Eric Abrahamsen <eric@ericabrahamsen.net> writes:

> Eric Abrahamsen <eric@ericabrahamsen.net> writes:
>
>> Eric Abrahamsen <eric@ericabrahamsen.net> writes:
>>
>>> Ted Zlatanov <tzz@lifelogs.com> writes:
>>
>> [...]
>>
>>>> The registry has ERT tests, which I thought covered this case.  Can you
>>>> look at `tests/gnustest-registry.el'?  As a first step, can you try
>>>> making tests to demonstrate the problems?
>>>
>>> Will do.
>>
>> Okay, turns out there were tests, but the bit that tests pruning was
>> commented out :)
>>
>> This is the first in a (gradual) series of patches. It does very little
>> except:
>>
>> 1. Adjust `registry-prune' to do what the docstring says: return the
>> total number of entries pruned
>>
>> 2. Uncomment the pruning test and adjust it so it correctly catches the
>> number of pruned entries.
>>
>> Test should fail. Over the next couple of days I'll add another test to
>> check that entries are sorted correctly before pruning, and then take a
>> stab at fixing the pruning itself.
>
> Okay I realized it didn't make too much sense to patch gnus just to make
> it fail, so here's a complete patch, with passing tests, that preserves
> precious entries when pruning. I haven't done the sorting yet, that will
> be a separate patch.

And of course seconds later I realized that the test didn't quite match
the logic of the function, here's a new version.

In essence, here's the logic of pruning, taken from the test:

(let* ((size (registry-size db))
	     (limit (min (oref db :max-hard)
			 (oref db :max-soft)))
	     (keepers (registry-search db :member '((extra "more data"))))
	     (desired-prune-count (- size (- limit (* size 0.1))))
	     (expected-prune-count (min (- size (length keepers))
					desired-prune-count))
	     (actual-prune-count (registry-prune db)))
	(ert-info
	    ((format "Expected to prune %d entries but pruned %d"
		     expected-prune-count actual-prune-count)
	     :prefix "Error: ")
	  (should (= expected-prune-count actual-prune-count))))

I hope that's what was actually intended! Seems to work okay, though.
Please test with byte-compiled files, otherwise it's brutally slow.

Eric