Data File Viewer (Parquet, JSONL, Excel)

English | 简体中文 | 繁體中文

Introduction

This extension previews three things in VS Code:

That is the whole list. Everything else this fork inherited — Word, PowerPoint, PDF, ePub, fonts, images, PSD, ICNS, XMind, HTML, Java .class, archives — was removed in 6.0.0: VS Code already covers some of them, and the rest have better dedicated readers. The rationale is in ADR-0015; the upgrade notes are in CHANGELOG.md.

This fork

This repository is a fork of cweijan/vscode-office (MIT, see NOTICE.md). It builds as GaussAA.office-viewer-plus and installs alongside the original. The fork is maintained for one person's data work, which is why it only keeps the formats that work involves.

If both extensions are installed, each one offers to open the same spreadsheet files, so VS Code picks one as the default. To choose, use Reopen Editor With… on a tab, or set workbench.editorAssociations (for example { "*.xlsx": "GaussAA.officeViewer" }). Disabling the original is the simplest way to avoid the question.

Parquet: the files upstream cannot open

Upstream bundles hyparquet without its decompressors, so a ZSTD-compressed file fails with parquet unsupported compression codec: ZSTD. This fork passes compressors from hyparquet-compressors, so SNAPPY, GZIP, BROTLI, ZSTD and LZ4 all decode.

Reading is lazy and stays usable on big files:

• One page at a time. Only the visible 50 rows are decoded, so a 377 MB file no longer materialises its whole table.
• Range requests when the host supports them. A 2-byte probe decides between ranged reads (footer plus the pages you look at) and a single full fetch; the console logs [parquet] source: ranged|full.
• Navigation for millions of rows. Jump to any row number; the footer shows the visible range (239,951–240,000 of 250,000).
• Type-aware cells. Numbers are right-aligned with tabular figures, strings, booleans and timestamps are coloured from the theme, plain BYTE_ARRAY values render as hex, nulls show as NULL, and clicking a cell opens its full value.
• Structured cells look structured. JSON arrays/objects — including list fields that Polars stores as a JSON string — are token-coloured instead of one flat string.
• URLs are links. A cell whose whole value is http(s) opens in the browser through the extension host; other schemes and URLs with credentials are never opened.
• Click a schema field to jump to its column. The table scrolls horizontally to that column and leaves a quiet tint behind, without changing rows or filters.

Find in file scans each column once with progress, marks matching rows, highlights the matching text inside cells, and jumps to the first hit. A 1M-row file scans in about 1.3 s.

The viewer follows the VS Code theme — including the antd widgets, which used to stay light under a dark view. There is no separate light/dark switch.

JSONL / NDJSON: a directory, not a grid

One JSON value per line is not a table, so the fork ships a two-pane viewer: the left pane lists entries, the right pane renders the selected entry as a tree (or as raw text).

• Bytes stay in the extension host. The webview receives summaries for the visible entries and one full value for the selected entry — never the file.
• Search across the whole file, streamed in batches with a snippet around the match and the term highlighted.
• Long values wrap instead of being clipped, and entries above 500 nodes fall back to a fixed-height virtual list so a pathological line cannot freeze the UI.
• Keyboard: ←/→ collapse/expand, Space toggles, Enter copies the value, Ctrl+Shift+C copies the path.

Spreadsheets: large files stay usable

Above 8 MB, CSV/TSV and workbook files open in a read-only windowed table: only the rows the viewport holds are in the DOM, so a 21 MB CSV paints in ~0.5 s with ~33 MB of heap where the editable grid needs seconds and hundreds of megabytes. Find scans the file in blocks, "go to row" only moves the scroll offset, and workbooks keep a sheet picker. Editing above the threshold stays in the text editor (Open in VS Code).

Verification

pnpm test        # unit tests for the parsing logic (node:test, no extra framework)
pnpm verify:ui   # headless DOM scenarios; screenshots land in .ui-verify/shots
pnpm verify:perf # byte/time budgets on large files

CI runs all three on every push, together with pnpm build and pnpm run package. Add --docs to verify:ui to refresh the screenshots above.

Known limits

• Whole-file parquet search on a very large file is I/O bound because it reads every row group it visits.
• Parquet is read-only; there is no writer.
• Spreadsheets above 8 MB open in a read-only windowed view (viewport rows only). Workbooks above 15 MB, and CSV/TSV from a virtual workspace above 25 MB, ask before loading at all; a local CSV/TSV may be windowed up to 200 MB. All workbook formats (.xlsx, .xlsm, .xls, .ods) are windowed as row text, so they are read-only above 8 MB too.
• JSONL search matches raw text, not JSON paths.
• Upstream telemetry has been removed entirely.

Git History

The extension still ships the Git History workspace (it is a commit browser, not a file preview, so the 6.0.0 narrowing did not touch it). Open it from the Source Control view, the editor title bar, the editor context menu, or the file explorer context menu.

Other features

• YAML: document outline and anchor navigation (Go to Definition for alias references)
• XML: document formatting
• Icon theme: 8 Material Icon Theme icons for spreadsheets, parquet and JSONL

Development

pnpm install
pnpm build        # production build into out/
pnpm run package  # create the .vsix (build first)
pnpm check        # format + lint + baseline + typecheck + unit tests
pnpm check:full   # check + build + smoke + verify:ui + verify:perf

Press F5 in VS Code and choose Extension to run the local build.

Credits

• Spreadsheet parsing: SheetJS/sheetjs
• Spreadsheet rendering: myliang/x-spreadsheet
• Parquet: hyparquet and hyparquet-compressors
• Material Icon theme: PKief/vscode-material-icon-theme

Sponsor

Database Client for Visual Studio Code, supporting MySQL/MariaDB, PostgreSQL, SQLite, Redis and ElasticSearch, and working as an SSH client. Get it now.