This ports the safe parts of #9619 and updates the tests. Because
existing tests had to change and I was concerned about changing people's
email messages in a bugfix release I'm targetting 2.next with this
change.
By removing a bunch of empty() guards we can make '' behave like all the
other key names. This does change the existing behavior/tests around ''
key, but I think that is ok given the need to manipulate ''.
Refs #9632
FormHelper should not run URLs through Router twice when determining the
the form's lastAction attribute. However, because we're using the helper
method (see #9414) we do need to HTML decode the URL before using it in
form token generation.
Refs #9455
When mock objects are created from models that access their datasource
in the constructor, an exception would be raised for the missing default
datasource. By changing how configuration data is handled in the mock
creation we can avoid this issue and not reopen#4867
Refs #8225
The current checks are way too relaxed, and are more like testing
for a substring, which makes it easy for invalid paths to slip
trough, for example `/foo/var/www` is falsely tested to reside in
`/var/www`.
Passing an empty path never worked properly, it was triggering a
warning, didn't worked on Windows, and the behavior that the current
top level directory would be assumed for empty paths wasn't
documented.
Similar is true for relative paths. While they did match at one point,
this was incorrect behavior, and matching actual path fragments seems
out of scope for this method.
This change makes the `$path` argument required, requires it to be an
absolute path, and throws an exception in case a non-absolute path is
being passed.
In eaa2bbbcae I changed the behavior to
now use the TitleCased name as I didn't understand the intended behavior
or how apache was working in the reporter's specific case.
Refs #9229
