| 1 |
Richard Freeman wrote: |
| 2 |
|
| 3 |
> John Lawles wrote: |
| 4 |
>> |
| 5 |
>> The concept is that there would be a separate team maintaining |
| 6 |
>> the stable tree. A precedent for this would be the Linux kernel's |
| 7 |
>> two-track development. |
| 8 |
>> |
| 9 |
> |
| 10 |
> Actually, in theory that is what already happens with stable keywording |
| 11 |
> - or is supposed to happen. It is a different group of developers who |
| 12 |
> maintain the stable tree (well, there is overlap, but some people wear |
| 13 |
> multiple hats). |
| 14 |
> |
| 15 |
I agree that what with unstable, stable and herds working in overlays, we |
| 16 |
have enough structure in place. Indeed some users complain that use of |
| 17 |
overlays means they have to add stuff to layman to get bleeding-edge like |
| 18 |
they "used to." :roll: IMO the balance is right. For anyone who hasn't |
| 19 |
tried it yet, I recommend autounmask -n. |
| 20 |
|
| 21 |
> Expat was a real mess - mainly because they break ABI and it doesn't use |
| 22 |
> slotting. I'm not sure why slotting couldn't have been used in this |
| 23 |
> case (others might be able to comment on this). |
| 24 |
> |
| 25 |
> What we need is a better mechanism of warning users that they're about |
| 26 |
> to do something that could cause them major headaches. ELOG is useless |
| 27 |
> when you find out after the fact. |
| 28 |
> |
| 29 |
Well, we added /etc/warning to update[1] after the expat thing so we could |
| 30 |
automatically protect users from ABI breakages. (That's why it took 2 or 3 |
| 31 |
months to bring out the new version; testing in chroot was a pita til we |
| 32 |
got binhosts working nicely over the web.) It also picks up on any |
| 33 |
elog/warn/info that tells the user to revdep on a lib; from a fresh 2007.0 |
| 34 |
install when it does the expat thing it also picks up on libintl.so.7 |
| 35 |
(iirc) which isn't in the warning file. |
| 36 |
|
| 37 |
That's basically a mechanism for us to do what John was discussing: pick up |
| 38 |
on breakages that hit us in #gentoo or -chat, and if they're more than a |
| 39 |
simple revdep, codify a workaround. It's a generic mechanism, to do with |
| 40 |
packages that must be built before the package (none atm) straight after |
| 41 |
(eg gettext XML-Parser and libtool for expat), things we know will break |
| 42 |
and so should be rebuilt before the revdep (eg dbus) and the actual lib/s |
| 43 |
to revdep on. For all of these, they're only rebuilt if installed and |
| 44 |
slotting is checked. |
| 45 |
|
| 46 |
It's far easier doing that (and a whole load of other stuff) in a script |
| 47 |
imo, since it's easy for users to change without requiring dev time. It |
| 48 |
also keeps a logical separation between the front-end UI and the actual |
| 49 |
package manager, which is rightly more concerned about maintaining your |
| 50 |
system's packages in a coherent state, and building literally any type of |
| 51 |
software. |
| 52 |
|
| 53 |
> revdep-rebuild working better would also be nice. If it won't work the |
| 54 |
> ebuild should abort prior to install with a link to a document |
| 55 |
> indicating what will need to be done, and then users can read and |
| 56 |
> understand it before they break things. |
| 57 |
> |
| 58 |
Yeah that was the real problem with expat: revdep was going through a |
| 59 |
rewrite at the time. The change so that the new version would install the |
| 60 |
best-visible by default (as opposed to exact same versions) was fine (it's |
| 61 |
what update does by default, since you sometimes need a package where the |
| 62 |
version installed has gone from the tree) but it affected the apache2 |
| 63 |
upgrade since apxs is slotted. (Sorting that out added another week or |
| 64 |
so ;) |
| 65 |
|
| 66 |
[1] YAF update script: http://forums.gentoo.org/viewtopic-t-546828.html |
| 67 |
|
| 68 |
|
| 69 |
-- |
| 70 |
gentoo-project@l.g.o mailing list |
Archives