[Python][Parquet] ParquetWriter reports incorrect schema when specified at wrong order

[rongcuid]

Describe the bug, including details regarding any error messages, version, and platform.

When I try to write a table to a ParquetWriter, specifying schema in a different order gives an error:

ValueError: Table schema does not match schema used to create file: 
table:
cmdline: list<item: string>
  child 0, item: string
time: double
lock: string vs. 
file:
time: double
cmdline: list<item: string>
  child 0, item: string
lock: string

Component(s)

Parquet, Python

[mapleFU]

Table schema does not match schema used to create file

Seems this report the reason. The table should same schema as file.

If you want you can add a select to reorder the column, this could be light weight because I believe it would not copy the underlying ChunkedArray: https://arrow.apache.org/docs/python/generated/pyarrow.Table.html#pyarrow.Table.select

[rongcuid]

I think Parquet spec doesn't require the schema to have same order. Did I misunderstand the spec?

EDIT: https://issues.apache.org/jira/browse/PARQUET-188 OK, maybe I misread the spec

[rongcuid]

Though I linked to the issue, I am unable to actually find any mention of order in specs, however.

[mapleFU]

Emmmm as a file format, parquet cannot reordering, it needs to maintain all schema in a file as same. So in parquet standard, it cannot reorder within a file.

During library writing to parquet, we don't support reorder now because handling this in parquet writer might be tricky. So I think a select as a projection would be convinient, efficient and ok here.

[rongcuid]

I see one comment in a source file saying that it is ordered. Could this info be added to the API documentation?

[mapleFU]

If you think it's confusion to not allow ordering, I think we can add this info