Fix ICP cutoff badness #2467
Fix ICP cutoff badness #2467
Conversation
|
It was broken by one of my recent optimisation refactor changes. |
|
(Test failure related to pycodestyle, see #2465 fix.) |
| @@ -1562,7 +1562,7 @@ def generate_triggers(self, lexpression, left_nodes, right, seq, | |||
| ltaskdef.intercycle_offsets.add((None, seq)) | |||
| else: | |||
| ltaskdef.intercycle_offsets.add( | |||
| (str(first_point - last_point), seq)) | |||
| (str(-(last_point - first_point)), seq)) | |||
There was a problem hiding this comment.
Fix here. It is basically a revert of a recent change.
| # There aren't any dependent tasks in other cycles for my_point. | ||
| return my_point | ||
| # There aren't any dependent tasks in other cycles for point. | ||
| return point |
There was a problem hiding this comment.
I was originally looking at this method for clues of what went wrong, hence the minor refactor. There is actually nothing wrong with this method, apart from the rather awkward interface, which is simplified somewhat in this change.
|
Can you explain exactly what problem was caused by this bug? |
|
Yes. |
|
That does sound like a bad bug. |
There was a problem hiding this comment.
Test failure is due to a deprecation notice fixed by #2465.
|
(Travis CI fail is just due to pep8 -> pycodestyle) |
Reported via email by @benfitzpatrick. The new test would fail on current master.