On Wed, 2009-12-09 at 23:08 -0500, Rob Crittenden wrote:
Jason Gerard DeRose wrote:
Okay, here's a revised patch.
Significant additions/changes from the previous version are:
1. The return value dict now includes a 'summary' value, something like
'Added user jdoe'. This summary is used by the CLI and webUI.
Previously I was generating the summary in the CLI and webUI separately.
This removes the duplication and allows the commands to easily produce
arbitrary summaries (before they were limited a single summary format
like 'Added user %(primary_key)s'. This also makes it easier for
3rd-party tools to provide UIs without having to introspect the Python
API (because they happen to be written in PHP, whatever).
2. I renamed the 'primary_key' member in the return value dict to
'value'. This is simpler and will be will be easier on translators
('Added user %(primary_key)s' vs 'Added user %(value)s'). I'm also
thinking of returning the name of the primary_key (e.g., 'uid') when
returning an entry or a list of entries, so this opens the door for me
to use 'key' in the future without confusion. Note this change is only
relative to my previous proposed patch. The use of the return value
dict hasn't yet hit master.
3. XMLRPC_test.setUp() no longer tests for server availability with
`user-show notfound` prior to each test running. Instead, I try to
connect to the server just once when the `xmlrcp_test` module loads,
which sets the `server_available` module attribute. XMLRPC_test.setUp()
will still raise nose.SkipTest for each test as before. This change
helps the XMLRPC tests run much faster and also makes problems easier to
debug server-side as there isn't all the `user-show notfound` background
noise.
4. This adds my new `Declarative` base class for the XMLRPC tests which
allows you to define the XMLRPC tests using simple data structures,
letting the base class do the tedious stuff. IHMO, the tests are
considerably faster and easier to write this way, but just as important
is the fact that Declarative takes care of reporting the errors when a
command's return value doesn't match what we expected. We have pretty
good coverage in the XMLRCP tests, but we don't have very good reporting
when something goes wrong. I've put a lot of effort into making sure
typical error reports contain the information needed to quickly focus in
on the problem. The most important part of the error reporting is in
the new tests.util.assert_deepequal() function, which can be used by any
test to compare two nested data structures. Currently only the
test_user_plugin and test_group_plugin tests are using `Declarative`,
but the rest will follow.
5. I rewrote the make-test script in Python and added a feature John
asked for and one I wanted. John wanted the ability to easily run only
the tests in one or more modules. You can now be specifying the module
in Python notation or the module file. For example:
./make-test tests.test_xmlrpc.test_user_plugin
Or equivalently:
./make-test tests/test_xmlrpc/test_user_plugin.py
I wanted an easy way to use the nosetests --stop option, which causes
the testing to abort upon reaching the first error, which I have found
useful when updating plugins to one of my incompatible API changes. Use
it like this:
./make-test --stop
Yup, big! May my patch reviewers one day forgive me.
-Jason
Ack. There are a couple of things we need to address such as porting the
rest of the plugins to work with this new return value scheme but we can
do that post-push. IMHO it is better to get this in now and clean up the
few remaining items than to delay any further.
We also need to try to avoid hardcoding domains in the tests. A couple
of user tests look for dc=example,dc=com instead of api.env.basedn.
rob
Thanks. Pushed to master. I'll get on porting the few remaining
plugins.
___
Freeipa-devel mailing list
Freeipa-devel@redhat.com
https://www.redhat.com/mailman/listinfo/freeipa-devel
