Field order¶
genson keeps fields in the order they first appear in the JSON. The order is part of the output: it's the order of the columns and of the fields inside each struct.
Within one object: document order¶
An object's fields keep the order of its keys, however many keys it has:
import polars as pl
import polars_genson
keys = ["zeta", "alpha", "mu", "beta"]
row = "{" + ", ".join(f'"{k}": 1' for k in keys) + "}"
print(pl.DataFrame({"j": [row]}).genson.infer_polars_schema("j"))
Across rows: first-seen order¶
When rows have different keys, a field's position is where it was first seen, reading the rows in order:
import polars as pl
import polars_genson
df = pl.DataFrame({"j": ['{"b": 1}', '{"a": 1, "b": 2}', '{"c": 3, "a": 4}']})
print(df.genson.infer_polars_schema("j"))
b comes first because the first row has it; a appears in the second row and c in
the third. The same holds for the fields of merged records, such as the values of a map
under unify_maps (see Maps and records).
Determinism¶
The same input always gives the same field order, on every run. (Before 0.9.0, objects with more than 32 keys could come out in a different order on each run.)
Combining output from several inputs¶
Because the order depends on which rows are seen first, two inputs with the same fields can still produce different field orders: another file may simply list its keys or rows in a different order. When you combine output from several inputs, put their fields into one fixed order first, for example the order of a schema you keep for the dataset, so the combined columns and structs line up.