Merge pull request #335 from RosettaCommons/vmullig/bgs_parallel
Minor bugfix to the BundleGridSampler.
Fixing an issue with reset mode. No test changes expected.
notify author
notify list [rosetta-logs@googlegroups.com]
Merge pull request #334 from RosettaCommons/vmullig/bgs_parallel
Fix the bundlegridsampler_design_nstruct_mode integration test
Fixes an integration test that's currently failing to finish.
notify author
notify list [rosetta-logs@googlegroups.com]
Merge pull request #330 from RosettaCommons/vmullig/bgs_parallel
Vmullig/bgs parallel
This allows simple parallelization of the BundleGridSampler (without violating the one-mover-per thread/process rule, or the movers-don't-send-MPI-messages rule).
-- The mover now has a "nstruct_mode" option. In nstruct_mode, one set of Crick parameters is sampled per job; otherwise, all sets are sampled every job. (This means that if a parallel job distributor is being used, different sets of Crick parameters will be sampled in different parallel trajectories.)
-- To allow this to be used with other movers that might also have a similar sampling option, there's also an "nstruct_repeats" option. Let's say that you wanted to sample 10 different sets of Crick parameters and 10 different sets of parameters for another mover (for a total of 100 pairwise combinations). You could set nstruct_repeats to 10, which means that each set of Crick parameters is sampled 10 times, and cycle through all parameters sampled by the other mover in jobs where the Crick parameter set is sampled over and over.
-- This is currently covered with a (conventional) integration test using the non-MPI job distributor, but I'd really like to cover this with an MPI-mode integration test, if we could add such a category.
-- I have updated the documentation.
Expected test changes:
-- New integration test bundlegridsampler_design_nstruct_mode.
-- Cosmetic change to bundlegridsampler and bundlegridsampler_design.
notify author
notify list [rosetta-logs@googlegroups.com]
Merge pull request #326 from RosettaCommons/vmullig/remodel_bug
Fixes a bug revealed when RemodelMover is used to introduce disulfide bonds.
The bug is actually in core/conformation/util.cc, in the form_disulfide() function. This function replaces residues with disulfide-linked cysteines (or D-cysteines, or D- or L-homocysteines) and bonds them, but was previously failing to preserve applied variant types. This meant that the cutpoint variants at termini were missing, leading to a crash. In the process of fixing this, I cleaned up some duplicated code, too. I've also added an integration test for placing disulfides with the RemodelMover -- it currently seems to be un-covered by tests.
notify author
notify list [rosetta-logs@googlegroups.com]
Merge pull request #322 from RosettaCommons/vmullig/goto_killer
Removing the last three goto statements in core
These really don't need to be here at all.
(To clarify -- by "removing", obviously I'm not just deleting them. I'm setting up the same program flow using other means.)
No test changes are expected, though I did see failures in tests that were already failing. I'll keep an eye on performance over the next little while -- I don't think this will affect performance in any way but the most negligible, but if it does, we can rework these.
notify author
notify list [rosetta-logs@googlegroups.com]