Blue CSV

A Zed editor extension that makes CSV files feel like a spreadsheet — without leaving the text buffer.

Status: pre-alpha. v0.6.0 adds markdown-table round-trip: bluecsv.toMarkdownTable renders the buffer as a GitHub-flavored pipe table, bluecsv.fromMarkdownTable parses one back. If a plan/ directory is present in your checkout, it holds design docs and roadmap notes.

What it does

Zed's extension API doesn't currently support custom file-renderers, so Blue CSV works with the text buffer you already have and makes it behave more like a grid:

• Rainbow columns — each column gets its own color, so rows and columns are scannable at a glance. Powered by tree-sitter-csv.
• Column alignment — pad fields so columns line up in a monospace view. Toggleable.
• Markdown-table round-trip — convert a,b,c ↔ | a | b | c | and back.
• Cell navigation — Tab / Shift-Tab hop between fields; commands for add-column, delete-column, duplicate-row, header-aware sort.
• Column-aware LSP — diagnostics for row-width mismatches and unterminated quotes; completions drawn from values already seen in that column; hover shows column name + row index.
• Type inference — columns are classified as int, float, date, or string. Hover on a data cell shows the inferred type; hover on a header cell shows a summary (count, distinct, min, max, sum, mean). Outlier cells in typed columns surface as diagnostics.

Supports .csv and custom delimiters.

Install

Not yet published to Zed's extensions registry. Until then, install it as a dev extension:

git clone https://github.com/anistark/bluecsv.git

In Zed, open the command palette and run zed: install dev extension, then point it at the cloned directory (the one containing extension.toml). The extension will auto-download a matching bluecsv-ls release on first use.

See CONTRIBUTING.md for the full local-dev setup.

LSP commands

Every transform is also surfaced as a code action: put your cursor in the file, open editor: toggle code actions (cmd-. by default), and pick one. Cell-scoped actions (delete column, duplicate row, sort by column) read the column / row from the cursor position.

The language server also exposes these raw workspace/executeCommand names, usable from keybindings or other clients:

Command	Arguments	Effect
bluecsv.align	[uri]	Pad every field with trailing spaces so columns line up.
bluecsv.unalign	[uri]	Strip alignment padding.
bluecsv.addColumn	[{uri}]	Append an empty column (with generated header if hasHeader).
bluecsv.deleteColumn	[{uri, col}]	Remove column at index col.
bluecsv.duplicateRow	[{uri, row}]	Duplicate row at index row.
bluecsv.sortByColumn	[{uri, col, ascending?}]	Sort rows by col; header row kept in place when hasHeader.
bluecsv.nextCell	[{uri, position}]	Request the editor move the cursor to the next cell.
bluecsv.prevCell	[{uri, position}]	Request the editor move the cursor to the previous cell.
bluecsv.toMarkdownTable	[uri]	Rewrite the buffer as a GitHub-flavored pipe table.
bluecsv.fromMarkdownTable	[uri]	Parse a pipe table back into CSV.
bluecsv.columnSummary	[{uri, col}]	Return {type, confidence, count, distinct, min, max, sum, mean, mismatchRows} as JSON.

Zed extensions can't contribute keybindings directly — Tab / Shift-Tab cell navigation will land once the extension API grows a keymap hook, or you can wire the commands above manually in your Zed keymap.json.

Alignment vs. markdown-table

Both commands make a CSV easier to read in a monospace buffer, but they serve different purposes:

• Align (bluecsv.align) pads fields with trailing spaces. The buffer is still a valid CSV — every downstream tool keeps working — and bluecsv.unalign restores the canonical form exactly.
• Markdown table (bluecsv.toMarkdownTable) rewrites the buffer as | a | b | rows with a | --- | separator. The output is no longer CSV; it's intended for previewing in Zed's markdown preview, pasting into docs, or diffing as prose. bluecsv.fromMarkdownTable reverses it.

Round-trip CSV → markdown → CSV is lossy for these cases:

• Leading / trailing whitespace inside a field is trimmed (markdown cells are whitespace-stripped).
• Fields that needed CSV quoting only for non-RFC reasons (e.g. a |) may come back unquoted — semantically equivalent, textually different.
• \r / \n inside a field are encoded as <br> in markdown and restored on the way back.

Simple canonical CSV (no padding inside fields) round-trips exactly.

Type inference

Columns are classified by scanning all non-empty cells. A column is assigned a type (int, float, date, string) when at least 90% of its non-empty cells match that type and at least 3 non-empty cells exist; otherwise it falls back to string. Numeric promotion: int cells satisfy a float column.

Recognized dates are ISO-ish: YYYY-MM-DD, YYYY/MM/DD, and YYYY-MM-DDTHH:MM:SS with optional Z or ±HH:MM suffix. Ambiguous shapes like MM/DD/YYYY are left as strings.

Settings

All settings live under the bluecsv.* namespace in Zed's settings.json (or lsp.bluecsv.initialization_options).

Key	Type	Default	Effect
hasHeader	bool	true	Treat row 0 as a header. Affects sorting, completions, inference, and hover.
alignOnSave	bool	false	Run bluecsv.align automatically on save.
inferTypes	bool	true	Enable hover type labels, header summaries, and type-mismatch diagnostics.
typeMismatchSeverity	"warning" | "hint" | "off"	"warning"	Diagnostic level for cells that don't match their column's inferred type.
maxBufferBytes	integer	10485760 (10 MB)	Size cap (in bytes) above which inferTypes and alignOnSave are skipped. Basic diagnostics (row-width, unterminated quotes) and completions still run. Set to 0 to disable the cap.

CLI

The bluecsv CLI exposes the library for use outside the editor.

bluecsv align data.csv              # pad columns to width
bluecsv unalign data.csv            # strip padding
bluecsv infer data.csv              # per-column: type, confidence, empty, mismatches
bluecsv stats data.csv 2            # min/max/sum/mean/count/distinct for column index 2

align and unalign stream the file through a two-pass algorithm when it exceeds 10 MB, keeping peak memory bounded. Override with --stream / --no-stream, or change the auto-detection threshold via the BLUECSV_STREAM_THRESHOLD environment variable (bytes). Stdin always buffers, because pipes can't be rewound for the second pass.

Contributing

See CONTRIBUTING.md.

License

MIT © Kumar Anirudha.