[C++, Parquet] dictionary(..., large_string) type not preserved when writing to Parquet

[mattaubury]

(tested on pyarrow-13.0.0, Linux x64)

When writing a dictionary encoded column of large_string type to Parquet file and reading it back, it is read back as a plain string type.

Repro in Python but I see the same in C++:

>>> import pyarrow as pa
>>> import pyarrow.compute as pc
>>> import pyarrow.parquet as pq

>>> strings = pc.dictionary_encode(pa.array(["foo, bar, foo"], pa.large_string()))
>>> table = pa.table([strings], ["strings"])
>>> table.schema
strings: dictionary<values=large_string, indices=int32, ordered=0>

>>> pq.write_table(table, "table.parquet")
>>> pq.read_table("table.parquet").schema
strings: dictionary<values=string, indices=int32, ordered=0>

I'd expect to get it back as a dictionary<values=large_string, indices=int32, ordered=0> type.

Component(s)

C++, Parquet, Python

[mapleFU]

Currently we're not able to read large-dictionary, and #35825 This patch is unmerged

[mattaubury]

Currently we're not able to read large-dictionary, and #35825 This patch is unmerged

Hmm, it seems to read fine, just as a non-large string.

[mapleFU]

@mattaubury Parquet Internal doesn't regard dictionary as a "type", instead, it regard it as an "encoding", that is saied, writing dictionary to parquet doesn't means the encoding is dictionary. During read, currently because of the patch is un-merged, so we're unable to read large-binary dict from parquet