Re: [cpp-netlib-devel] localhost_tests failures

[Dean M. B. <mik...@gm...>]

Hi Kim,

On Wed, Sep 3, 2008 at 2:29 PM, Kim Gräsman <kim...@gm...> wrote:
> Hi Dean,
>
> On Wed, Sep 3, 2008 at 07:47, Dean Michael Berris
> <mik...@gm...> wrote:
>>
>> I think it's all about whether we want to live with the test failure
>> in Windows based on a perfectly valid test that exposes a bug in
>> Python in Windows. In this case I think it's alright for us to live
>> with it -- we can mark it as an expected failure if we're in Windows.
>
> I don't see how that's an improvement over saving test.xml with
> LF-only line endings, and get consistent test results...? I understood
> Divye's point to be that if we did that, we would mask something else.
>

It's not about changing the test data to make sure the tests run, it's
more about making sure we get the correct data that we expect from the
HTTP Server -- and if I requested for a file that has CRLF line
endings in from the server, that I would get the exact same file if I
read it from disk in binary mode.

> Wait... OK, I think I see what's happening:
>
> 1) test.xml always contains CRLF, even on Linux
> 2) On Linux, the Python server serves test.xml with CRLF endings intact
> 3) If we change test.xml to LF-only, we have worse test coverage for
> line ending independence
>
> Is that it?
>

Yes, we want to make sure CRLF is preserved -- and that the file is
intact when sent over the HTTP Server, and received by the client.

>> Perhaps a patch to the test to reflect this would be very welcome. :)
>
> It's included in the patch I sent last night -- in Win32
> configurations the test code strips all CRs from the source text
> document. But maybe a dedicated test would be better to demonstrate
> this.
>

I would like to see though a windows-specific flag which sets the
expected failures to 1 in the localhost_test case that checks for file
lengths.

I haven't taken a look at your patch yet, so I'll see what happens
when I get to it later.

HTH

-- 
Dean Michael C. Berris
Software Engineer, Friendster, Inc.