I didn't review it deeply, but by just compiling the example, it seems that the switch is implicitly set when using this PR:

~/Develop/popl/build $ ./popl_example
Allowed options:
  -h, --help                   produce help message
  -v, --verbose                be verbose
  -d, --double arg (=3.14159)  test for double values
  -f, --float arg (=2.71828)   test for float values
  -i, --int arg (=23)          test for int value w/o option
  -s, --string arg             test for string values
  -m, --implicit [=arg(=42)]   implicit test

verbose_option  - is_set: 1, count: 1, reference: 0
hidden_option   - is_set: 1, count: 1
double_option   - is_set: 0, count: 0, value: 3.14159
...

but output without any command line flag should be:

~/Develop/popl/build $ ./popl_example 
verbose_option  - is_set: 0, count: 0, reference: 0
hidden_option   - is_set: 0, count: 0
double_option   - is_set: 0, count: 0, value: 3.14159
...

also the count will not go higher than one. When using -vvv the verbose count should be 3, but is 1.

I think you can actually leave the Switch untouched.
OptionParser::parse will clear all options, i.e. will clear the switch values_ and set count = 0, which makes the switch "not set".
Also there seems to be a bug in the current ValueTemplate<T>::set_value function. It calls values_.clear(); and add_value(value);. It should rather call the new ValueTemplate<T>::clear() function, which will also correctly reset the count_ to 0, followed by add_value(value);.
Only problematic case are the assign_to pointers, which should be set to the default value (if there is any)

I merged and removed the clear() from Switch. I need to think about the default and assign_to stuff.
I think a Switch should have a fixed default (false).
This might change the whole class structure (maybe the ValueTemplate class will not be needed anymore)

OK, it's up to you.

I've implemented one more feature that my project needs, a "mandatory" or "required" option: it must be set from command line, otherwise an exception is thrown. It is useful for making a "filename" option, as you cannot set any default value it. Actually it serves as intermediate level between ValueTemplate and Value, and it can be merged with ValueTemplate completely. I'll push it to my branch, but it might be not 100% completed yet.

I merged and removed the clear() from Switch. I need to think about the default and assign_to stuff.
I think a Switch should have a fixed default (false).

That does not work in my environment. If Switch is assignes a remote value at the 1st parse, it is still assigned to true after 2nd parse, regardless of its input.

did you try with the current version? 722a55d

Thanks, works well now!