package (samba) is re-merged infinitely [solved]

[christoph_peter_s]

Dear Gentoo experts,

I have a few small machines, which act as Domain Controllers in my home network. On one of them, samba does not stop from being re-emerged whenever I update.

[Naib]

[christoph_peter_s]

Hi Naib,

here it is:

[Naib]

by USE flag's forcing this, one of samba's USE flags must be linked to another packer (ie set flag foo --> libfoo is needed).
This additional package must have been updated around the time you experienced this.

what is the output of: (3 additional options have been added

[christoph_peter_s]

Dear Naib,

I did as You advised...

[cboldt]

--deep and --complete-graph are redundant, I think, with -deep controlling. Both options " consider the deep dependencies of all packages from the world set, and will bail out if it determines that the given operation will break any dependencies of the packages that have been added to the graph."

The difference is that --deep can cause more packages to be updated. --complete-graph considers all the dependencies, but doesn't necessarily update them.

Add the "--tree" option to view these dependencies. "--unordered-display" can make the output from "--tree" more readable.

[christoph_peter_s]

It seems my problem is a tough one... here's the output from a (slightly) faster machine, which is also affected...

[cboldt]

Samba-4.6 are "hard masked" in /usr/portage/profiles/package.mask. I don't know that this observation is relevant, just making it.

Without "--tree" in the output, we don't get a clue what is causing portage to think your samba needs to be rebuilt. When I have "forced rebuilds," the emerge command tells me WHY those rebuilds are being ordered. After updating my portage tree, I use this to figure out the effect on my system ...

[christoph_peter_s]

Dear cbolt,

Thank You for Your recommendations. I will do so in the future. It is always better to understand, what's going on...
For the time being, I need some other methods to isolate the problem, as even the verbose option doesn't tell me, why exactly portage thinks the samba package needs to be rebuild.

Best regards
Peter

PS: will immediately try to emerge a depclean samba...

PS2: OK, doing it the hard way. Uninstalled samba, and then

[cboldt]

I tried the "-p --depclean" and didn't get any useful information. I also tried "emerge -pe samba" and was overwhelmed. BUT, this one might shed some light ...

[cboldt]

You don't by chance have "--verbose-slot-rebuilds" set to "n"?

[cboldt]

My experience with adding the -v switch is that it lists USEflag and size of download, and for --pretend emerges, adds nothing else.

The --backtrack option isn't relevant for your issue. The samba rebuild is found and ordered regardless. --backtrack is used to resolve BLOCKED emerge commands, and will uncover additional packages that, when rebuilt or removed, resolve the blockage.

[christoph_peter_s]

OK, now I have three 'threads'... on one box I run the reinstallation after removing samba and its immediate dependencies via --depclean. This should get first results later this evening.
On the second box I started a reinstall with the --emptytree option. About 300 packages on a Raspi 2 - this will take one to two days. But that doesn't bother me much...
And then I will run the update with the -d option and tee its output to a file. If all these fail, I will file a bug on net-fs/samba.
I keep You informed... maybe some other guys will hit the same troubles.

[christoph_peter_s]

The first approach has failed. I.e. unmerging samba + emerge --depclean + re-emerging samba.
I run the update now with the -d option and write the output via tee to a file. Maybe I'll find some hints here.

[christoph_peter_s]

OK, with the -d option, I found the point, where the problem is....

[Tony0945]

Maybe this helps?

[cboldt]

both versions of dnspython can play ...

[cboldt]

It appears dnspython is being brought in by samba (I ran `equery d -a dnspython`), when samba USEflags includes "addns"

The default for that USEflag is -addns, meaning the flag is turned off. Also, samba is the only package that takes advantage of USE=addns

Besides samba, sys-custer/swift is the only package that uses dnspython

Are you using samba as an active directory (AD) DNS? If not, just turn off the addns USEflag.

[christoph_peter_s]

I solved it on one of the machines by upgrading to dev-python/dnspython-1.15.0, which is still masked testing for the arm architectures. As the eix output for dnspython does indicate, there are to distinct verisions for python 2 and python 3, which reside in different slots. Something goes wrong here. I wonder what will happen with the emptytree-installation - but this is still running.
I will now upgrade to dnspython-1.15.0 for the second machine. Hoping that this is really the solution. I came to the idea of upgrading, as the new version seems not to need different slots for Python 2 and 3. So it seemed worth the effort for trying.
These small machines are my domain controllers (fully sufficient for a two user home domain with two sites). The file servers are capable machines, the bigger one with an Sandy Bridge core i7 CPU and a big H/W raid. Obviously the file servers don't have the ADDNS flag set.

PS:
I don't have too many positive experience with --emptytree installations. I tried it a few time as the means of last resort. It helped to rescue a near completely hosed installation, but for fixing issues like a slot of use flag problem, it seems pretty unsuited.

[cboldt]

I don't believe anybody here would have advised using -e (--emptytree) to resolve the problem. Many unnecessary compile operations result. I've used --emptytree in conjunction with --pretend to troubleshoot, basically to find a "too big" list of dependencies for a package.

You should be able to adjust your machines with "--upgrade --deep" after unmasking dnspython-1.15.0, followed by a --depclean to get rid of the unnecessary versions.

[christoph_peter_s]