Files with gz extension are transparently compressed and decompressed.

Closes #17

Added diagnostic

I fixed it so that it *Really* works with both internal and external
lapack/blas. The only drawback (I could not do it better) is that
both lapack and blas should be either internal or external.

(thanks to Toni Giorgino)

I fixed a missing include.

Original comment:
Fixes in backup output files - cant rename open file under windows - print reason for failure - remove leak of FILE*s

(cherry picked from commit 448e84fc)

(thanks to Toni Giorgino)

I fixed a missing include.

Original comment:
Fixes in backup output files - cant rename open file under windows - print reason for failure - remove leak of FILE*s

Changed the internal enum in switching function (for clarity)

Closes #43.

Adding STRETCH flag to switching functions they go smoothly to zero
at D_MAX

Closes #52.

When adding objects to Communicator class so that they can be used in MPI
operations one should add the conversion operator to both Data and
ConstData. Otherwise it is not possible to e.g. Isend() an object

One can e.g. specify ATOMS=10-1:-3

Closes #45

The fix to #39 was incorrect since adding an int to a string
does not print it.

I also removed the redundant check that index is positive (it is unsigned)

I removed an assert and added an if then plumed_merror

close #39

I removed an assert and added an if then plumed_merror

close #39

Two fixes:
1. with autoconf, the proper macro for exceptions is __PLUMED_HAS_EXCEPTIONS
I fix it (breaking the compatibility with existing Makefile.conf)
because I think this is not used by anyone actually
2. uses macro __PLUMED_HAS_EXECINFO to decide whether execinfo.h should
be included or not

Neighbor list in multicolvar now has three input parameters

TOL - all values less than this are ignored
NL_TOL - all values less than this are ignored until next neighbor list time
NL_STRIDE - frequency with which to update neighbor list

Neighbor list has been tested for DISTANCES and ANGLES.  Still needs more
testing

I set it to 2 instead of 2.0, so that it will not be necessary
to update it

Reference coordinates are saved already shifted to their average.
Weights are stored already normalized.

A regtest has been fixed (it contains numerical derivatives which
slighlty changed with this commit)

class OptimalAlignment is not used anymore in RMSD

I did it in the boring way - full derivatives of rotation matrix.
It seems to work now, but still breaks some test. It is probably
some numerical issue, I will investigate before merging.

It is slower than the align==displace version (~ a factor 1.5),
but I think it is more than acceptable.

For a number of complex multicolvars eg. Local_q4 it is useful
to store the quantities calculated in one multicolvar and to use them later.
I had written functionality for this in a somewhat ad hoc manner intially and
as such there was much duplication of code and non uniformity in implications.
This resolves these issues.  Data is now stored more consistently and it is possible
to use a low memory version of this functionality at all times.

The nature of the backbone is now specified by a class called MolDataClass.
A warning is given about GLY residues

Removed many scans of the output buffer by keeping trace of buffer length.
Time needed to write a long line with a sequence of file.printf() is
now linear with line length. Before it was quatradic, causing problems
especially when logging huge atom lists at startup.

* Better algorithm for int powers
* Simplification when 2*N=M (which is the usual case)

I replaced everywhere it made sense the new Communicator class
syntax. This commit is not strictly necessary, but provides
examples of the simpler syntax so I think should be merged

Now all the MPI wrappers come in two flavors. They can be used
with a pointer as usual, e.g.

double*data;
comm.Sum(&data[0],n);

or they can be used with a reference to a single number

double x;
comm.Sum(x);
// same as comm.Sum(&x,1);

Notably, the second version also works with std::vector and std::string
and automatically detect the size, e.g.

std::vector<double> v;
comm.Sum(v);
// same as comm.Sum(&v[0],v.size());

Furthermore, everything works transparently with Vector and Tensor
either as single objects or elements
of vectors or of C arrays:

std::vector<Vector> v;
comm.Sum(v);
// same as comm.Sum(&v[0],v.size());
// same as comm.Sum(&v[0][0],3*v.size());

All the regtests are satisfied

This allows to perform test on minimal image from outside the class

There was a potential bug when stripping blanks from short strings (1 char)

I think the way pbc were applied was not correct (Fix #19).

This fixes #23
Notice that it slows down (with icpc) generation of Gaussian numbers
by a factor 2.5. I think it is better now to have reproducible regtests.