Implemented Selenium tests and login-help.

Implemented Selenium tests and login-help.

This branch introduces integration tests based 
on Selenium. This way it is possible to check that
the GUI actually works, is connected to the API 
backend, etc.

Firefox is the webdriver used here, it is installed 
by default in Ubuntu. All the graphical operations 
are executed in memory thanks to xvfbwrapper.

This branch also includes the changes Gary made to 
add a charm option for the login help text.

https://code.launchpad.net/~frankban/charms/precise/juju-gui/bug-1097191-selenium/+merge/142787

(do not edit description out of merge proposal)

[frankban, 2013-01-10 21:44:39]

Please take a look.

[gary.poster, 2013-01-10 22:26:11]

land as is, or maybe with changes, see the discussion of bug 1090716 below. :-)

Hi Francesco.  Unfortunately, the side-by-side diffs in Rietveld are broken for
some reason.  I just looked at the unified diff.

This looks great, thank you!  I have two questions.

First, for my own edification, why does the test FF connect to the self-signed
cert without complaining?  Does Selenium set that profile up for us or
something?

Second, as a comment, as you might have noticed I put Benji's branch for bug
1090716 up for review.  You have an XXX for that bug.  Note that with the
hidden-GUI branch, the GUI service will still be in the DOM.  It will just have
display: none or something like that.  Given that approach/implementation, can
you remove the XXX reference to the bug and simply make juju-gui charms
disappear from the test results?

Thanks!

Gary

[frankban, 2013-01-11 09:23:20]

On 2013/01/10 22:26:11, gary.poster wrote:
> land as is, or maybe with changes, see the discussion of bug 1090716 below.
:-)

Thanks for the review Gary.

> Hi Francesco.  Unfortunately, the side-by-side diffs in Rietveld are broken
for
> some reason.  I just looked at the unified diff.

Yes... it happens sporadically: usually it can be fixed by just re-proposing the
branch.

> This looks great, thank you!  I have two questions.
> 
> First, for my own edification, why does the test FF connect to the self-signed
> cert without complaining?  Does Selenium set that profile up for us or
> something?

Yes it does. One of the default capabilities Selenium uses when creating 
a driver is 'acceptSslCerts': True: whether the session should accept 
all SSL certs by default.

> Second, as a comment, as you might have noticed I put Benji's branch for bug
> 1090716 up for review.  You have an XXX for that bug.  Note that with the
> hidden-GUI branch, the GUI service will still be in the DOM.  It will just
have
> display: none or something like that.  Given that approach/implementation, can
> you remove the XXX reference to the bug and simply make juju-gui charms
> disappear from the test results?

If the element is hidden, it will not be found by Selenium. It seems that
the Selenium parser only finds elements that the user can see. 
See
http://code.google.com/p/selenium/wiki/FrequentlyAskedQuestions#Q:_Why_is_it_...

So, once that branch is landed, the current implementation of 
`get_service_names` should just work as expected, without changes. 
I am going to remove that XXX and re-propose.

[frankban, 2013-01-11 09:25:13]

Please take a look.

[teknico, 2013-01-11 10:49:18]

Land with possible changes.

Very nice, significant improvement to tests. Please find one doubt and one
observation below.

https://codereview.appspot.com/7064068/diff/8001/tests/deploy.test
File tests/deploy.test (right):

https://codereview.appspot.com/7064068/diff/8001/tests/deploy.test#newcode67
tests/deploy.test:67: password.send_keys('admin')
Isn't this overwriting the "password" function parameter? And isn't the code
using the hard-wired "admin" password, rather than the passed in one?

https://codereview.appspot.com/7064068/diff/8001/tests/deploy.test#newcode147
tests/deploy.test:147: 'haproxy', 'mediawiki', 'memcached', 'mysql',
'wordpress'])
This depends on the contents of RAPI's sample.json file, maybe it should be
somewhere a little more visible than just baked in the code.

[frankban, 2013-01-11 11:00:06]

Please take a look.

[frankban, 2013-01-11 11:00:54]

Thanks for the review Nicola!

https://codereview.appspot.com/7064068/diff/8001/tests/deploy.test
File tests/deploy.test (right):

https://codereview.appspot.com/7064068/diff/8001/tests/deploy.test#newcode67
tests/deploy.test:67: password.send_keys('admin')
On 2013/01/11 10:49:19, teknico wrote:
> Isn't this overwriting the "password" function parameter? And isn't the code
> using the hard-wired "admin" password, rather than the passed in one?

You are right, good catch!
Fixed.

https://codereview.appspot.com/7064068/diff/8001/tests/deploy.test#newcode147
tests/deploy.test:147: 'haproxy', 'mediawiki', 'memcached', 'mysql',
'wordpress'])
On 2013/01/11 10:49:19, teknico wrote:
> This depends on the contents of RAPI's sample.json file, maybe it should be
> somewhere a little more visible than just baked in the code.

Done.

[teknico, 2013-01-11 11:03:41]

frankban wrote:
> Thanks for the review Nicola!

Perfect, land it! :-)

[frankban, 2013-01-11 11:40:36]

*** Submitted:

Implemented Selenium tests and login-help.

This branch introduces integration tests based 
on Selenium. This way it is possible to check that
the GUI actually works, is connected to the API 
backend, etc.

Firefox is the webdriver used here, it is installed 
by default in Ubuntu. All the graphical operations 
are executed in memory thanks to xvfbwrapper.

This branch also includes the changes Gary made to 
add a charm option for the login help text.

R=gary.poster, teknico
CC=
https://codereview.appspot.com/7064068