I checked in that patch.  Marking this post-0.6.0 now, updating title.

I am inclined to fix this bug by adding a suggestion to the documentation that one can read data directly from a URL by using a pipe file name that uses a helper program, e.g. "|curl http://janda.org/c10/data%20sets/vote92/vote92.por".  A patch that implements this, and fixes a few bugs I encountered while testing it, is attached.

I considered two alternatives:

• Link against curl or libwww (from w3c), as jmd suggested.  Long term, this might be a good solution.  The problem is that we'd have to add an additional layer of wrappers over stdio to make this reasonable for the client code.  I can certainly do that, but I'm slightly concerned about making it efficient in the common case (though I don't have any evidence that it would be inefficient), readable, and simple all at the same time.  Also, there is an impedance mismatch between curl/libwww and PSPP: curl wants to provide data through callbacks, which isn't convenient, so we'd have to do buffering. 

• Fork a "curl" or "wget" process and communicate with it over a pipe.  The only problem here is that mingw32 doesn't have fork, and I'd like to make this a general-purpose solution.  We could use popen, which is implemented by mingw32, but then error reporting would suffer: the subprocess would only be able to report errors via its exit status (popen doesn't provide a way to intercept stderr).

(file #13525)