format_exc_info should add structured data about the exception to the event_dict

[wbolster]

The stdlib's Logger class has a .exception() method that is basically .error(..., exc_info=True). When porting an existing application that used logger.exception("oops something went wrong") in many exception handlers, all this code had to be updated to logger.error("oops...", exc_info=True).

Since the whole point of structlog.stdlib.BoundLogger is that it emulates the stdlib's Logger API, would you consider adding an .exception() method?

[wbolster]

Fwiw, the stdlib's Logger also logs the original exception message, regardless of the message passed to the .exception(...) method.

[wbolster]

Okay, it seems this is already (mostly) addressed by #22.

The stdlib's .exception() also includes the original exception message; I can't quickly see whether structlog's emulation does that as well?

[hynek]

Please try and file a bug if it doesn’t. (respectively transform this ticket into that in that case to make use of it :)).

[hynek]

(I’m not 100% sure what you mean so I figure it’s faster if you just test it before I start asking clarification questions and test it then to tell you)

[wbolster]

This Python file:

logging.basicConfig()
logger = logging.getLogger(__name__)

try:
    raise ValueError("This is the exception message.")
except ValueError:
    logger.exception("This is the log message.")

shows this output:

ERROR:__main__:This is the log message.
Traceback (most recent call last):
  File "blah.py", line 8, in <module>
    raise ValueError("This is the exception message.")
ValueError: This is the exception message.

As you can see, both the log message and the exception message show up in the log. A similar setup with structlog loses the exception message.

[wbolster]

The structlog configuration is from the docs:

    structlog.configure(
        processors=[
            structlog.stdlib.filter_by_level,
            structlog.processors.StackInfoRenderer(),
            structlog.processors.format_exc_info,
            structlog.processors.JSONRenderer(),
        ],
        context_class=dict,
        logger_factory=structlog.stdlib.LoggerFactory(),
        wrapper_class=structlog.stdlib.BoundLogger,
        cache_logger_on_first_use=True,
    )

[hynek]

So, I’ve tested it with

import logging

import structlog


logging.basicConfig()

structlog.configure(
    processors=[
        structlog.stdlib.filter_by_level,
        structlog.processors.StackInfoRenderer(),
        structlog.processors.format_exc_info,
        structlog.processors.JSONRenderer(),
    ],
    context_class=dict,
    logger_factory=structlog.stdlib.LoggerFactory(),
    wrapper_class=structlog.stdlib.BoundLogger,
    cache_logger_on_first_use=True,
)

logger = structlog.get_logger()

try:
    raise ValueError("This is the exception message.")
except ValueError:
    logger.exception("This is the log message.")

And the reformatted output I’m getting is:

{
  "exception": "Traceback (most recent call last):\n  File \"t2.py\", line 22, in <module>\n    raise ValueError(\"This is the exception message.\")\nValueError: This is the exception message.",
  "event": "This is the log message."
}

So the error message clearly is there (the stdlib output is completely within the formatted string), but what you’re saying is that you’d like the exception data in a more structured form?

Right now, format_exc_info really just formats the traceback: https://github.com/hynek/structlog/blob/master/structlog/processors.py#L169

It’s totally thinkable to pull data out of the exception I guess? I’m removing the milestone though because this is a nice but not too important feature (and you can implement it for yourself for now if you want to).

[wbolster]

Okay maybe I was just being blind. The original message is part of the formatted traceback, right?

[hynek]

Yes, it should. Scroll to the right above to verify. :) The job of format_exc_info is to emulate the output you would have gotten on the console.

[wbolster]

In that case I must've missed
it the first time in the huge json spaghetti on my screen. I'm sorry for the noise.

Hynek Schlawack notifications@github.com schreef op 10 januari 2015 09:40:00 CET:

Yes, it should. Scroll to the right above to verify. :) The job of
format_exc_info is to emulate the output you would have gotten on the
console.

---

Reply to this email directly or view it on GitHub:
#31 (comment)

Sent from my Android device with K-9 Mail. Please excuse my brevity.

[wbolster]

I think extracting the actual error message from a stack trace is not that useful anyway. The message will likely not be that useful without the additional info from the stack trace, so I guess the current output is fine.

Closing this issue.