For fun (having read the changelog of Martin's shim package) I overloaded my system so much that suspend took slightly over a minute to complete, at which point I was able to reproduce the same failure as before installing the updated shim package. I have not otherwise experienced the problem (in numerous tests) since installing the updated shim.
This indicates conclusively that at least in my case, the timeout/race is the culprit. It also suggests a way that Martin (or another dev) might be able to reproduce. Find an older/weaker system, launch some VMs and start some compiles, and then run a suspend. Hopefully that will be enough to trigger it.
For fun (having read the changelog of Martin's shim package) I overloaded my system so much that suspend took slightly over a minute to complete, at which point I was able to reproduce the same failure as before installing the updated shim package. I have not otherwise experienced the problem (in numerous tests) since installing the updated shim.
This indicates conclusively that at least in my case, the timeout/race is the culprit. It also suggests a way that Martin (or another dev) might be able to reproduce. Find an older/weaker system, launch some VMs and start some compiles, and then run a suspend. Hopefully that will be enough to trigger it.