- From: Lou Montulli <montulli@mozilla.com>
- Date: Wed, 16 Aug 95 16:39:16 -0700
- To: Shel Kaphan <sjk@amazon.com>
- Cc: Lou Montulli <montulli@mozilla.com>, Larry Masinter <masinter@parc.xerox.com>, bne@bne.ind.eunet.hu, http-wg%cuckoo.hpl.hp.com@hplb.hpl.hp.com
In article <199508162233.PAA14427@bert.amazon.com> Shel Kaphan <sjk@amazon.com> wrote: > > Lou Montulli writes: > > > > You are forgetting that this only solves a third of the problem. > > > > It doesn't solve the problem of two dates being exactly the > > same but the file has been modified, or the problem of a file getting > > modified and having it's date set into the past. Date's alone > > are not a strong enough versioning system. And don't try and tell me > > that these don't happen. > > *I* won't try to tell you that these don't happen, though they seem to > be evidence of bugs or human activity that it is not clear HTTP > should try to correct for after the fact. I can't think of a reason > for this happening other than someone or something changing a system > clock or deliberately falsifying a modification date. In any case, it > doesn't seem like the responsibility of a communication protocol to > try to figure out what's "really" going on if a server is too lame to > get it right for itself. With most current implementations of HTTP servers it is impossible to "get it right for itself". HTTP servers reference a file system that can be changed at random. The HTTP server can only rely on the last modified date of a file and that date can be inaccurate. Adding an additional size checksum allows the server to "get it right for itself" much more often. > > The percent of the time that there is going a problem like this seems > like it must be small, certainly not worth imposing additional > overhead on the protocol to correct for, especially if that involves > scanning whole files for checksums. Protocols like HTTP are not about getting is right 99% of the time, they need to get it right 99.99% of the time. Adding a size checksum significatly reduces the error rate. Adding an MD5 checksum reduces that error rate even furture. I don't want to require all servers to interpret size= or MD5=, but client should be allowed to send them optionally so that the server can use them if it really wants to make an informed decision about sending a 304. Perhaps reliability is not as important to your systems as it is to ours? If that's the case, make your servers ignore size= and only use the date. :lou -- Lou Montulli http://www.mcom.com/people/montulli/ Netscape Communications Corp.
Received on Wednesday, 16 August 1995 16:41:10 UTC