Models can define default order conditions that will cause issues with
postgres. By setting order=>false on all the queries emitted by
TreeBehavior, we can avoid this issue and not have issues with models
that define a default ordering.
Refs #5155
When using subqueries or joining against tables containing FROM the
results were bizzare and unexpected. By tokenizing the entire query we
can more accurately grab the fields.
Refs #3972
Using camelize() does the transformation we want without also
singularizing the name. This makes user input more transparent and fixes
issues around plugin schemas with plural names.
Refs #4993
The HTTP specs seem to indicate that the Location header should contain
an 'absoluteURI' which includes encoded data. In order to avoid
a regression with the issue fixed in b9ee4fc9f1
we'll continue to replace `%2F` and no longer decode the entire URI.
Fixes#5076
If create() is called with a relative directory and the path does not
exist, it should be created inside of the folder->pwd() and not inside
the process' cwd.
Refs #4990
disabled multiple checkbox input should not generate an enabled hidden input,
as it leads to saving empty data upon submission -> carrying the disabled attribute to the hidden input prevents this
The Memcached extension does not provide enough information to PHP's
Reflection API to allow mock objects to be generated correctly. We'll
skip the affected tests until php-memcached-dev/php-memcached#126 has
been resolved.
Because getMockForModel() does not go through the test datasource
injection in ClassRegistry::init() we need to duplicate the basics of
that logic here. Thankfully, we already have a mock so we can do that
datasource switching without reflection. Of course this means there will
be limitations to how/when this will work, but I feel those scenarios
can probably be solved by not using mocks, or by mocking out the
problematic methods. This set of changes makes getMockForModel() work
with secondary datasources, as one would expect it to do, but I'm not
sure it ever did.
Refs #4694
While it *did* work for single fields, isUnique could not be used to
validate the uniqueness across multiple fields as documented. Because
validation methods pass arguments in an order the validator did not
expect the validation method would not work correctly.
Fixes#4571
