package-moves: Allow moving back, and describe how to chain updates #326
Conversation
|
Does the relevant section of PMS need clarification? It says:
I suppose it doesn't say "has ever". |
Probably yes. I suppose it's going to be hard to formally convey the right meaning. |
It does happen. And when it does, I think it makes sense to be explicit on how to make it work without stuff accidentally exploding in user's face. |
|
We had a previous discussion in bug 296713, where the conclusion was that mistakes can be corrected before they hit the rsync mirrors. Apart from that case, moves are pretty much permanent and cannot be reverted. Are we sure that this change won't open up possible undefined behaviour, especially when considering multiple repositories and binary packages? |
Make the rule forbidding reusing old package names less strict. There is no real reason not to move packages back — we can safely assume that the package manager will handle that just fine. The only real restriction is that no other package must reuse any of the old names, i.e. effectively cause the package manager to attempt merging the packages together. Add explicit rules for updating old package move entries. These match the logic used in pkgcheck, and therefore ensure both correct results and no warnings from pkgcheck. Signed-off-by: Michał Górny <mgorny@gentoo.org>
I can't think of anything worse than what we have right now. The worst mess would be if overlay copies our My understanding is that if you move package back, and remove the original move entry, then:
|
|
What do we do for PMS (which I'd rather not expand too much)? Would something like the following be sufficient? |
Wouldn't this imply we can never cleanup |
That would work for me. Though TBH I think PMS is stepping too much into the "policy" area here. Perhaps it would be cleaner to explain what the consequences would be. |
|
Thinking about it, I'd also be happy to revert https://gitweb.gentoo.org/proj/pms.git/commit/?id=3ba90e4a4918cfe820bb7c627bef2fc0d5a9fed5 and leave the policy for the devmanual. |
See gentoo/devmanual#326, oops. Signed-off-by: Sam James <sam@gentoo.org>
How about this? |
|
perfect, thanks! |
|
Yes, that sounds about right. Also clearly splits specification from policy. |
| That is, rather than the chain: A:S₁ → A:S₂, A → B; we prefer to have | ||
| the chain: A → B, B:S₁ → B:S₂. All package and slot move entries must reside |
There was a problem hiding this comment.
I think that Unicode superscript and subscript characters should be avoided. They are a workaround for plain text but have no place in a markup language like XML or HTML. (In other words, a 1 is a 1 regardless of its appearance in regular text or in a subscript.)
Edit: I suggest to change it to either A:X → A:Y or A:S1 → A:S2. Unless we would expect to see frequent future use of subscripts and superscripts, in which case <sub> and <sup> could be added to DevBook XML.
|
Merged (with the suggested markup change) in commit 5cb6b40. |
Make the rule forbidding reusing old package names less strict. There is no real reason not to move packages back — we can safely assume that the package manager will handle that just fine. The only real restriction is that no other package must reuse any of the old names, i.e. effectively cause the package manager to attempt merging the packages together.
Add explicit rules for updating old package move entries. These match the logic used in pkgcheck, and therefore ensure both correct results and no warnings from pkgcheck.