Wetlook World Forum

I always feel contempt for anyone who wilfully violates specifications. There is always another answer that doesn't involve lying to the client software. The worst one I saw was the MacHTTP web server which would actually deliberately send back corrupted files under certain circumstances, which was liable to crash your software and possibly the whole computer back before Mac OS X came out.

Download managers provide numerous services: a) download several files at once, b) download different segments of the same file at once, c) resume broken downloads, d) queue files so you don't have to spoon feed the downloading. MP3 stores tend to zip up all your files for you, but that's not as practical when every download is larger than a zip full of MP3s. The fact that you need a download manager is a testament to the lack of imagination on the part of browser authors: only iCab on the Mac had comprehensive download manager built in, and Internet Explorer for Macintosh had a download manager that supported resuming and queueing. The Windows world loses out hugely here.

Now, you don't have an issue with downloading multiple files at once. The problem of software grabbing multiple segments of the same file at once, should be an easy one to deal with: when an incoming Range header is received, check whether the final byte offset equals the final byte of the file. If it does, then the user agent is attempting to resume a broken download. If it isn't, the user agent is trying to download multiple portions at once. In this case, you can either return 416 Requested Range Not Satisfiable to reject the request, or do what I did at WetInRed, and simply return 200 OK together with the complete file to simply pretend that we don't understand ranges at all (may not be the right thing to do ;-) That way, you annoy greedy download managers, and permit both queueing and resuming. (The only reason this isn't all still running at WetInRed is their incorrect understanding of memory usage limited me to five concurrent PHP script executions so my passthrough script became a denial of service bottleneck. It's also probably a WTF that Linux doesn't offer a more accurate way to measure memory consumption, but that IS a difficult problem to deal with as text pages can presumably be shared by different users.)