Fix converter, formatter, logging and import errors

[wolph]

Found by an adversarial pass over the modules that are unchanged since 4.0.1. Every fix started as a failing test. Only errors whose fix cannot break working code are in here. The rest is listed at the bottom for a decision.

The branch is based on #53, so the test hooks pass on any machine. Merge #53 first, or merge this and #53 is included.

Logging

class Spam(Logged):
    def eggs(self):
        try:
            1 / 0
        except ZeroDivisionError:
            self.exception('boom')    # no traceback in the log
        self.info('hello')            # record says logger.py:info

class Bacon(Logurud): ...
Bacon().info('payload {"a": 1}')      # KeyError: '"a"'

class LoggedInt(Logged, int): ...
LoggedInt(5)                          # 0

• exception() logs the traceback.
• Records name the caller instead of logger.py and the wrapper.
• Logurud accepts a message with a brace in it. An explicit extra still reaches the loguru record.
• __new__ forwards its arguments, so LoggedInt(5) is 5. A base __new__ that does not take them is called without, as before.

Converters

• to_int('abc123', regexp=r'\d+') and the to_float equivalent raised IndexError for a pattern without a group. They use the whole match now.
• scale_1024(0.0001, 9) returned the power -1. It is clamped at 0.

Formatters

Call	Before	After
timesince(timedelta(days=30))	1 month and 2 days ago	1 month ago
timesince(timedelta(days=365))	1 year and 1 day ago	1 year ago
timesince(timedelta(seconds=-1))	12 months and 4 weeks ago	1 second ago
timesince(aware_datetime)	TypeError	works
camel_to_underscore('HTTP_OK')	htt_p_ok	http_ok
camel_to_underscore('HTML5Parser')	htm_l5_parser	html5_parser

Import

import_global('email.mime.text') raised ImportError unless other code had already imported that submodule. A nested module that exists and fails to import, for example on a missing dependency, shows its own error.

Docstrings corrected, no behaviour change

• to_float uses the first group, and its default is the int 0.
• to_str returns bytes. Its examples showed str and only passed because of ALLOW_BYTES.
• regexp=True drops a sign, a leading decimal point and an exponent.
• remap is only exact when every value with a fraction is a Decimal.
• import_global takes a list of names for modules.

Compatibility

What working code can notice:

• Code that passes stacklevel=2 to a Logged method to work around the wrong caller now points one frame too high.
• A loguru record no longer carries exc_info, stack_info and stacklevel in its extra.
• timesince and camel_to_underscore return a different string for the inputs in the table.
• Logged.exception() outside an except block logs NoneType: None, as logging does.
• A Logurud message without arguments is logged as written. Code that doubled its braces to avoid the old crash now sees the doubled braces.
• to_int and to_float with a group-less pattern and an exception argument that caught the old IndexError return the matched value where they returned the default.

A second adversarial pass compared this branch with 4.0.1 and found three regressions in its first version. All three are fixed and pinned by tests:

• The logger of a Logged class is created at its first instance, as in 4.0.1. Creating it when the class is defined let dictConfig with its default disable_existing_loggers switch it off.
• A subclass inherits a logger from the class body of its parent.
• A base class whose __new__ takes no arguments still works next to Logged.

The same pass found nothing that went from right to wrong across 45150 to_int and to_float combinations, 213008 scale_1024 combinations, every timesince day count from 0 to 2000 and 66853 generated identifiers for camel_to_underscore. The caller in the log record is right in 49 call shapes on CPython 3.10 to 3.14 and PyPy 3.10 and 3.11.

Found and left alone

These change what working code sees, so they need a decision:

• MyClass.info(...) on a Logged class raises AttributeError before the first instance, and a subclass logs under the name of its parent until then. A lazy logger attribute would fix both.
• camel_to_underscore('IOError') gives ioerror while MyIOError gives my_io_error.
• import_global(..., modules=[]) imports everything.
• apply_recursive does not recurse into a dict inside a list and never passes its **kwargs on.
• python_utils.__all__ lacks containers and exceptions, and logger.__all__ lacks LoggerBase.
• get_terminal_size returns width 0 for COLUMNS=1.
• from python_utils.types import * fails on Python 3.15.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, add credits to your account and enable them for code reviews in your settings.