logging levels

[andycasey]

The documentation advises that:

Available levels are .debug, .info, .error, and .critical. For warnings, use warnings module.

But in fact logging an .error level does nothing, logging a .critical raises an exception (trace below), and there are leftovers from when "warnings" were available.

In [3]: log.critical("what")
---------------------------------------------------------------------------
UnboundLocalError                         Traceback (most recent call last)
<ipython-input-3-53d8881f152d> in <module>()
----> 1 log.critical("what")

~/anaconda2/envs/py36/lib/python3.6/logging/__init__.py in critical(self, msg, *args, **kwargs)
   1351         """
   1352         if self.isEnabledFor(CRITICAL):
-> 1353             self._log(CRITICAL, msg, args, **kwargs)
   1354 
   1355     fatal = critical

~/anaconda2/envs/py36/lib/python3.6/logging/__init__.py in _log(self, level, msg, args, exc_info, extra, stack_info)
   1440         record = self.makeRecord(self.name, level, fn, lno, msg, args,
   1441                                  exc_info, func, extra, sinfo)
-> 1442         self.handle(record)
   1443 
   1444     def handle(self, record):

~/anaconda2/envs/py36/lib/python3.6/logging/__init__.py in handle(self, record)
   1450         """
   1451         if (not self.disabled) and self.filter(record):
-> 1452             self.callHandlers(record)
   1453 
   1454     def addHandler(self, hdlr):

~/anaconda2/envs/py36/lib/python3.6/logging/__init__.py in callHandlers(self, record)
   1512                 found = found + 1
   1513                 if record.levelno >= hdlr.level:
-> 1514                     hdlr.handle(record)
   1515             if not c.propagate:
   1516                 c = None    #break out

~/anaconda2/envs/py36/lib/python3.6/logging/__init__.py in handle(self, record)
    861             self.acquire()
    862             try:
--> 863                 self.emit(record)
    864             finally:
    865                 self.release()

~/research/projects/astra/python/astra/utils/logger.py in colored_formatter(record)
     95             message = '{} {}'.format(color_text(warning_text, ''), warning_category_colour)
     96 
---> 97     sys.__stdout__.write('{}{}\n'.format(header, message))
     98     sys.__stdout__.flush()
     99 

UnboundLocalError: local variable 'header' referenced before assignment

[albireox]

Ok, I think this is fixed. Indeed, the critical level was raising an error because it was not being handled in the colored_formatter function. That should be fixed.

I've also remove the warning leftovers and redone how the system handles the warnings. The main problem is that the warning redirection can only happen by overriding the function that warnings uses to handle the warning. astropy already does that so there can be conflicts if astropy is imported. What I've done is to disable the astropy handling of warnings and redirect everything to our _showwarning. If this still produces problems the only solution would be to give up on trying to catch warnings and use log.warnings instead.

log.error should not do anything apart from logging a message with the level CRITICAL. It's not supposed to raise an exception (unless raise or log.exception). print_exception_formatted is called when an exception is raised, but not with log.error.