fix(stat_summary_bin): emit the fallback warning instead of building it

[Rodrigo-Palma]

The bug

stat_summary_bin.setup_params builds a PlotnineWarning and throws it away:

def setup_params(self, data):
    keys = ("fun_data", "fun_y", "fun_ymin", "fun_ymax")
    if not any(self.params[k] for k in keys):
        PlotnineWarning(                                  # instantiated, never raised
            "No summary function, supplied, defaulting to mean_se()"
        )
        self.params["fun_data"] = "mean_se"

There is no warn(...), and the module does not even import it, so the fallback to mean_se() happens in complete silence:

>>> s = stat_summary_bin(bins=5)
>>> s.setup_params(df)
before   : None
warnings : []
after    : mean_se

It is the only instantiated-and-discarded warning or exception in plotnine/:

$ grep -rnE "^\s+(PlotnineWarning|PlotnineError|UserWarning)\(" --include="*.py" plotnine/
plotnine/stats/stat_summary_bin.py:110:            PlotnineWarning(

Why it is an oversight and not a decision

Three independent signals:

1. The commit that wrote the line. 827fd09 ("Add default function to stat_summary_bin", closes stat_summary_bin no default summary function #528) deliberately turned an error into a warning, switching the import along with it:

   -from ..exceptions import PlotnineError
   +from ..exceptions import PlotnineWarning
   ...
   -            raise PlotnineError('No summary function')
   +            PlotnineWarning(
   +                "No summary function, supplied, defaulting to mean_se()"
   +            )
   +            self.params['fun_data'] = 'mean_se'

   The message was written on purpose; only the warn() call never made it in. The raise that used to make this visible went away and nothing took its place.

2. The sibling. stat_summary.setup_params still raises on the very same condition (identical keys tuple and if not any(...) block), and every other warning in plotnine/stats/ goes through warn(msg, PlotnineWarning).

3. ggplot2 informs here. R/stat-summary-bin.R: cli::cli_inform("No summary function supplied, defaulting to {.fn mean_se}"). The wording in plotnine is the same sentence, stray comma aside.

What it costs in practice: someone who mistypes the argument, fun= instead of fun_y= coming from ggplot2 3.4+, gets error bars nobody asked for and no hint that a default was substituted.

The change

Call warn, and drop the stray comma while the line is being touched, which makes the message read exactly like ggplot2's.

Test

test_no_summary_function_warns asserts the warning via p._build(), the same way tests/test_stat.py does.

Before the fix:

E       Failed: DID NOT WARN. No warnings of type (<class 'plotnine.exceptions.PlotnineWarning'>,) were emitted.
1 failed

After: 4 passed in that file.

Full suite with the fix: 906 passed, 5 skipped, 1 xpassed, including tests/test_lint_and_format.py. The new warning surfaces in the existing test_setting_binwidth, which is exactly the documented fallback and does not fail anything, since the project does not set filterwarnings = error.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 88.50%. Comparing base (cb81719) to head (3d6b7bc).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1129   +/-   ##
=======================================
  Coverage   88.50%   88.50%           
=======================================
  Files         226      226           
  Lines       16613    16614    +1     
  Branches     2130     2130           
=======================================
+ Hits        14703    14704    +1     
  Misses       1295     1295           
  Partials      615      615           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[has2k1]

@Rodrigo-Palma, thank you.