Backends

Both backends inherit all data-access logic from _EUROCONTROLBase. The only difference is how a query is executed and what type is returned.

	EUROCONTROLpy	EUROCONTROLSpark
Connection	SQLAlchemy engine (oracledb thin mode)	Oracle JDBC over Spark
Return type	pandas.DataFrame	pyspark.sql.DataFrame
Geometry methods	✓ acc_sf / ansp_sf / es_sf / fir_sf	✗ (use EUROCONTROLpy then pass GeoDataFrame to polyfill_h3)
polyfill_h3	Pure-Python h3 per row	PySpark UDF (runs on workers)
Best for	Notebooks, interactive, moderate data sizes	Large-scale production pipelines

How the shared base works

_EUROCONTROLBase (abstract)
│  _build_flights_sql(...)        ← SQL built once, used by both
│  _build_airspace_profiles_sql(...)
│  ... (all _build_* methods)
│
│  flights_tidy(...)              ← calls _execute_query(_build_flights_sql(...))
│  airlines_tidy(...)
│  ... (all public data methods)
│
│  generate_so6(trajectory)       ← pure pandas, backend-independent
│  polyfill_h3(gdf, resolution)   ← default pandas impl (overridden in Spark)
│
├── EUROCONTROLpy
│    _execute_query(sql)  → pd.read_sql(text(sql), engine.connect())
│    _execute_table(tbl)  → pd.read_sql("SELECT * FROM tbl", ...)
│    polyfill_h3(...)      → pandas (inherited default)
│    + acc_sf / ansp_sf / es_sf / fir_sf
│
└── EUROCONTROLSpark
     _execute_query(sql)  → spark.read.jdbc(url, f"({sql}) T", props)
     _execute_table(tbl)  → spark.read.jdbc(url, tbl, props)
     _select_columns(df, cols) → df.select(*cols)
     _fillna_trajectory(df) → withColumn(...when isNull...)
     airspace_profiles_tidy(...)      ← overrides with two-query Spark join
     flights_airspace_profiles_tidy(...) ← same
     polyfill_h3(...)                 ← overrides with @F.udf

Switching backends

Because both classes expose the same method signatures, you can swap them with a single line change:

# Development / notebook
from eurocontrolpy import EUROCONTROLpy
ec = EUROCONTROLpy()

# Production (Spark)
# from eurocontrolpy import EUROCONTROLSpark, build_spark_oracle_session
# spark, url, props = build_spark_oracle_session()
# ec = EUROCONTROLSpark(spark, url, props)

# Everything below works unchanged:
flights = ec.flights_tidy("2024-01-01", "2024-01-02")
so6     = ec.generate_so6(flights.toPandas() if hasattr(flights, "toPandas") else flights)