On Mon, Feb 23, 2009 at 9:57 PM, Sebastian Nowicki <sebnow@gmail.com> wrote:
On 24/02/2009, at 2:56 AM, Aaron Griffin wrote:
On Mon, Feb 23, 2009 at 1:47 AM, Brendan Hide <brendan@swiftspirit.co.za> wrote:
Again, this will be properly fixed as soon as pacman correctly takes into account the errors given when doing the actual deployment. Personally, I like the idea of adding sanity checks - but only if they're done properly. ;)
I wouldn't say that - we've talked about "proper transactions" for a long time, and are still nowhere near a proper solution. So don't expect this to be fixed soon, unless you're writing the patches :)
Has there been any (real) discussion about transactions? Perhaps a wiki about design ideas would be good.
I know lots of people like wiki pages for this sort of thing, but I never have - to use a relevant analogy (due to list discussion): a wiki page is like looking at a working dir, whereas for design decisions, it's important to see the changelog :) I like seeing WHY certain decisions were made, and a wiki page generally won't tell me that. It will tell me that Bob deleted 3 lines and Jim added a new bullet point. But, I digress. I think the only conversation on this topic was between Dan and myself over email, and Dan, Xavier, and myself on IRC. But this was long ago and it may have been brought up since then. The process is simple: if any stage in a given package install or upgrade fails, we should be able to "roll back" to the same state before that install/upgrade began. Ideas are welcome, but I haven't seen a really good idea regarding this that isn't slow as hell.