[PATCH 2 of 2] test-commit: move test requiring symlink to the end

[Adrian Buehlmann]

On 2012-05-31 16:36, Mads Kiilerich wrote:
> On 31/05/12 16:15, Adrian Buehlmann wrote:
>> On 2012-05-31 16:06, Adrian Buehlmann wrote:
>>> # HG changeset patch
>>> # User Adrian Buehlmann<adrian at cadifra.com>
>>> # Date 1338469397 -7200
>>> # Node ID b16c22df608298617dc23e8f5546541a6ce22194
>>> # Parent  ca1dcebca73680d1b874e51e7515b696c095a4a1
>>> test-commit: move test requiring symlink to the end
>>>
>>> Before, the whole testfile was untested on Windows. Now, only a small portion
>>> is really skipped.
>>>
>> I will do more of these, as soon as patch 1 is accepted.
> 
> It looks like a good start to get partial skipping - thanks.

Which you have killed right away, with the next statements. Thanks as well.

> I have no doubt that we also will need - and will have - "skip a 
> section" instead of just "skip the rest".

I disagree here.

> It might thus not be a good
> idea to mass-adapt tests to the "skip the rest" style - it will be 
> boring work and will make many tests less readable and it will be even 
> more boring work to make the tests readable again.

I have almost expected this reaction.

The net result will be, that no progress is made at all (again) with
regards to increasing test coverage on Windows. Something you have
declared yourself as bitterly required.

Now, on to the bike-shedding:

I think skipping isolated sections is a non-solution and quite some work
to implement and get right.

The tests are pretty heavily woven into each others. Having to track
what side effects affect later tests is way too complicated. No one will
ever do that.

Moving feature-demanding tests to the end of the testfile is

(a) easy to implement support for in run-tests.py
(b) easy for understanding the interdependecies inside a testfile

Most feature-demanding testcases mostly happen to be in the place they
currently are due to the (more or less random) order in which testcases
were added when amending testfiles. Yet they sometimes create hard to
disentangle accidental interdependencies between test cases, most of
which weren't following any larger plan at all.

So again, I think the postulated need for section-wise skipping is false.