ebuild.installed returns unhelpful errors

[kaithar]

I don't know where the error is actually being formatted, but after digging it turns out this error:

State: - pkg
Name:      baseinstall
Function:  installed
    Result:    False
    Comment:   The following packages failed to install/update: <long package list here>. The following packages were already installed: <more packages here>.
    Changes:   

Is actually being generated by a missing keyword (clarification edit: a missing entry in the accept_keywords file, not anything within Salt). The way the comment is phrased gives the impression that the update failed /because/ the packages were already installed, rather than that being a separate bit of status information.

Any thoughts on how this could be made more helpful to the user? I had to run salt-minion in the foreground to find out what emerge was actually complaining about.

[basepi]

Thanks for the report.

As to how to fix it, any idea what you would want the output to look like?

[kaithar]

That's the tricky bit... parsing the output of emerge to get a more accurate answer is a pain since it was intended to be read by humans.
In a perfect world the comments section would tell you the actual fault (in this case, "Keyword changes are necessary to proceed"), though I think that's unrealistic for the current design.
Might be helpful to have the output give something like "The system package manager returned an error for :" and then the actual command executed. The exact output is perhaps a bit much to send back, but maybe have it write the output to /var/log/salt/minion if the command fails?

[tmessi]

I have been wanting to improve the output of the ebuild module for failure cases like this. I think it is possible to parse the output from emerge and provide some helpful details back to the user. Perhaps we can start by documenting the various failure cases (and some example emerge output would be a bonus).

[basepi]

@kaithar That's a good idea, to output the command that was executed and try to restrict the super-verbose output.

Now we just need an emerge expert to help us with all the failure cases! =P

[tmessi]

Some off the top of my head:

• keywords changes
• use flag changes
• license changes
• mask changes
• fetch failed
• circular dependencies
• blocked packages
• fetch restricted
• slot conflicts

[thatch45]

@shadowfax-chc when you have a moment can you please give us an update here? No rush :)

Thanks!

[tmessi]

I haven't had a chance to address the last 3 errors in that list. They tend to be less common so I haven't had example output to work with. If anyone encounters an emerge error that salt doesn't handle, please provide the out put of the emerge command. For example, if pkg.install fails, provide the output of emerge --quiet --ask n <package>

Also I can start trying to craft some scenarios to make those errors repeatable.

[basepi]

That would be great! We've just been trying to clean up the issue list and wondered if there had been any undocumented progress on this issue. =)

[kaithar]

Recent pains have reminded me that qt is an excellent source of weird blockers.
Oracle Java is a nice manual fetch example.

[terminalmage]

Looks like we still lack reporting for "fetch failed" and "circular dependencies"