Parquetglass: Column Lab

Parquetglass opens a Parquet file and reads only its footer. It verifies the PAR1 magic at both ends of the file, reads the four-byte footer length, and decodes the Thrift compact protocol FileMetaData in pure Swift. Data pages are never decompressed or decoded; every figure on screen comes from the footer. The Summary screen reports format version, writer string, row count, row group count, leaf column count, footer length and offset, data region size, total compressed and uncompressed bytes, the overall compression ratio, and the codecs and encodings used. Footer bytes can be viewed as hex and key-value metadata opened in full. The Schema screen rebuilds the flat element list into a tree. Groups expand, LIST and MAP annotations are labelled, and each leaf shows its physical type, logical or converted type, repetition, field_id, and its max definition and repetition levels. A parquet-tools style message definition is rendered alongside. The Row Groups screen lists each row group with row count, compressed and uncompressed bytes, ratio, start offset, chunk count and sorting columns, with a bar chart of compressed versus uncompressed bytes per group; on iPad the groups appear as a table. Opening a row group shows every column chunk with its codec, encodings, value count, sizes, and dictionary and data page offsets. Each column detail aggregates its chunks across row groups: total values, null count, overall min and max decoded from the plain-encoded statistics, compressed and uncompressed bytes, share of the file, and whether a dictionary page exists. Statistics are decoded for INT32, INT64, FLOAT, DOUBLE, BOOLEAN, DATE, TIME, TIMESTAMP, DECIMAL, UUID and UTF8 columns; other byte arrays are shown as hex. The engineer can attach a flag (watch, problem, verified) and a note to any column, and a note to the file. Notes are keyed to a fingerprint of the footer, so they follow the file when it is reopened from another folder. Export produces a Markdown report or a JSON document with the same content through the share sheet or the clipboard. Files are opened from the Files app or any document provider and remembered in a Recent list through security-scoped bookmarks. Three pyarrow-written samples are bundled: nested telemetry under ZSTD, an orders table using SNAPPY with dictionary encoding plus DECIMAL and DATE columns, and a strings and dates file mixing GZIP, an uncompressed chunk and delta encodings. Encrypted footers (PARE) are reported, not decoded.