[twill] Bug in FakeResponse

[Jacob Hallén]

Ahh, I see. The response object as defined in mechanize rewinds the cursor in 
the file as a side effect of going through a copy. Personally I think that is 
not the best of designs. A copy of an object should really be a verbatim copy 
and not something that is just similar to the original. 

In any case, the FakeResponse needs replacing.

Jacob Hallén

torsdag 28 juni 2007 23:32 skrev John J Lee:
> On Wed, 27 Jun 2007, Jacob Hallén wrote:
> [...]
>
> > As it turns out, there is a bug in mechanize as well.
>
> I don't see a bug, actually.
>
> I'll repeat: the mechanize code here is nasty, do not try and use your own
> response object.  I suspect you're missing the fact that mechanize
> response objects have independent seek pointers:
>
> import copy
>
> import mechanize
>
> r = mechanize.make_response(
>      "blah", [], "http://example.com/", 200, "OK")
> data = r.read()
> assert r.read() == ""
> r2 = copy.copy(r)
> data2 = r2.read()
> assert data2 == data
> assert r2.read() == ""
>
>
> If you still think there's a bug, can you provide a specific failing test?
>
> > My full fix for the problem is as follows:
> >
> > 1. Re-implement FakeResponse:
>
> Bad idea!
>
> > 2. Modify the set_response method of the class RobustFactory:
> >
> > You have to make the copy before you do the response.read(), otherwise
> > you have an exhausted stream in the copy you make.
>
> Nope.
>
> > Apparently analyzing forms worked at some point in time. I wonder how it
> > could be broken in two different places in two different packages.
>
> The only bug I see is the one in twill, which is really my fault for the
> deeply obscure mechanize code here (the whole seek thing "fills a much
> needed gap", as does some of the backwards-compatibility of various kinds,
> which is one of the reasons has taken such a long time for mechanize to
> get to stable).
>
> > I think some tests would be a good idea.
>
> Any tests in particular?  You will have noticed that there are "some
> tests" already.
>
> Certainly there aren't enough tests (especially functional tests).  I wish
> I could spend more time working on it...
>
>
> John