I recently read through this thread from the IETF HTTP working group, circa 1995, on Netscape’s then-current proposal to add byte-range support to HTTP. What I found fascinating about the thread is that at the time, there was great resistance to the idea. Netscape wanted support mainly so they could retrieve individual pages of embedded PDF documents, but many people objected that general byte range support was too expensive and complex to support this one use, and that byte ranges were the wrong mechanism for this anyway (which is true).
What’s amazing is that while the resumption of interrupted documents—which is probably what 99% of byte range requests are used for today—was mentioned as a potential use (and Netscape Navigator 2.0 did implement it), no one seemed to consider it a worthy goal. Many people pointed out that a lot of documents (server-translated HTML, CGI scripts, etc) could not be reliably byte-served, or would be prohibitively expensive to do so. Static binary files were mentioned, but mostly in the context of, e.g., a 100k image, where it would be a handy convenience rather than a huge time-saver.
It is remarkable today to remember that in the mid-’90s, HTTP simply wasn’t considered a viable way to transfer files. If you wanted to download a large file, you switched to FTP. Nine years later, of course, I find myself routinely using HTTP to transfer multi-gigabyte files, for which the possibility of resuming a failed transfer without starting over is much appreciated.
I also noted with satisfaction that it was my idea to encode the range request in a request header, which is the design that ended up in HTTP 1.1.
Alexei, I am encountering a problem trying to serve a dynamically built linearized PDF from a servlet to MSIE. The servlet is invoked thru the POST method and streams back the pdf response with a pdf mime type. The transmission is terminated (by MSIE or Acrobat)and a GET is issued with the URL only. This GET is typically followed by a byte range POST request from Acrobat. However, because the GET is issued without the initial POST’s payload, the servlet does not know what data to return and so the subsequent POST from Acrobat is never generated.
Why is MSIE generating the intermediate GET? It appears to make byte-serving from a servlet incompatible with the POST method.
Requested Range Not Satisfiable
None of the range-specifier values in the Range request-header field overlap the current extent of the selected resource.
I am getting this error message on http://www.linkage.nl, which is really strange and only in IE 5.2.3 for the Mac. In other browsers there seems to be no problem…
How can I solve this ? (without NOT using IE anymore I mean of course…)