29 apr. 2019 kl. 14.18 skrev Michael Albinus <michael.albinus@gmx.de>:
> 
>> I decided to maintain it as a derived state
>> because it felt silly to replace O(1) code with O(N), and the
>> invariant is clear enough (stated in its doc string). (Some of the
>> places where the variable is updated are O(N) but less frequently
>> executed.)
> 
> Yes, but is N large enough to experience the difference?

These things are tricky to measure, but obviously inefficient code just doesn't feel right to write. For example, generating a list just to see if it is non-empty, when that could be determined in a more straightforward way.

> My proposal was to use it NOT as a predicate, but as a function
> returning the buffer list.

Very well; here is an incremental patch (to make the differences clear). It's a compromise: the derived state is gone, but there are two functions: one for the list of buffers that need to be polled, and one for whether that list would be non-empty.

By the way, the patch now uses functions from cl-lib, not just macros. Is there any reason not to?