March 28, 2023

Tabular knowledge is in every single place. I help studying and writing tabular knowledge in varied codecs in all 3 of my software program software. It is a vital a part of my data transformation software. However all of the tabular knowledge codecs suck. There doesn’t appear to be something that’s fairly area environment friendly, easy and fast to parse and textual content based mostly (not binary) so you possibly can view and edit it with a typical editor.

Most tabular knowledge at present will get exchanged as: CSV, Tab separated, XML, JSON or Excel. And they’re all extremely sub-optimal for the job.

CSV is a large number. One quote within the improper place and the file is invalid. It’s troublesome to parse effectively utilizing a number of cores, because of the quoting (you possibly can’t begin parsing from half approach by way of a file). Totally different quoting schemes are in use. You don’t know what encoding it’s in. Use of separators and line endings are inconsistent (generally comma, generally semicolon). Writing a parser to deal with all of the totally different dialects is by no means trivial. Microsoft Excel and Apple Numbers don’t even agree on interpret some edge instances for CSV.

Tab separated is a bit higher than CSV. However can’t retailer tabs and nonetheless has points with line endings, encodings and many others.

XML and JSON are tree constructions and never appropriate for effectively storing tabular knowledge (plus different points).

There’s Parquet. It is rather environment friendly with it’s columnar storage and compression. However it’s binary, so can’t be considered or edited with normal instruments, which is a ache.

Don’t even get me began on Excel’s proprietary, ghastly binary format.

Why can’t we’ve a format the place:

  • Encoding is all the time UTF-8
  • Values saved in row main order (row 1, row2 and many others)
  • Columns are separated by u001F (ASCII unit separator)
  • Rows are separated by u001E (ASCII report separator)
  • Er, that’s your complete specification.

No escaping. If you wish to put u001F or u001E in your knowledge – robust you possibly can’t. Use a distinct format.

It might be fairly compact, environment friendly to parse and simple to manually edit (Notepad++ reveals the unit separator as a ‘US’ image). You would write a quick parser for it in minutes. Typing u001F or u001E in some editors could be a faff, however it’s hardly a showstopper.

It could possibly be known as one thing like “unicode separated worth” (hat tip to @fakeunicode on Twitter for the title) or “unit separated worth” with file extension .usv. Possibly a distinct extension might used when values are saved in column main order (column1, column 2 and many others).

Is there nothing like this already? Possibly there’s and I simply haven’t heard of it. If not, shouldn’t there be?

And sure I’m conscious of the related XKCD cartoon ( ).

** Edit 4-Might-2022 **

“Javascript” -> “JSON” in para 5.

It has been pointed on the above will provide you with a single line of textual content in an editor, which isn’t nice for human readability. A fast repair for this may be to make the report delimiter a u001E character adopted by an LF character. Any LF that comes instantly after an u001E could be ignored when parsing. Any LF not instantly after an u001E is a part of the info. I don’t find out about different editors, however it’s simple to view and edit in Notepad++.