We do a proper length check when using the sockets in IPC mode and fail when it reaches 103 which is the lower value for this error on BSD systems, Linux and Solaris is 107. We just exit the master/minion since continuing would mean leaving it crippled.
* Fixed the `brew` and `sysbench` modules which were returning `None` instead of `False`.
* Additionally, regarding the module `sysbench`, @sharan-monikantan, why are the parsed returned keys white-space padded?
I noticed earlier tonight when I added some information to the "hacking"
docs that my "root_dir" and "sock_dir" links were not actually linking
to anything, and found that they were missing from the minion
configuration page. It turns out that there were a good many params that
hadn't made their way into the docs yet, so I added as many of them as
I could identify.
* On some occasions it seems that `salt.utils.event.SaltEvent.destroy()` is called twice and the errors exhibited on #3015 would show up on the second time we would try to set the linger socket option since that socket was already closed. So, even though on some parts of code which I changed in this commit weren't giving any problems, I ended up using the same recipe all around, if the socket is open, set the linger option and close it, plus, if open, also close the context.
Platform-specific code from #3019 has been moved to pkg_resource module.
This code may evolve a bit down the road. Presently it will only check
syntax if the package is coming from the portage tree via ebuild.
However, future support for entropy (binary pkg manager), as well as
expanding syntax checking for binary package files should be considered
TODO items.
Do not proceed unless package names are provided using the
category/pkgname format. For packages submitted without the category
name, offer suggestions from the portage tree.
acd949c and 0b5ce09 broke the pkg state, such that for single pkg
installs (i.e. no "pkgs" or "sources"), if the version arg is not
provided, a package will always be interpreted by the pkg state as being
installed, and the package will never be installed. This commit fixes
that regression.
It was discovered that pkg module for Gentoo-based systems does not work
as expected if the full package name (including the category) is not
specified. This is not documented anywhere, so this should give users
better changes to know about it.
Check issue #3008 on GitHub to know about further development.
Several modules where __virtual__() were returning None were being
improperly loaded. Any __virtual__() that returns a string will
never evaluate to False when cast to a bool() so "if not virtual"
will do the right thing as we get more and more modules that might
not necessarily be written up to standard (until we can fix them)
