Created attachment 26831 [details] [review]
Test case

Is this something that broke during 0.11 -> 0.12 phase?

Don't think so (if someone wants to fix it for 0.12 that would be nice though).

It definitely didn't break during the 0.11->0.12.  I tried it with 0.11 and
highlight for the test case is the same (actually, I reproduced the test case
because it wouldn't load in 0.11).

I'll see to this one.

OK, great, then it's fine to keep this on 0.13, if you have the time to fix it
before 0.12 it'll be all the better.

Cheers!

The trouble here is that the B task latest end is computed as the minimum value
between the project end and the latest end with successor. But B has no successor.
The latest end computation should also take into account the subtasks latest end
when computing the latest end.

Yay, good catch, great! :)

The description above is a bit wrong, it should be:

Latest finish of B is minimum of project finish and latest START of any successor.

I suspect that the problem really is that the dependency list (the order in
which to go through tasks) is calculated for the forward pass, which puts
children before parents. When we then reverse the list for the backwards pass,
parents are handled before children. I think we should go through children
before parents and do like in the forward pass, i.e. to get the min/max values
from the children for summary tasks.

Should be fixed in CVS.