Tootfinder

Proper #security nightmare time.
#LMDB is a database that's designed to operate on trusted input. Upstream has historically rejected all bug reports regarding problems with malformed input.
Py-LMDB project provides #Python bindings to LMDB that are normally built against bundled LMDB. Someone recently started mass-filing "untrusted input" vulnerabilities against py-lmdb, and py-lmdb started #slop - coding fixes to their bundled LMDB. Of course, nobody even bothered reporting most of these bugs upstream, and the one that I've seen reported was rejected as "don't do that".
Py-LMDB supports building against system LMDB, and #Gentoo was doing that so far. However, now we are facing a problem: system LMDB operates under the assumption that it is working on trusted input, while py-lmdb (and its bundled LMDB) operates under the assumption that it may be working with untrusted input. The guarantees no longer align.
If we continue to use system LMDB (and skip all the added slop tests that literally cause Python to crash), then Gentoo's py-lmdb package will now have different input expectations than upstream py-lmdb. And of course we can't just remove that crap because someone added exactly one package (TorchVision, i.e. part of the plagiarism machine suite) depending on it.
https://bugs.gentoo.org/971352