Lester Hedges

Okay, I can get the correct behaviour if I null the `SystemTrajectory` object when calling `System::deleteAllFrames`, e.g.: ``` diff --git a/corelib/src/libs/SireSystem/system.cpp b/corelib/src/libs/SireSystem/system.cpp index 874d0abb1..868a2b883 100644 --- a/corelib/src/libs/SireSystem/system.cpp +++ b/corelib/src/libs/SireSystem/system.cpp @@...

Note that I'm not sure how to fix the code when calling `deleteFrame` rather than `deleteAllFrames`, since there seemingly isn't a way to remove a single frame from the `SystemTrajectory`...

I can confirm that the above fix gives the correct number of trajectory frames for `somd2` when we delete all frames from within the dynamics object each checkpoint. Just need...

No, it appears the the memory still isn't being freed correctly with this approach. For example, this is the profile for a `somd2` repex run. The jumps in memory are...

Zoomed in it looks like this. There is a big jump in memory each checkpoint, then a small drop again, but not back to the baseline. If all of the...

These are the profiles for a standard (non-repex) and repex runs for 100ps, checkpointing every 10ps. The standard run is the profile for a single window, whereas repex is all...

Looks like this doesn't completely fix #330, rather the specific case of SMILES input. Note that the test that fails is passing locally. Will investigate further.

Okay, it works locally using the same environment as for the CI. The test involves generating a mapping with RDKit, so I had assumed that this had changed between versions....

Hmm, it seems to fail when _all_ of the tests are run in one go, not individually. This would suggest some possible file caching issue, but the fixtures for this...

It's definitely this PR that makes the test fail, but I can't see what is causing it. (The test passes in isolation, or even when run as part of the...