|
From: | Aaron Whitehouse |
Subject: | Re: [Duplicity-talk] On replacing tar, why not dar ? |
Date: | Wed, 9 Sep 2015 14:38:41 +0100 |
User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.2.0 |
I haven't personally used dar. It looks as though it has some very
interesting features, but mainly only one (quite active) developer.
I see that public key encryption is only available in the
pre-release version: http://sourceforge.net/p/dar/feature-requests/67/ I'm always nervous when people talk about a complete rewrite, particularly in a small project and particularly where it is something critical like backups (and you may not find issues until you come to restore the backup many years later). It also reminds me of: http://onstartups.com/tabid/3339/bid/2596/Why-You-Should-Almost-Never-Rewrite-Your-Software.aspx http://www.joelonsoftware.com/articles/fog0000000069.html It tends to create a pretty bad user experience: all the developers want to work on the new version; all the users are on the old version; and you spend all your time telling users that their bug will be fixed in the new version, but that they can't use that yet because it isn't finished. Is there any reason that we couldn't do things more incrementally? Say: 1. add in dar to do exactly what we currently use tar for (and no more; potentially activated by a commandline option?); 2. switch the default to write dar files (but read either type); 3. as developers have time, we could replace pieces of home-brew code to use dar's features when duplicity is using dar files; and 4. if maintenance of the tar-writing codepaths is considered too much of a burden, turn off tar file output (though keep tar reading code). I would expect it would create a much better user experience, if nothing else, and would mean that we could more gracefully deal with things like the user having an old version of libdar that doesn't contain the required feature (taking the above example, use our encryption code for public key encryption until the user is both outputting to dar files *and* has a version of libdar that supports public key encryption). Just my thoughts and happy to be proven wrong by somebody who knows more! Aaron On 08/09/15 16:54, Kenneth Loafman
wrote:
|
[Prev in Thread] | Current Thread | [Next in Thread] |