In the original fix I was making convert() throw exceptions.
However, we use a lot failed conversions when reading inputs.
Since catching an exception is much more expensive than inspecting
a returned bool, this was slowing down a lot some parts of the code

In 0ed75e9d I solved the issue
by allowing to throw only a few functions.

Here I take a radically different approach: a new function (convertNoexcept)
can be used like to old one, and returns a bool. The function
convert just calls convertNoexcept and, if false, throws.

In this manner there is no overhead expected at all with respect to the
previous version of the code.

In addition, the modifications are less intrusive. It is sufficient to
replace convert with convertNoexcept in all the places where the returned
bool was checked

This reverts commit 8f0d0628.

This reverts commit 0ed75e9d.

I changed the code in private convertTo* functions to the original
form (return bool). This is actually faster when exceptions are raised
frequently. In this manner, I solve the performance issue on Linux.

See #717

Error is this one:
../maketools/makelinks.sh: line 24: echo: write error: Broken pipe

It's triggered by the comparison with a <(echo) file, since the comparison
might fail before echo completes. If this works I will backport it

This reverts commit cc471663.

minor improvements to OPES_EXPANDED

minor improvements to OPES_METAD

In this manner later we could make the two modules independent

probably these writes should be formatted to avoid inconsistencies
btw compilers

Not used yet, but could be useful

See https://github.com/plumed/plumed2/pull/708#issuecomment-871624325

Otherwise, it is not possibly to copy correctly arrays of plumed objects

There are currently problems building with latest intel

Fix incorrect EXPANSION module in manual

trigger build