Red Hat Bugzilla – Bug 1023908
RFE: support for user configuration
Last modified: 2014-07-28 10:54:03 EDT
Description of problem:
I do lot of builds with mock. Each time I have to give "--resultdir=~/rpmbuild/mock/PACKAGENAME-PACKAGEVER -v" or do it in shell script. Would be great if all mock options could be set in some user configuration file.
Users building for custom distributions could use own distro configs without asking system administrator for copying their configs into /etc/mock directory.
All options have to be given on each invocation of mock.
"mock path/to/package.src.rpm" builds according to user specification.
I came from Debian and used pbuilder a lot.
That would be nice indeed.
If I may suggest, please use ~/.config/mock/ rather than ~/.mock/
Mock is clearly not a desktop application, but it would still be nice to follow the FreeDesktop.org spec here, to avoid polluting people's HOME dire with more of .files, and make it easier to backup user configuration. :)
There's the pyxdg module in Python to get the right folder to use:
>>> from xdg.BaseDirectory import xdg_config_home
Forgot to mention: I'd be willing to work on a patch if that's a desired change. :)
I'd be happy to accept a patch from you Mathieu :)
You may want to wait for 1.1.34 before doing any serious work. I'm testing it now and once it gets past basic checks I'll create an update for it and will update my git tree.
Huh, this seems to already exist, actually:
$ cat ~/.mock/user.cfg
config_opts['root'] = 'fedora-rawhide-x86_64-modified-from-user-config'
config_opts['resultdir'] = "/var/lib/mock/myresultdir"
$ mock -r fedora-rawhide-x86_64 --rebuild ipset-6.20.1-1.fc20.src.rpm
INFO: mock.py version 1.1.32 starting...
Start: init plugins
INFO: selinux disabled
Finish: init plugins
INFO: mock.py version 1.1.32 starting...
INFO: Start(ipset-6.20.1-1.fc20.src.rpm) Config(fedora-rawhide-x86_64-modified-from-user-config)
[... snip ...]
INFO: Results and/or logs in: /var/lib/mock/myresultdir
INFO: Cleaning up build root ('clean_on_success=True')
Start: lock buildroot
Start: clean chroot
INFO: chroot (/var/lib/mock/fedora-rawhide-x86_64-modified-from-user-config) unlocked and deleted
Finish: clean chroot
Finish: lock buildroot
The code to do that was added in this commit:
Author: Masatake YAMATO <firstname.lastname@example.org>
Date: Thu Jun 7 14:29:21 2012 -0500
added option to set a plugin parameter value from the commandline [BZ# 754321]
This patch adds the --plugin-option command line option, which allows
a plugin parameter to be set on the commandline:
Ex: mock --plugin-option="ccache:max_cache_size=5G" foo.src.rpm
The above will set the cccahe max to be 5 gigabytes
Signed-off-by: Clark Williams <email@example.com>
Given that the commit message doesn't mention user configurations at all, I have a feeling that this wasn't expected to be committed, as if Masatake just added this to test his patch, and forgot to remove it (or to send it as a separate change)
So, Marcin, good news: you can already use a per-user configuration. :-)
(I still think it would be better in ~/.config/mock/, though)
Now to the other part of your question: can we use variables in the config so that, for example, the resultdir would be defined relatively to the package name/version?
Not at the moment, because the configs are read long before mock starts looking at the srpm to build.
In py/mockbuild/backend.py, the resultdir is obtained with this line:
self.resultdir = config['resultdir'] % config
So maybe if py/mock.py could proactively add the package name and version to the config, then it should all just work.
However, the problem is that mock doesn't always have a package to process. For example, when running the "--init" command, mock still needs to write the log files into the resultdir. But because there is no package associated, then the resultdir string wouldn't be interpolated properly, and as a result mock would store the logs in a folder like
/path/to/my/resultdir/%(package_name)-%(package_version)" which doesn't strike me as highly desirable.
So given all this, maybe this isn't such a good idea, and the best really is to specify --resultdir on the command line?
Note that the above is what I got from reading the code, but I don't know it that well and could very well be wrong :)
That looks like a good summary.
I do like the idea of moving configs into ~/.config/mock so if you want to gin up a patch for that feel free.
You *can* use variables in the config, but as you point out it's a hit-or-miss thing based on how mock was invoked. Of course you can create your own configs or edit existing configs and put reasonable (for you) defaults into them.
Given all that I'd probably say it's safer to just specify the resultdir on the command line.
Thanks for feedback guys. Looks like I will stay with my build scripts then.
So what is status of this RFE? Do you still want it? Or the findings in #4 is enough?
CLOSED/TOBEIGNORED would be best.
I looked at comment #4, created ~/.mock/user.cfg but it is far from what I had in pbuilderrc under Debian. Those tools differ too much so I will stay with my scripts around mock as they are good workarounds for existing code.
Note that I use mock only to build packages. So all I want is "take this srpm, build, drop into shell if failed" which my script does:
$ cat ~/bin/hrwmock.sh
LC_ALL=C LANG=C mock $SRCRPM \
--uniqueext=$PACKAGENAME && X=Y
if [ $X == X ]; then
mock --shell --uniqueext=$PACKAGENAME
mock --clean --uniqueext=$PACKAGENAME
This way I can have several builds going at same time as they do not reuse the same chroot and build dir is dropped after build of package. I also use squid with few GB of on-disk cache instead of yum caching.