I think Rod's solution to this problem was to provide a test webapp, which
can provide feedback in much the same way as Watchdog.
----- Original Message -----
From: "Vincent Massol" <[EMAIL PROTECTED]>
To: "Morgan Delagrange" <[EMAIL PROTECTED]>;
<[EMAIL PROTECTED]>
Sent: Monday, August 27, 2001 11:16 AM
Subject: Re: [httpclient] unit test suite
> I understand very well your point but I don't think you understood mine
...
> :-)
>
> What I am saying is that unit tests are fine-grained and can exercise all
> needed tests. For example I don't see any unit test to verify that when
the
> server returns 2 cookies in 2 different HTTP headers it works fine ! ....
> Why ? Because if you test this by hitting real servers it is going to be
> *very* hard to simulate. In principle you'll want to exercise the full
> client HTTP RFC and this is going to be impossible by hitting real
servers.
> It is however very easy to do with Mock Objects.
>
> I alsol agree that we also need some coarse grain functional tests which
hit
> real servers but it comes as a complement to the unit tests.
> Thanks
> -Vincent
>
> ----- Original Message -----
> From: "Morgan Delagrange" <[EMAIL PROTECTED]>
> To: "Vincent Massol" <[EMAIL PROTECTED]>;
> <[EMAIL PROTECTED]>
> Sent: Monday, August 27, 2001 5:01 PM
> Subject: Re: [httpclient] unit test suite
>
>
> > I don't think I got my point across here. I'm all in favor of having
unit
> > tests. I just think that the unit tests should bang on real servers,
not
> > mock objects.
> >
> > ----- Original Message -----
> > From: "Vincent Massol" <[EMAIL PROTECTED]>
> > To: "Morgan Delagrange" <[EMAIL PROTECTED]>;
> > <[EMAIL PROTECTED]>
> > Sent: Monday, August 27, 2001 10:57 AM
> > Subject: Re: [httpclient] unit test suite
> >
> >
> > > Hi Morgan,
> > >
> > > There are 2 kinds of tests :
> > > * unit tests
> > > * functional tests
> > >
> > > I was referring to the unit tests which are supposed to fully test the
> > code
> > > logic. It is also easily automatable and need to be run every time you
> > > modify something to ensure you don't break anything. It is quite
> > > fine-grained tests compare to functional tests.
> > >
> > > Functional tests are harder to automate but can be done (see Cactus).
> > > However, we'll have to say which test can be run on which server
because
> > as
> > > I saw in the comment of some test Tomcat 3.x does not support OPTIONS
> > > request for example so these tests should not be executed on Tomcat
3.x,
> > ...
> > > Also we need to control the page that is hit by the test so we'll
> probably
> > > need to provide some test pages to deploy to the test server.
> > >
> > > I don't see what would be the problem of providing unit tests ! The
more
> > > tests the better (provided they are executed all the time and
corrected
> > when
> > > modifications or new APIs are introduced). I am also a partisan of
test
> > > first .... but I won't ask you to do that ... (although you should try
> one
> > > day, I'm sure you'll like it) :-)
> > >
> > > Thanks
> > > -Vincent
> > >
> > > ----- Original Message -----
> > > From: "Morgan Delagrange" <[EMAIL PROTECTED]>
> > > To: <[EMAIL PROTECTED]>; <[EMAIL PROTECTED]>
> > > Sent: Monday, August 27, 2001 3:00 PM
> > > Subject: Re: [httpclient] unit test suite
> > >
> > >
> > >
> > > Vincent Massol wrote on 8/25/01 12:02 pm:
> > >
> > > >I think that if we wish to
> > > >provide a strong and reliable
> > > >http client framework we
> > > >ought to have a good suite of
> > > >unit test that we will enrich
> > > >as we go along and find bugs.
> > > >I have had a brief look at the
> > > >current tests and we really
> > > >need to refine them.
> > >
> > > Agreed, we could definitley use more tests.
> > >
> > > >They also rely on the
> > > >installation of a web server
> > > >on your machine which I don't
> > > >like too much as the test will
> > > >pass and fail depending on
> > > >which one you use ...
> > >
> > > Isn't that an advantage? After all, wouldn't you want to try running
> your
> > > tests on the actual server you will be communicating with?
> > >
> > > >Instead I propose to simply
> > > >use Mock Objects to provide
> > > >a reliable suite of tests,
> > > >which is independent of the
> > > >web server installed. The
> > > >principle is simple: we simply
> > > >need to create a mock
> > > >Socket class and a way to
> > > >pass this mock to the
> > > >HttpClient class. The idea is
> > > >that it is possible from our
> > > >test case to set the
> > > >behaviour that we expect
> > > >from the mock Socket class,
> > > >like what data it will return
> > > >on the output stream, ...
> > > >there is absolutely no logic in
> > > >the mock implementation,
> > > >just setters and getters.
> > >
> > > I think we want to be testing real servers, so we can uncover a wider
> > > variety of problems.
> > >
> > > >I'll write one ASAP but I'd like
> > > >to know if I can be voted in ....
> > > >I am yearning to go forward
> > > >especially as Cactus now
> > > >relies on HttpClient.
> > >
> > > You don't need our approval to be a committer, but I don't know if
this
> > > particular task is a good idea.
> > >
> > > - Morgan
> > >
> > >
> > > _________________________________________________________
> > > Do You Yahoo!?
> > > Get your free @yahoo.com address at http://mail.yahoo.com
> > >
> > >
> > >
> >
> >
> > _________________________________________________________
> > Do You Yahoo!?
> > Get your free @yahoo.com address at http://mail.yahoo.com
> >
> >
_________________________________________________________
Do You Yahoo!?
Get your free @yahoo.com address at http://mail.yahoo.com
