[Python-Dev] Make stacklevel=2 by default in warnings.warn() (original) (raw)

Nick Coghlan ncoghlan at gmail.com
Mon Sep 21 11:05:34 CEST 2015


On 21 September 2015 at 17:21, Nathaniel Smith <njs at pobox.com> wrote:

On Sep 21, 2015 12:15 AM, "Victor Stinner" <victor.stinner at gmail.com> wrote:

Would it be too verbose to display two frames or more by default? Maybe depending on the action (ex: only if the warning is emitted only once). It's not just about how it gets displayed -- the frame that it gets "attributed" to is also used when comparing against the warnings filters to determine what action to take. What if stacklevel=1 makes it match a filter with action "ignore", and stacklevel=2 makes it match a filter with action "raise"? (This is a common example I've encountered in the context of wanting to set up CI tests so that if my code uses deprecated functionality then I want to fail the test so I notice, but if some library I'm using uses deprecated functionality internally then that's not my problem. This breaks down when the library makes the common error of issuing DeprecationWarnings with stacklevel=1, because that makes python think that they are deprecating themselves, not warning that I did something deprecated.)

As Victor notes, though, that's not the right default for scripts that are issuing deprecation warnings to end users - there, they really are deprecating themselves. It's also not the right default for pretty much any warning other than DeprecationWarning - for those, you're generally wanting to say "we hit this problematic code path", not report anything about your caller.

Passing "stacklevel=2" for API deprecations is by no means obvious though, so perhaps it makes sense to add a "warnings.warn_deprecated(message)" function that's implemented as:

def warn_deprecated(message, stacklevel=1):
    return warnings.warn(message, DeprecationWarning,

stacklevel=(2+stacklevel)).

Cheers, Nick.

-- Nick Coghlan | ncoghlan at gmail.com | Brisbane, Australia



More information about the Python-Dev mailing list