GitLab

Removed 'void' declarations from getCutoff/getCutoffDistance methods so that swigInputBuilder.py would correctly wrap return values.

Fix Clang warning about unhandled enum values.

In travis, use multiple cores.

Travis VM's have up through 2 cores. Passing a -j2 flag should make/ctest the builds/tests faster.

Clang had been generating the following warning:

```
ParsedExpression.cpp:123:13:
warning:
      30 enumeration values not handled in switch: 'CONSTANT', 'VARIABLE',
      'CUSTOM'...
            [-Wswitch]
                switch (node.getOperation().getId()) {
                                ^
                                1 warning generated.
```

To get rid of this warning, I added a default case, which does nothing.

XmlSerializer generates XML directly rather than going through TinyXml

Fix residue assignment in PSF parser

(which means that consecutive residues may have the same residue number _and_
name, but are different residues).

GromacsTopFile adds exclusions based on [ exclusions ] directives

Force field XML files can specify CustomNonbondedForces

Added documentation for using a CustomNonbondedForce in force fields.  Also updated documentation on how to specify a tabulated function.

Add virtual sites

Improve unit handling in charmm crd file classes.

Fixed a crash with virtual sites in double precision mode

Fix Amber inpcrd file in the examples directory

the 3rd and 4th columns, making it an invalid Amber inpcrd file (which the new
parser cannot handle).

Fixed bug when multiple virtual sites depend on the same particles (see bug 2019)

Make separate lists for dihedrals and the dihedrals with loaded parameters.

Fix the handling and units of box lengths in CHARMM files

former has a single entry for all unique dihedrals. The latter has a separate
entry for every _term_ of every dihedral (and the latter is the one with the
parameters loaded into it). This way, parameter sets can be reloaded at will
without relying on the hack that createSystem previously employed (i.e., backing
up the dihedral array, clobbering it, then putting it back). This way is cleaner
and safer (although functionally equivalent for typical OpenMM use).

boxVectors, so they should be a property computed from the vectors). Also make
sure that any attempts to change the boxVectors correspondingly deletes the
cached copies of the box lengths so they are recomputed if necessary.