make order features applied in consistent #1177
Conversation
hcoles
force-pushed
the
bug/consistent_feature_order
branch
from
63854cb
to
1ceafb3
Compare
|
Is it an option to use https://stackoverflow.com/a/51945053/643007 ? |
|
That would keep the order consistent, but not make it (easily) predictable. |
|
True. I'm just thinking that it would preserve the order of the input (e.g. the list of features), whereas if the alphabetical output order isn't viable, it needs to be shuffled on the receiving end. It all depends on the intended contract for |
|
Order wasn't meant to be important, and hence hadn't been considered before now. Without the explicit alphabetical order, it would be determined by the order plugins are picked up off the classpath. That would introduces further unpredictability and the possibility of different behaviour for the same codebase when run on a different os/some other factor changes. It's easier going forward to have a simple explicit order that a human can understand. |
|
Got it. It's true that the crash in the interceptors was only coincidentally related to ordering, it was just a consequence of the argument swapper interceptor being buggy. Let's just hope that the new order doesn't cause any other intermittent issues. |
Fixes #1175