CSV Splitter

Nobody splits a file for fun

The file itself is fine. Something waiting downstream of it is not, and the list of somethings turns out to be short:

• A spreadsheet that stops counting. Excel worksheets hold 1,048,576 rows. Hand one a longer file and it loads the first million-odd rows and tells you the file was not loaded completely, which is polite for "the rest is missing". Google Sheets has its own cell budget and gives up in the same spirit.
• Import forms with row caps. Mailing tools, CRMs, ad platforms, payroll portals and payment processors all publish a per-upload limit, and the error message never says which row it stopped at. Chunks that fit under the cap are the entire fix.
• Work that has to be shared out. Six people cannot annotate the same file. Six files of equal length, one each, and the merge at the end is a concatenation.
• Batch jobs that fall over. A loader that times out or exhausts memory on one enormous file usually walks through twenty smaller ones without complaint, and a failure costs you one part instead of the whole run.
• Attachment limits. Email gateways and ticket systems bounce big files, and so do plenty of SFTP partners with a ceiling written into the contract.

The advice people are usually given is to open the thing in a text editor and cut it in half by hand. That works exactly once, and every piece after the first arrives with no column names on it.

Two ways to say where the cuts go

The Split by control offers Rows per part and Number of parts. Which one you want depends on where the constraint lives.

• Rows per part is for a limit someone else set. A spreadsheet that stops at 1,048,576 rows, an importer that refuses more than 10,000 at a time, a partner who asked for batches of 5,000. Type their number and stop thinking about it. The box starts at 100,000, which clears most import caps and opens instantly in any spreadsheet.
• Number of parts is for a limit you set. Four analysts, four files. The rows are divided as evenly as whole numbers allow, so the last part is the one that comes up short. This mode starts at 2.
• Typing is forgiving. Commas, underscores and stray spaces are stripped before the number is read, so 1,000,000 and 1000000 mean the same thing. Anything that is not a whole number of at least 1 falls back to the default, and the widget says it did rather than pretending you typed something sensible.
• Asking for more parts than you have rows is capped at one row per part, with a note. A seven-row file requested as 20 parts comes back as 7 parts.

Gotchas worth knowing

• The parts are always comma separated. A semicolon or tab file is read correctly and written back out with commas. If the delimiter has to survive the trip, run the parts through the delimiter changer afterwards.
• 500 files is the ceiling. Request a split that would go past it and the parts are made bigger until the count fits, with a message naming the new size. Six thousand rows asked for at one row each come back as 500 files of 12 rows.
• A header with nothing under it is an error, not an empty zip. The widget says the file has a header row and nothing to split, which is more useful than handing you a zip of one empty file.
• Quoted fields survive intact. A cell holding a comma, a line break or an escaped quote stays one cell, and it is re-quoted correctly on the way into whichever part it belongs to.
• The invisible marker at the start of Excel exports is removed. Left in place it attaches itself to the first column name, and that bug tends to survive several rounds of debugging because the header looks perfectly normal on screen.
• 100 MB is where this page stops. No row limit under that, and no metering of any kind. Past it the browser tab is the wrong place to be doing this, and the widget offers the full editor, which streams the file rather than holding it all at once.