CSV Ultra

Separated values as a spreadsheet and as text, with neither view treated as the poor relation.

Open a .csv, .tsv, .tab or .psv and you get an Excel-style table: frozen column titles, in-place editing, sorting, row and column selection, columns and rows you can drag to size. Click </> in the title bar and the same tab becomes a text editor with every column in its own colour — which is the one thing that makes a forty-column CSV readable as text, because the problem was never the characters, it was telling which comma you are between.

Every edit the table makes is an edit to the file's own text. So undo, redo, the dirty dot, ⌘S, hot exit, revert, the diff view and a text editor open on the same file in the next group are all VS Code's own, and all behave exactly as they do everywhere else.

The table

• Excel-style grid. Column letters (A, B, … AA) with the file's own header names beneath them, a row number gutter carrying each row's place in the file, zebra stripes, grid lines, and numbers right-aligned the way a spreadsheet aligns them.

• Virtualized. Only what is on screen exists: a few hundred recycled boxes positioned from arithmetic, over a table of any size. A million-row file scrolls like a hundred-row one.

• Edit in place. Double-click, press Enter, press F2, or just start typing. Enter commits and moves down, Shift+Enter up, Tab right, Shift+Tab left, Escape throws it away, Alt+Enter puts a line break inside the cell (which is written back quoted, so it stays one field). Delete clears whatever is selected.

• A blank row and column past the end. Typing in them extends the file. Growing a table is the same gesture as editing one.

• Select like a spreadsheet. Click and drag for a block, click a column letter or a row number for the whole thing, ⌘/Ctrl-click to add a second band that is nowhere near the first, Shift-click to extend, ⌘A for everything. ⌘/Ctrl with an arrow jumps to the edge of the run of data you are in — the Excel rule, including skipping a gap to the next filled cell.

• Sort by clicking a column head (or double-clicking it): ascending, descending, back to the file's own order. Numbers sort as numbers, ISO dates as dates, text with a natural collator so item2 comes before item10, and blanks sink to the bottom in both directions. The header row never sorts into the data.

  A sort is a view, not a change. Nothing is written and nothing is dirty until you press Write on the sorted chip in the toolbar — right for a format whose row order is often the data itself.

• Resize anything. Drag a column's right edge or a row's bottom edge. Double-click either edge to fit it to its content. Select several columns or rows first and the edge speaks for all of them — one drag to give six columns the same width, one double-click to fit each of them to its own. Widths, heights, the sort and the header choice are remembered per file, per workspace.

• Copy and paste as a grid. Copying puts tab-separated text on the clipboard, so it lands as cells in Excel, Numbers or Sheets. Pasting reads tabs, commas or semicolons — with quotes honoured, so "Smith, John" stays one cell — and a single value with no line break is pasted as a value rather than split on the comma inside it. Copying three ⌘-clicked columns gives you those three columns side by side.

• Find in the table (⌘F/Ctrl+F) with match-case and whole-cell toggles, a running count, and F3 / Shift+F3 to step through.

• Right-click a cell, a row number or a column letter for insert, delete, sort and fit. Right-clicking inside a selection keeps it, so the menu acts on everything you picked and not on the one row the pointer landed on.

• A read-only lock. The padlock in the toolbar puts the table into look, don't touch: no cell editing, no paste, no insert or delete, no writing a sort down. Sorting, finding, resizing and copying all still work, and the right-click menu shrinks to what the table can actually do.

  It is a setting (csvUltra.readOnly), not a per-tab switch, so it stays on for every table you open until you turn it off — which is the point when you are picking your way through a production export.

• A footer that adds up. Select a block and it reports the count, how many cells are filled, and the sum, average, minimum and maximum of the numbers in it.

Rainbow columns in the text editor

Open the source and each column is painted in its own colour, cycling through ten. The status bar names the cell your cursor is in — R412 · C price — using the file's own header for the column, which is the fastest way to read a wide CSV as text.

The ten are not in spectral order, and that is the point: neighbouring columns would then get neighbouring hues, and a salmon column beside an apricot one is one column as far as the eye is concerned. Instead the hues are handed out in steps of a third of the wheel, and lightness alternates on top of that, so a column and the one beside it differ twice over.

The colours are theme colours, csvUltra.column1 … csvUltra.column10, with defaults for dark, light and both high-contrast themes. Override them where you override every other colour:

"workbench.colorCustomizations": {
  "csvUltra.column1": "#ff8ab3",
  "csvUltra.column2": "#8ad0ff"
}

Those same values reach the table's column heads, so a column is the same colour in both views of the same file. Only what is on screen is painted, so a hundred-thousand-line file costs nothing to scroll.

CSV ⇄ TSV

• Convert Delimiter…, Convert to TSV and Convert to CSV rewrite the open file in place — undoable like any other edit. Values are untouched; what changes is which of them need quotes, because a field holding a comma needs them as CSV and does not as TSV, and one holding a tab is the other way round. A search and replace gets this wrong and produces a row one field too long.

• Save a Copy as TSV… / as CSV… writes a sibling file under the matching extension and leaves the original alone. This is usually the one you want: a .csv full of tabs is a file every other tool reads wrong, so converting in place warns you and offers to finish the job.

• Read With Delimiter… changes how this tab reads the file without changing a single byte — for a file whose extension is lying.

Delimiters are detected automatically: .tsv/.tab are tab-separated and .psv pipe-separated whatever is inside them, and anything else is scored against comma, tab, semicolon, pipe and colon over the first few dozen records. The scorer parses each candidate properly, so a comma inside "Smith, John" never counts as a separator.

Writing back

A cell or a row edit rewrites only the record it touches. Every other byte of the file — the other rows' quoting, its line endings, its trailing newline, the raggedness of a file whose rows are not all the same length — is left exactly where it was, so the diff of a one-cell edit is one cell. A column insert, a written sort and a delimiter conversion touch every record and are written as one edit, which is one undo step.

csvUltra.quoteStyle decides how a rewritten field spells its quotes. preserve (the default) keeps the file's own habit: a value quoted on the way in stays quoted on the way out. minimal quotes only what has to be, always quotes everything.

Keyboard

Inside the table (the arrows and their friends are handled by the page, so they are never taken away from other editors):

Contributed, and rebindable — all of them require the table to be the active tab and the keyboard not to be in the terminal or an input box:

Settings

Prefer the text editor?

The table registers as the default editor for these extensions. To keep the text editor and reach the table only when you ask for it (Open in CSV Ultra):

"workbench.editorAssociations": {
  "*.csv": "default",
  "*.tsv": "default"
}

Privacy and safety

Nothing in the table reaches the network: no CDN, no web fonts, no telemetry, no remote origin of any kind. Your data never leaves your machine, and a CSV opens the same offline as online.

The only file the extension writes is the one you are editing — through VS Code's own edit and save, so nothing is written until you save it. Save a Copy as TSV… writes a new file, through a save dialog, and that is its only other write.

A cell holding <script> or a javascript: URL is shown as the text it is. It is never treated as markup, whoever sent you the file.