need a level of control for the print statements though. Otherwise, it won't be structured and we can't use them easily afterwards. For example, if I print a "Hello World" somewhere, I may not want to have it in the txt file.

[carterbox]

I guess you could use the __debug__ switch to save files instead. For example:

if __debug__:
    np.save('./debug_data/', some_data)

But also, if the user doesn't use the -O switch, it's better to dump a bunch of text to stdout instead of making files on their disk?

I think __debug__ can only be set before runtime, so we can't protect the user by setting it for them in the code.

[carterbox]

Today I learned that the logger in a python library should be set to use the NullHandler. This is how it is set up for TomoPy and it allows the library to inherit whatever logging environment the end user has set up.

https://docs.python.org/3/howto/logging.html#library-config