Reading and writing files

The two things every user of this library does: read a parquet file's columns into Fortran arrays, and write Fortran arrays out as a parquet file — plus the reference for choosing which rows a read returns and in what order (filtering, sorted order, random downsampling).

Parquet in, parquet out. This library reads and writes parquet and no other format; any conversion — FITS, CSV, HDF5, anything else — happens outside it, before the library is called or after it returns. It is a settled scope decision rather than a gap waiting to be filled, stated in full under Important behavior in the README.

• Reading parquet files from your Fortran code — open a file, read whole columns, random access by row or element, shape/type queries, prefetching, reading a file's table metadata, and chunked reads.
• Filtering, sorting and sampling rows — filter=, sort_by= and sample_fraction=: the rule grammar, three-valued Null logic, row-group pruning, and how the three compose.
• Writing parquet files from your Fortran code — open a writer, write columns, declare the file's contents with a MAML schema, writer options and compression, row groups and chunked writes, and dropping rows on the way out with a row mask.

Prefer whole files over single columns? The tables group is the shortcut: one parquet_open_table call presents the whole file as a parquet_table, with every column available by name and read only when touched.