[R] [C++] Allow me to write_parquet() from an arrow_dplyr_query

[asfimport]

Right now, I can:

ds <- open_dataset("some.parquet")
ds %>% 
  mutate(
    o_orderdate = cast(o_orderdate, date32())  
  ) %>% 
  write_dataset(path = "new.parquet")

but I can't:

tab <- read_parquet("some.parquet", as_data_frame = FALSE)
tab %>% 
  mutate(
    o_orderdate = cast(o_orderdate, date32())  
  ) %>% 
  write_parquet("new.parquet")

In this case, I can cast the column as a separate command and then write_parquet() after, but it would be nice to be able to us write_parquet() in a pipeline.

This will require a libarrow addition to / another version of WriteParquet that takes a RecordBatchReader instead of a fully-instantiated Table

Reporter: Jonathan Keane / @jonkeane

Related issues:

• write_parquet() / write_csv_arrow() cannot stream a dataset object back to S3 (is duplicated by)
• [C++] Allow ParquetWriter to take a RecordBatchReader as input (depends upon)

_{Note: This issue was originally created as ARROW-14428. Please see the migration documentation for further details.}

[thisisnic]

The reprex above now works, closing this ticket.