remove 'continuation' mdp option
This is probably a hangover from pre-checkpoint days. At least one user has found it strange and not well enough documented for them to feel they understood it.
When we've had a checkpoint restart to mdrun, we should (and probably do) assume that it is sane. If a user has done a restart via a tpr file (whether generated by grompp, convert-tpr or otherwise) then it is fine for us to assume it is not attempting an exact continuation, and to do things like apply initial constraints (and whatever the shell code needs).
#3 Updated by Berk Hess over 2 years ago
I would like to keep the continuation option. This is the only way to avoid initial constrainting etc when modifying tpr settings. I use that now and then for testing or research purposes. Also we still need the switch in the tpr file to tell mdrun what to do, so then we might as well have it in the mdp options as well.
#4 Updated by Mark Abraham over 2 years ago
Berk Hess wrote:
I would like to keep the continuation option. This is the only way to avoid initial constrainting etc when modifying tpr settings. I use that now and then for testing or research purposes.
When is initial constraining a problem? If it is a continuation, then the positions are already constrained and a small change to another configuration that satisfies the constraints is only a problem if you want a binary exact continuation having changed mdp settings. But I think you only need such exactness if you need to compare to another run that did not have that discontinuity, and have a reproducible run.
Also we still need the switch in the tpr file to tell mdrun what to do,
I think mdrun -cpi finding a file is sufficient to tell mdrun what to do. But if we identify a need, then e.g. having grompp -t set such a switch is plausible.
so then we might as well have it in the mdp options as well.
Not unless we document it well enough that people who don't write the code understand why it exists and when it should be set (or not). For example, http://manual.gromacs.org/documentation/2016.3/user-guide/mdp-options.html#run-control doesn't specify the behaviour when doing mdrun -cpi vs grompp -t vs alternatives with no checkpoints. Having more than one way to do things makes more work for us documenting how it works...