Fixes #236

See #253

[makedoc]

[makedoc]

This bug was pointed out by Jutta Rogal.  Also added regtest so
that bug stays fixed in future

This reverts commit fba71ad4.
Looks like it fails on linux. I have to investigate the problem,
meanwhile I revert

Since isnan() is only available in c++11, I prefer not to backport
this so v2.3

(thanks to Pratyush Tiwary)

[makedoc]

See #247

[makedoc]

[makedoc]

[makedoc]

[makedoc]

it has some problem with travis-ci

(no impact on plumed src)

[makedoc]

[makedoc]

Since matheval evaluator is not thread safe, it is necessary to allocate
one evaluator per threads

Closes #249

[makedoc]

* speed up of sharing and applying in plumed core
the main differences are:
- new member of class atoms to track unique local atoms (i.e. all the atoms needed by plumed action in a domain)
- positions are retrieved only for unique local atoms
- forces are zeroed only for unique local aotms
- forces are applied only for unique local atoms
in a system of 200K atoms on 4 mpi rank with plumed using only 4 atoms the
sharing and applyign time improves by 2 order of magnitudes

* removed leftover printf

* core: make it a bit faster also without domain decomposition

* astyle

* fix-241: unique atoms are update only if:
- particle decomposition has been updated
- domain decomposition has been updated
- replica-exchange step
- neighbour lists have been updated
overall this should make the sharing code even faster

* actionAtomistic have now a list of local atoms that is updated on:
- neighbour list step
- update of the pd/dd list
this allow to simplify the loop for the generation of global local list of unique atoms

* atoms: small tweaks

* fix: reallocate memory less often

* fix: for the scalar case there is no need to generate the list of unique atoms

* style

* few comments

These tests were marked as needing matheval incorrectly

[makedoc]

After d669afb2 there was still
a fix in EffectiveEnergyDrift.cpp when a domain had zero atoms.

This could affect real calculations, so I think this fix should be
backported to v2.3

Fixes #246

(cherry picked from commit 97d7c422)

There was a bug in driver with --debug-pd when, unluckily, the
last domain had by chance zero atoms

(cherry picked from commit d669afb2)

After d669afb2 there was still
a fix in EffectiveEnergyDrift.cpp when a domain had zero atoms.

This could affect real calculations, so I think this fix should be
backported to v2.3

Fixes #246

There was a bug in driver with --debug-pd when, unluckily, the
last domain had by chance zero atoms

[makedoc]

[makedoc]

[makedoc]