zentralopensource / zentral / 35736321662

91%

push

github

inventory: A CSV format for the full export

The export writes .csv files in a directory of the storage, like the Parquet
export, and the task result carries the same manifest. A customer who cannot
read Parquet loads the tables into a spreadsheet, a script, or a warehouse.

Each file starts with the header row. A table with many rows has more than one
file, and a reader that skips the header of each file reads them together;
without the header, one file of such a table is not a CSV a tool can open. A
table without a row has one file with the header alone, so the columns stay on
disk, as for the Parquet export.

The files follow RFC 4180: a comma separates the values, a quote in a value is
doubled, and a value with a line break is quoted. A NULL and an empty string
are both an empty field, because a CSV cannot say both; the Parquet export
keeps the difference. A boolean is true or false, a timestamp is ISO 8601 with
a Z - the database holds naive UTC values - and json and jsonb are the JSON
text, as in the Parquet export. A type the Parquet export knows and the CSV
export does not is a test failure, not an error during an export.

The two formats write their parts the same way, so the loop that saves them and
fills the manifest is shared now, and each format gives the columns of a table
and its parts. The test fixture gains a boolean value: the exports only ever
saw a NULL in a boolean column.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

77 of 77 new or added lines in 3 files covered. (100.0%)

52752 of 58037 relevant lines covered (90.89%)

0.91 hits per line