Had this crazy idea the other day to make an improved file uploader using just a localhost little webserver that websites can check to see if it exists. If it does they can punch people over to their uploader, and then it'll let them pick a local file and push it to an S3 bucket.

This a total hack to just see if it works. That means if you want to use it you have to do some hacking on this little bit of code and do some yak shaving. Don't ask me for help because I sort of don't care about it, it's just a random idea I cranked out.

I have it on good authority, after showing this to a friend, that at least one company almost has this feature, but they're too stupid to realize they could solve the "upload problem". Let's hope they read this, and then get their head out of their ass to make it a reality.

But, with that being said, I came up with this independently and then was told they had it.

Basically the receiver site creates an iframe to localhost:8080 and if it loads it sends a pushMessage to it. The poc-uploader is then a little Python app that hangs out there and when it gets a pushMessage, sends a reply to tell the parent site it exists.

After this handshake, the user can click a link and go to the localhost:8080 app to pick a file and upload it. Currently it just uploads to a fixed S3 bucket but that information could be passed along with the initial push messages. It also is just picking files in the local uploads directory in poc-uploader/uploads, but since it's localhost you could do almost anything. That's also kind of scary so...yeah.

Once the poc-uploader has pushed it to S3 it then tell the user and sends them back to the main site.

Pretty simple, and it lets the parent site find out if the user has the uploader installed and tell them to install it or run it.

Of course, to make this generally useful you'd need some more features and a good security review, but this is the general first idea.

I've opened up a bucket on S3 so people can try it out. If it gets abused I'm going to close it up. Oh, I'm sure it'll last a day but here's how to try it.

1. git clone git://github.com/zedshaw/poc-uploader.git
2. cd poc-uploader
3. pip install lpthw.web boto
4. python app.py
5. Go to http://zedshaw.com/uploader.html
6. Do what it says, maybe it'll work.

If it blows up then you have to setup your own S3 because I disabled it.

I'm assuming you can get the code with git. After you do that do this:

1. Get an Amazon S3 account.
2. Create a bucket where you want the test files to go.
3. Change the permissions on the bucket so that it is Upload/Delete and List accessible to everyone. Yes, this sucks but it's just a test. Talk to the boto guys about why they try to list a bucket and explode violently when they can't.
4. Open the app.py file and change the bucket_name variable to use the name of your bucket.

Now that you've got that working, do this:

1. python app.py
2. http://zedshaw.com/uploader.html
3. Go through it again.
4. File should be in your S3 account, even though you were on my site.

Put the file templates/receiver.html on your site and just hit it with a browser.

What? Heellllsssss no, not right now. It should be alright since the browser can't touch what's inside the iframe, but you never know what some jackass can figure out or what all browsers allow.

I would recommend in the future that this locally running server use some Desktop notifications and confirmations to make sure that rogue sites can't force uploads.

I'd also say that having an uploader like this is only half of solving the "upload problem" on the web. If you can combine this with the following features you'll have something:

1. Chunking the files into 4M pieces and storing them in the bucket as a bunch of chunks. Then sending a metadata to the target site so they know how to rebuild it.
2. Adding encryption on the uploaded file chunks.
3. Solid status notifications from the uploader to the site on progress.
4. Making it work reliably with crap internet that goes down. If you have 4M chunking, good encryption on each chunk, and a way to check what's been uploaded, then you've got the basis of reliable async uploads.
5. Doing all this crap for target sites so they just get pure uploads with nothing to worry about.
6. Improved native minimalist GUI for managing the app. Let the web browser do all the actual picking so it fits with the websites, but having a good installer, a tray notifier, and control panel will make it easier to do other stuff.

I just did, now I'm bored. I'd rather write books and play guitar than try to figure out how to scale an uploader service.

It's copyright, by me. You can't use this code in jack shit, but you can steal the idea all you want. Hurry though because Apple might patent it and sue you.