| 1 |
This list is pretty quiet unfortunately. |
| 2 |
|
| 3 |
On 24/04/15 22:12, Andrew Savchenko wrote: |
| 4 |
> Hi, |
| 5 |
> |
| 6 |
> On Sun, 12 Apr 2015 02:20:33 +0300 Andrew Savchenko wrote: |
| 7 |
>> Hello, |
| 8 |
>> |
| 9 |
>> I have a couple of concerns about a proper way of handling |
| 10 |
>> complicated dependencies during version bumps or stabilization. |
| 11 |
>> |
| 12 |
>> I. With some USE flags combinations package foo depends on bar and |
| 13 |
>> vice versa. While with a properly set deps (e.g. PDEPEND) this is |
| 14 |
>> not a problem for users during updates; but this is a problem |
| 15 |
>> during commits: repoman doesn't allow to commit multiple packages |
| 16 |
>> at once, so for a short period of time deptree will be broken |
| 17 |
>> during version bumps or stabilization. |
| 18 |
>> |
| 19 |
>> E.g.: |
| 20 |
>> sci-physics/root: PDEPEND on ~app-doc/root-docs |
| 21 |
>> app-doc/root-docs[api] DEPEND on ~sci-physics/root |
| 22 |
>> |
| 23 |
>> I see three possible solutions: |
| 24 |
>> |
| 25 |
>> 1) commit both packages with short window between (e.g. minute or |
| 26 |
>> two); |
| 27 |
|
| 28 |
This is almost always used and is usually fine - even fairly large sets |
| 29 |
of changes such as KDE bumps run without issue. You can always time your |
| 30 |
commits to avoid the cvs -> rsync update if you're worried. |
| 31 |
|
| 32 |
>> 2) ask infra to block cvs <-> rsync tree syncs for a while; |
| 33 |
|
| 34 |
This has been indeed been used rarely in the past for mass changes, such |
| 35 |
as qt x11-libs -> dev-qt pkgmove. |
| 36 |
|
| 37 |
>> |
| 38 |
>> 3) temporarily mask USE="api" for root-docs, commit root-docs, |
| 39 |
>> commit root, unmask USE="api". |
| 40 |
>> |
| 41 |
>> Second option looks like overkill and IMO should be used only for |
| 42 |
>> large scale operations like mass eselect package moves. |
| 43 |
>> |
| 44 |
>> Third option is formally correct, but it requires double |
| 45 |
>> modification of tree-wide file where any mistake may broke the |
| 46 |
>> whole tree. I think this approach brings more risks than benefits |
| 47 |
>> for a simple issue described above. |
| 48 |
>> |
| 49 |
>> Currently I use first option, since probability of broking deptree |
| 50 |
>> is low and affected audience will be very limited. |
| 51 |
>> |
| 52 |
>> Any idea how to handle this issue better? |
| 53 |
>> |
| 54 |
>> |
| 55 |
>> II. What to do if package stabilization requires to stabilize deps |
| 56 |
>> maintained by other devs? |
| 57 |
>> |
| 58 |
>> Should I just file a bug with stabilization request to a |
| 59 |
>> corresponding dev or team? |
| 60 |
>> |
| 61 |
>> What to do if I have no version requirements for a |
| 62 |
>> version to be stabilized? E.g. app-doc/root-docs needs |
| 63 |
>> dev-haskell/pandoc-citeproc, but it seems to work with any version |
| 64 |
>> in tree. Probably in will be the best for a relevant team to pick |
| 65 |
>> up a stabilization candidate for their convenience. |
| 66 |
>> |
| 67 |
>> Should I add arches there? Since I'm not a maintainer of a |
| 68 |
>> dependency package I'm not sure if I can do that. Package may |
| 69 |
>> require some dependencies to be stabilized as well... |
| 70 |
>> |
| 71 |
>> If I'm sure that some package (e.g. dep of my package) works fine |
| 72 |
>> and I see that it has no unstable deps or critical bugs, may I CC |
| 73 |
>> arch teams and ask for stabilization myself? Pros here are that |
| 74 |
>> this will speed up stabilization process and lower overhead for |
| 75 |
>> other devs, cons are that package maintainer may be aware of some |
| 76 |
>> issues preventing stabilization of this specific version. |
| 77 |
|
| 78 |
File a bug against whichever version looks sensible, but let the |
| 79 |
maintainer CC archs (or just ask them beforehand). |
| 80 |
|
| 81 |
> Any comments? |
| 82 |
> Should I redirect these questions to gentoo-dev mail list? |
| 83 |
> |
| 84 |
> Best regards, |
| 85 |
> Andrew Savchenko |
| 86 |
> |
Archives