nx_parquet.nx

nx_parquet.nx -- THE PARQUET CLI over nx_parquet_lib (search R0d be_bright: BRIGHT ships as parquet). usage: nx_parquet meta <file> nx_parquet count <file> nx_parquet show <file> <leaf name or index> <n> nx_parquet tsv <file> <leaf a> <leaf b> <out.tsv> meta prints the footer as data (schema tree with max levels, row groups, every column chunk's codec, encodings, counts and offsets) and then WALKS every BYTE_ARRAY column chunk's pages, so what it prints is what the reader decoded, never what the header claims; the last line is verdict=OK or verdict=REFUSED with a count. count the decoded entry, present-value and row counts per leaf across all row groups beside the declared rows: the known-answer test for a dataset card (BRIGHT declares num_examples per split; this must reproduce it). show the first n entries of one leaf: row, definition level, repetition level, value (a list leaf shows its elements one per entry with the row they belong to). tsv two flat leaves as <a>TAB<b> rows (tabs, newlines and carriage returns inside a cell become spaces), written to <out>.tmp and renamed into place, one row group in memory at a time; rows where either side is null are COUNTED and skipped, never written as empty strings. Exit 0 OK, 1 REFUSED (the refusal named on the last line), 2 usage. license_tier: ORIGINAL No hw writes (Rule 26).