Dev.to WebDev 🛠 Dev 👁 0 📖 5 min read

Five weeks after I deleted the helper: what shipped and how many people showed up

In October I wrote about LinkUnzip, a browser extension that reads a zip's table of contents over HTTP Range requests and downloads only the files you pick. A few days later I wrote about moving the whole engine into the

In October I wrote about LinkUnzip, a browser extension that reads a zip's table of contents over HTTP Range requests and downloads only the files you pick. A few days later I wrote about moving the whole engine into the browser and deleting the Rust helper it used to need.

That was five weeks ago. Here is what has happened since, including how many people use it, which is the number I was most nervous about writing down.

The numbers

Before 1.1.0, in the week of 8 to 15 October, the Chrome Web Store showed 32 installs and a peak of 7 weekly users. The gap between those two numbers was the helper: most people installed the extension, hit the "install this .exe too" step, and stopped there. Four of my first five uninstalls were on Linux or ChromeOS, where the helper couldn't run at all.

Today the dashboard says 38 weekly users and 570 installs in total. It's still small. There was no spike and no day where it jumped, just a line that goes up a little each week, and since the helper is gone the weekly users stay close to the installs instead of lagging far behind them. That gap closing is the part I care about most.

Edge Add-ons approved the extension on 11 Nov, so it's now in both stores.

What I shipped

1.2.0 (5 November). This one was mostly things people asked for, or things I kept tripping over:

  • Password-protected zips: ZipCrypto and WinZip AES (128, 192 and 256 bit). The popup asks for the password before it starts and checks it against the first protected file, so a wrong password fails before anything is written. It stays in memory and is never saved. AES runs at about 70 MB/s in plain JavaScript, which is fine until someone points it at a 40 GB archive.
  • GitHub's green "Download ZIP" button, in the browser. That server can only send the zip from start to end, so there's no file list and no size up front. The extension reads it in one pass and compares the zip's own index with what arrived at the end.
  • Zips inside zips. A zip in the file list gets an Open button that shows its own contents, read through the outer zip with Range requests, up to four levels deep.
  • Previews of video, audio and PDF with no size limit, streamed out of the zip. A video that is stored rather than compressed can be seeked, and only the part you jump to is downloaded.

The least glamorous change was If-Range. Every Range request now names the version of the file it wants, so if a server swaps the file between my first request and my tenth, it sends the new file instead of a slice of it, and the job stops instead of writing a corrupted mix. I tested it with servers that honour If-Range, ignore it, and always refuse it. Two real ones, GitHub's and Hugging Face's CDNs, answer a partial response to every If-Range whatever the version, so for those the extension checks each answer after the fact.

1.3.0 (11 November). The extension now comes in eight languages: English, German, Spanish, French, Italian, Japanese, Brazilian Portuguese and Romanian. It follows the browser's language, and there's a switch in Settings. The English sentence is the key for each translation (395 sentences plus 19 that depend on a count), and a script fails CI if any language is missing one or changes a placeholder. Sizes are written the local way, so 925,6 MB in German and 925,6 Mo in French. I'm a native Romanian speaker, so that file is the one I can vouch for. The others could use a native reader and I'd take corrections.

Speed, measured again

I had written that extracting Blender 4.2 (5,518 files) into a folder takes 52 seconds in the browser. I re-measured and it's 55 to 68 seconds. Without writing any files it would take about 10, so the rest is Chrome's file system calls, roughly 10 ms each, and it made no difference whether 4 or 256 files were in flight. Into Chrome's Downloads it saves about 8 files a second, and more at once makes it slower.

I had one scare along the way. A test run that took 8 to 10 minutes looked like a regression. It turned out my test script was still extracting into Downloads and never switched to the folder. The code was fine and the benchmark wasn't.

If you need the fast path, the Windows helper is still there as an option in Settings, and it does Blender in 15 to 22 seconds. It's off by default and nothing needs it.

Where I've tested it

Windows, since the start. For 1.1.0 I said Mac and Linux should work but I hadn't seen it. I have now: Chrome 157 on Ubuntu 24.04 (in WSL), where every test passed twice, and macOS 15 on a GitHub Actions runner, 13 runs passed with 226 files in Downloads matching byte for byte. I didn't read any timings off the Mac. Not tested: the folder picker by hand on Linux and Mac, Intel Macs, older macOS, and Edge or Brave on Mac. If you're on one of those, I'd like to hear it either way.

What still doesn't work

  • RAR and 7z, because their formats can't be read in parts the same way.
  • Deflate64, which Windows Explorer sometimes uses for very large zips.
  • Zips with millions of files, where the file list gets heavy.
  • Firefox. The engine needs changes (no offscreen page, no folder picker, noisier downloads panel) and I want a one-day feasibility check before promising anything.

What's next

Three small things first: a list of recent zips in the popup, a queue for the zip links on a page, and an "Open in LinkUnzip" link that dataset READMEs can use so a reader lands straight in the file list. After that, Firefox.

It's free and open source, no account and no telemetry:

What's the zip you most wish you could peek into without downloading it? That's the list I'll build the preview work around.

📰 Read the original article on Dev.to WebDev

Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.