Charles Kerr

OK whoever is building from source, https://github.com/transmission/transmission/pull/4953 has a one-liner that should fix the blocklist memory issue. If you're using `4.0.0` from tarball, you may want to apply that line...

> Note to self when learning to code: Are there enough, or is there any such thing as too much cow-bell/braces. https://www.reddit.com/r/ProgrammerHumor/comments/7sk1rx/a_spy_manages_to_steal_the_last_50mb_of_the_lisp/

I'm leaving this ticket open in case anyone else has heaptrack traces they want to submit. Shaving 650 MB off of Transmission's runtime is what call a pretty productive morning!...

Yep that second snapshot is very similar to the first, 717MB in the blocklist.

I'm not sure how this would work in practice? If all the peers did this, you'd get everyone dropping in and out of the swarm in pulses :smile_cat:

I think I'm going to decline this feature. It has some practical issues we discussed above, but also I'm not sure what problem it really solves and am not sure...

Hmm, failing with this error: ``` [ RUN ] PlatformTest.defaultConfigDirXdgConfigHome /home/runner/work/transmission/transmission/src/tests/libtransmission/platform-test.cc:83: Failure Expected equality of these values: expected Which is: "/home/runner/work/transmission/transmission/obj/tests/libtransmission/transmission-test-mdjvK5/home/user/.config/appname" actual Which is: "/home/runner/.config/appname" [ FAILED ] PlatformTest.defaultConfigDirXdgConfigHome (102...

There's some pre-existing condition that's not right with that file test that I haven't figured out yet. :thinking:

I think it may be something to do with passing a non-nullptr `OVERLAPPED` struct when calling `ReadFile()` but I'm not super familiar with the win32 API. Also it seems there...

> If swapping the blocks "read from closed file" and "read_at from closed file" still fails in the second block it would indicate that. The read_at block is the problematic...