An HTTP filesystem for Linux ?

On the 28th of Dec 97, I submitted in comp.os.linux.developement.kernel the idea of building an HTTP filesystem for Linux. The main idea behin such a development is to avoid the redundancy in all the "web" aware application, and being able to manipulate easily the informations on the Web using the good old' tools.

Here is an attempt to sumarize two weeks of discussions in the newsgroup:

• first I think there is a point here, be it pros or cons, the message didn't went through unnoticed, so I guess the nechanisms of access to the Web from applications raised attention.

  I tend to conclude that people are not completely satified with the current set of tools, but I'm probably biased

• There was a few voices expressing lack of interrest. I guess the main point was related to the expected lack of directory listing capabilities. It is right that providing the readir() on an HTTP filesystem will certainly be impossible on a lot of servers, or with quite a different semantic (parsing the HTML file in case of error and returning the links contained in the pages, another possible way would be to use the Webdav extensions when available).
• There was also a number of answers expressing interest but reluctant with the idea of implementing such a big thing directly into the kernel. Most of them suggested the use to userfs allowing a program to implement the filesystem access from user space. This seems definitely a good idea at least in the first steps of an implementation. This would allow to estimate the amound of required ressources to run it and depending on the result a pure kernel implementation may or may not be done.
• Another possibility suggested in a private mail would be to extend the libc to still provide the ease of use without tempering with the kernel.
• I also end-up with the feeling that people still think that such a filesystem would be read-only allowing only content to be retrieved and not saved. While for the majority of the HTTP space one have only read access, I think that being able to save to the server hosting your Web pages is a "good feature". Otherwise I agree that except for searching data in a large amount of pages, read only access somewhat restrict the usefulness of such a filesystem.

Considering the amount of feedback received, I suggest to start a small user-level implementation based on userfs and some existing HTTP code (for example the libWWW, one of the reference implementation for HTTP/1.1).

I have created a mailing-list httpfs@rufus.w3.org for people interested, to subscribe, send a mail to majordomo@rufus.w3.org with "subscribe httpfs" in the body of the message.

Archive of the messages is available here.

$Id: index.html,v 1.2 1998/01/11 08:45:50 veillard Exp $