Geospatial Postprocessing#

Geographic postprocessing in Superset transforms coordinate data in pandas DataFrames as part of the query postprocessing pipeline. All three functions live in superset/utils/pandas_postprocessing/geography.py and are exported via superset/utils/pandas_postprocessing/__init__.py so callers can import them from superset.utils.pandas_postprocessing.

Functions#

Each function appends or overwrites columns using _append_columns from utils.py, which calls DataFrame.assign (overwrite) or pd.concat (append) depending on whether source and target column names are identical.

Error Handling#

All three functions catch ValueError from the underlying library calls and re-raise as InvalidPostProcessingError with these messages:

• geohash_decode → "Invalid geohash string" 1
• geohash_encode → "Invalid longitude/latitude" 2
• geodetic_parse → "Invalid geodetic string" 3

Known Issue: Empty DataFrame Raises Misleading Error#

geohash_decode and geodetic_parse both fail with a misleading error when the input DataFrame has zero rows. 4

The root cause is the zip(*...) unpack pattern used by both functions 5 6:

lonlat_df["latitude"], lonlat_df["longitude"] = zip(
    *df[geohash].apply(geohash_lib.decode), strict=False
)

On an empty DataFrame, apply returns an empty Series, zip(*empty_series) produces nothing, and the tuple assignment raises ValueError: not enough values to unpack. The except ValueError block then surfaces this as InvalidPostProcessingError("Invalid geohash string") — even though no geohash string was malformed. 4

The same structural bug exists in geodetic_parse 7, which would raise "Invalid geodetic string" under the same condition.

Effect: Any chart using geohash-decode or geodetic-parse postprocessing will show a data-format error message when the underlying query returns zero rows, rather than rendering a "no data" state. This causes debugging effort to be misdirected toward data quality rather than query filters. 4

Suggested fix: Guard with an early return of the input DataFrame before the zip unpack when df is empty. This reserves the ValueError/InvalidPostProcessingError path for genuine decode failures. No merged fix exists as of this writing. 4

Testing#

Unit tests live in tests/unit_tests/pandas_postprocessing/test_geography.py. They cover the happy path for all three functions using the lonlat_df fixture 8 but do not cover empty DataFrame inputs or error paths — the known empty-DataFrame bug is untested.

Related Files#