May provide strings as attributes rather than unicode

[James Westby]

Hi,

While trying to land the launchpadlib update in to Launchpad itself, I ran in to test failures
with the latest release.

Confused I went to try and reproduce in lp:launchpadlib.

I ran python bootstrap.py && ./bin/buildout && ./bin/test and found that test_no_credentials_calls_get_token_and_login
was failing, as it was expecting the "beta" webservice, when 1.0 is default. This is bug 488448.

I didn't see the failures I was seeing when using in launchpad, and so dicsovered
that ./bin/test doesn't run any of the doctests, and I think this may mean that the tests
haven't been getting exercised.

Writing a small test_docs.py to load the doctests I was able to run them.

The failure I got, which was the same as when trying to land the upgrade was:

File "/home/jw2328/devel/launchpadlib/launchpadlib/launchpadlib/src/launchpadlib/tests/../docs/people.txt", line 34, in people.txt
Failed example:
    me.name
Expected:
    u'salgado'
Got:
    'salgado'

which is what was failing when I tried to upgrade launchpad's copy.

This one is caused by the

  332 representation = simplejson.loads(representation)

in lazr.restfulclient.resource. If we want unicode strings then pass unicode to
simplejson.

Thanks,

James

[James Westby (james-w)]

Ah, I get it now, these doctests aren't supposed to be run outside of the Launchpad test suite.

Still, there are a couple of real bugs here, one of which is caught by the lazr.restfulclient
testsuite, but it seems that if run with an older simplejson or something it doesn't trigger
the problem. The other can be caught with a test there too.

Thanks,

James

[Gary Poster (gary)]

> these doctests aren't supposed to be run outside of the Launchpad test suite.

Right.

> it seems that if run with an older simplejson or something it doesn't trigger the problem.

We've had very similar simplejson/unicode related problems in the past. I think the newer ones work better but there's also some odd edgecase that Benji encountered in which he simply rebuilt simplejson (with the same version as before) to correct a test failure like this.

I'm triaging low because we do run the tests in Launchpad's test suite, and they are passing. Counterarguments welcome.