The `--noreplace` argument to `emerge` is generally coupled with
`--newuse` or `--changed-use`, and can be used instruct Portage to
rebuild a package only if necessary. Simply checking to see if the
package is already installed using `equery` is not sufficient to
determine if any changes would be made, so that step is skipped when
the `noreplace` module argument is specified. The module then falls back
to parsing the output from `emerge` to determine if anything changed. In
check mode, `emerge` is called with `--pretend`, so it produces
different output, and the parsing fails to correctly infer that a change
would be made.
This commit adds another regular expression to check when running in
check mode that matches the pretend output from `emerge`.
Signed-off-by: Dustin C. Hatch <dustin@hatch.name>
When running in check mode, the *portage* module always reports that no
changes were made, even if the requested packages do not exist on the
system. This is because it was erroneously expecting `emerge --pretend`
to produce the same output as `emerge` by itself would, and attempts to
parse it. This is not correct, for several reasons. Most specifically,
the string for which it is searching does not exist in the pretend
output. Additionally, `emerge --pretend` always prints the requested
packages, whether they are already installed or not; in the former case,
it shows them as reinstalls.
This commit adjusts the behavior to rely on `equery` alone when running
in check mode. If `equery` reports at least one package is not
installed, then nothing else is done: the system will definitely be
changed.
Signed-off-by: Dustin C. Hatch <dustin@hatch.name>
- Backport config file handling from the DNF module rewrite #527
(Current config handling does not work with dnf and leads to
tracebacks when run as an unprivileged user).
- Make a mandatory requirement on yum-utils (for /usr/bin/repoquery)
because none of the fallback code works for dnf (it's unported yum API
code).
Both of these issues will be fixed better in the dnf rewrite when it is
feature complete.
Fixes#471
The documentation for the `state` field is not very clear.
It says possible values are "installed, uninstalled" and default value is "present"
The examples below alow uses `present` and `absent`.
This patch uses "absent" and "present" instead of "installed" and "uninstalled"
Moreover, this is consistent with other packaging modules, like homebrew itself
package_latest was calling package_present but did not care
about the return code so errors in package_present were hidden
and everthing look ok on the console when zypper update did not fail,
but no packages where installed.
* Add support for check mode
* Use "pkgin search" to guarantee 0 or 1 result
* Edit documentation for style, new feature, etc.
* General refactoring
* Lay some groundwork for future support of "state=latest"
Some packages attempt to prompt the user for certain settings during
installation. Thus, this parameter sets the environment variable
$BATCH to 'yes', which forces package installation scripts to accept
default values for these interactive prompts. This should work for all
prompts that have a default value and aren't implemented through a
custom script (as this variable is built into the ports/package system).
FIXME: Package install should fail if it prompts and batch isn't set;
currently, the install hangs indefinitely.
TODO: Allow user to specify the answers to certain prompts.
I (github.com/mwpher) have NOT tested this with any packages besides
bsdstats. It's a small improvement, but not a complete answer to all
the complexities of package installation.
In case of release repositories or other special cases you might not
need the autorefreshing of the repos. This patch adds a configure
option instead of hard enabling this.
Signed-off-by: Justin Lecher <jlec@gentoo.org>
Fix bug in ansible get_package_state and get_current_version that breaks when there are multiple versions of a package installed and there is a list of packages to install.
The previous implementation used 'zip' to match requested names to installed names which fails, because rpm outputs multiple lines per package when there are multiple versions.
Testcase: Install opensuse, install multiple kernel versions (happens by update)
Before patch: calling
zypper: state=present for name={{item}}
with_items:
- kernel-desktop
- git
leads to ansible aborting.
After the patch ansible performs as expected and makes sure both packages are present.
Also the last version number is used for further update information in this version (before if only one package name was given the oldest version number was used).
