nx_dataset_export.nx

nx_dataset_export.nx -- IMS Thrust D R2: BYTE-FAITHFUL multi-format dataset export. ONE source dataset -> THREE serializations (CSV / JSON / TSV) that are ROUND-TRIPPABLE: parse any of the three back and you recover the IDENTICAL records (same headers, same cells, byte-exact). This is the "dataset in lots of formats to download" half of the arc -- the data a chart visualises, offered for download in the format the consumer wants, with no lossy drift between them. REUSE-vs-NEW: the UXF family (nx_uxf_tsv_migrate / nx_canon_cid) models "one source -> ONE content- addressed CID for dedup/round-trip-to-CID" -- a DIFFERENT goal (it does not emit human-readable CSV/ JSON/TSV text files a browser can download). The CSV escaping RULES here follow RFC 4180 exactly as nx_chem_report_csv established (quote a field with , " CR LF; double internal "); the JSON string escaping follows RFC 8259 as json_emit established. We implement the 3 generic emitters + 3 parsers directly over a tiny generic ROW MODEL (no per-schema code) -- the gap those organs do not fill. ROW MODEL (NxDataset): nrows records, each with ncols string fields, in flat parallel arrays hdr_ptr[c]/hdr_len[c] -- the c-th column header cell_ptr[r*ncols+c]/cell_len[..] -- the field at (row r, col c) Fields are arbitrary bytes (may contain commas, quotes, tabs, newlines). PURE: no I/O here; the caller owns buffers (sys_mmap scratch only). license_tier: ORIGINAL