[PATCH 1 of 2 stable] fileset: add tests of generated working copy states

[Matt Mackall]

On Fri, 2015-01-23 at 17:34 +0000, Martin von Zweigbergk wrote:
> The disadvantages I can see:
> 
> * It's easy to leave a stray "BROKEN" marker after fixing the issue. This
> is a little annoying. I don't know if some infrastructure could help with
> it.

Actually, I think the biggest issue is that when you fix something
(possibly as a side-effect), you now have false-positives from the test
suite to dig through.

I'm at -0 on this sort of thing. If people are actually working through
issues in the area, I'm lately inclined to take documented-broken tests.
But I don't really like to take them on their own unless I think
someone's going to follow up soon.

(And I'm definitely -1 on tests that actually FAIL, rather than document
incorrect behavior.)

Perhaps we can augment the test suite to be more explicit here. For
instance, we could flag entire blocks including descriptions as
known-broken:

---
!This should do something:
!
! $ hg something
! abort: huh?
! [255]
!
---

..and then it'd be really obvious from the test output when you'd hit
this, and also obvious from diffs if you hadn't fixed up the whole
block.

-- 
Mathematics is the supreme nostalgia of our time.