Fix `RecentlyUsedContainer` regression

#2115 added typing to the _collections module. Unfortunately, it also introduced a regression that breaks test_urllib3_pool_connection_closed in the requests test suite, as noticed in https://github.com/urllib3/urllib3/pull/2316/checks?check_run_id=3049608329. __setitem__ is implemented like this in 1.26.x:

https://github.com/urllib3/urllib3/blob/ba95e9eac73452d3bccfb5413b00d9a4fe3e4c31/src/urllib3/_collections.py#L61-L74

And it’s now implemented like this in main:

https://github.com/urllib3/urllib3/blob/baba8c2b4aabda8907a777039baa3f8758e4333a/src/urllib3/_collections.py#L133-L155

When len(self._container) == self._maxsize == 0, self._container[key] = value is now called after self._container.popitem(last=False), which is why that popitem call raises an exception.

Applying this hacky diff fixes the requests test:

diff --git a/src/urllib3/_collections.py b/src/urllib3/_collections.py
index 32d5330e..798ba432 100644
--- a/src/urllib3/_collections.py
+++ b/src/urllib3/_collections.py
@@ -139,15 +139,15 @@ class RecentlyUsedContainer(Generic[_KT, _VT], MutableMapping[_KT, _VT]):
                 # size of the pool. Because accessing a key should move it to
                 # the end of the eviction line, we pop it out first.
                 evicted_item = key, self._container.pop(key)
+                self._container[key] = value
             except KeyError:
+                self._container[key] = value
                 if len(self._container) >= self._maxsize:
                     # If we didn't evict an existing value, and we've hit our maximum
                     # size, then we have to evict the least recently used item from
                     # the beginning of the container.
                     evicted_item = self._container.popitem(last=False)
 
-            # Finally, insert the new value.
-            self._container[key] = value
 
         # After releasing the lock on the pool, dispose of any evicted value.
         if evicted_item is not None and self.dispose_func:

@haikuginger What are your thoughts on this? Is it possible to revert to the original logic?

[haikuginger]

@SethMichaelLarson ah good point; I misunderstood. That approach looks sound to me.

[pquentin]

Thank you @haikuginger and @sethmlarson! Opened https://github.com/urllib3/urllib3/pull/2327 to address this