Fix ElementHandle::setAs.*Array methods

[dawagner]

Direct accesses to array parameters were incorrectly identified, resulting in
an error when the client tried to set them. This was caused by a programming
mistake in using overloaded methods because T, const T, T & and const T& are different. This is fixed by using the SFINAE principle and correctly
handing all the cases described above.

•  File: parameter/ElementHandle.cpp:L41-77
• <img border=0 src='https://avatars.githubusercontent.com/u/6862950?v=3' height=16 width=16'> Consider using SFINAL to test for size() method existence

/** @return 0 by default, ie for non overloaded types. */
size_t getUserInputSize(...) { return 0; }
/** @return collection size if it has a size() method. does not participate to overload otherwise. */
template <class T, class..., class = decltype(&T::size)>
size_t getUserInputSize(T collection) { return collection.size(); }

• <img border=0 src='https://avatars.githubusercontent.com/u/6430928?v=3' height=16 width=16'> I'm not sure it is worth it: It works but only if I add an overload for std::string (it has a size method but we don't want to consider it as a collection). I can't guarantee, ahead of time, that no future "scalar" type will have a size() method. We would rely too much on future tests begin correctly written.
• <img border=0 src='https://avatars.githubusercontent.com/u/6862950?v=3' height=16 width=16'> You are right, then this should work:

size_t f(...) { return 0; }
template <class T>
size_t f(const std::vector<T> &t) { return t.size(); }

•  General Comment
• <img border=0 src='https://avatars.githubusercontent.com/u/6430928?v=3' height=16 width=16'> @krocard @tcahuzax please review.
• <img border=0 src='https://avatars.githubusercontent.com/u/8655789?v=3' height=16 width=16'> ## Current coverage is 79.54%

  Merging Fix ElementHandle::setAs.*Array methods #351 into master will increase coverage by +0.73% as of 30a7c5f

@@            master    #351   diff @@
======================================
Files          210     210
Stmts         7161    7163     +2
Branches         0       0
Methods          0       0
======================================
+ Hit           5644    5698    +54
Partial          0       0
+ Missed        1517    1465    -52

Review entire Coverage Diff as of 30a7c5f

Powered by Codecov. Updated on successful CI builds.

  

[dawagner]

@krocard @tcahuzax please review.

[codecov-io]

Current coverage is 79.54%

Merging #351 into master will increase coverage by +0.73% as of 83192a1

@@            master    #351   diff @@
======================================
  Files          210     210       
  Stmts         7161    7163     +2
  Branches         0       0       
  Methods          0       0       
======================================
+ Hit           5644    5698    +54
  Partial          0       0       
+ Missed        1517    1465    -52

Review entire Coverage Diff as of 83192a1

Powered by Codecov. Updated on successful CI builds.

[krocard]

Consider using SFINAL to test for size() method existence

/** @return 0 by default, ie for non overloaded types. */
 size_t getUserInputSize(...) { return 0; }

/** @return collection size if it has a size() method. does not participate to overload otherwise. */
template <class T, class..., class = decltype(&T::size)>
size_t getUserInputSize(T collection) { return collection.size(); }

[dawagner]

I'm not sure it is worth it: It works but only if I add an overload for std::string (it has a size method but we don't want to consider it as a collection). I can't guarantee, ahead of time, that no future "scalar" type will have a size() method. We would rely too much on future tests begin correctly written.

[krocard]

You are right, then this should work:

template <class T>
size_t getUserInputSize(const T&) { return 0; }

template <class T>
size_t getUserInputSize(const std::vector<T> &t) { return t.size(); }

[krocard]

👍