All this talk of probabilities and possibilities took us on a tangent. It's
occurred to me that, if CVS is to compute MD5's without getting confused with
line endings, one of the following must occur:
1. The server sends the file to the client so that the client can compute and
compare the "real" file's MD5 with the local copy's MD5.
2. The client sends the local file to the server so the server can compute and
compare MD5's.
3. The client computes the local file's MD5 (taking into consideration line
endings for ASCII files). The server does the same computation and sends it
over to the client so it can be compared.
4. I completely missed something in my analysis.
The problem with the first two (if it's not already obvious) is that it defeats
the purpose of the MD5 comparison (which is to decrease the amount of net
traffic).
The third option makes the client a somewhat more complicated.
If it's the fourth option, I'm sure others will point out what's not obvious to
me.
Noel
[EMAIL PROTECTED] on 2000.04.19 18:58:32
To: [EMAIL PROTECTED]
cc: [EMAIL PROTECTED] (bcc: Noel L Yap)
Subject: Re: Could timestamps be replaced with MD5?
[EMAIL PROTECTED] on 04/18/2000 05:20:22 PM
>I'd like to see this as a switch. Then, I'll modify my cvs alias to be
>'-z9 -q -md5', and the people developing aircraft software can keep it off
>to satisfy their ISO standard requirements.
Having been on the sidelines all this time, I'm at a point where I have an
opinion (oh no!). I think the switch thing is the way to go (although I think
the switch should be "--md5" instead of "-md5"). IMHO, the people who won't
want to use the switch are the ones who have a misplaced sense of danger.
Also, the "download instead of upload" idea is arguable since the upload could
potentially save more info (ie the diffs themselves when the server needs them)
from going through the wire.
Noel
