Use unique_ptr for Pimple Transform

[blowekamp]

Remove raw pointer usage for the pimple transform with unique_ptr which enforces the logic of ownership to the heap allocated object.

[N-Dekker]

Thanks for asking me to have a look!

The && are not really necessary here, you may just pass the unique_ptr "by value". Example (https://godbolt.org/z/h4cs6fzWo):

void MyFunc(std::unique_ptr<MyObject> myObj)
{
}

int main()
{
    MyFunc(std::make_unique<MyObject>());
}

Of course, you can still use &&, but pass by value is more common, for unique_ptr. My two cents.

[blowekamp]

Good to know that is the conventions. The move is still required with the following:

int main()
{
    auto uptr = std::make_unique<MyObject>();
    MyFunc(std::move(uptr));
}

I suppose there is no more cost to moving the unique_ptr object that just copying a ptr.

[blowekamp]

@N-Dekker After following your suggestion of changing the r-value reference parameter to pass by value There are some compilation errors on linux system ( before others were canceled ):

https://open.cdash.org/viewBuildError.php?buildid=8664641

/usr/include/c++/9/bits/unique_ptr.h:79:16: error: invalid application of 'sizeof' to incomplete type 'itk::simple::PimpleTransformBase'

Any suggestions?

[blowekamp]

@N-Dekker After following your recommendation of passing std::unique_ptr by value instead of r-value, I are getting errors on most compilers related to the pimple object type being only a forward declaration. Any further suggestions or recommendations?

[N-Dekker]

@N-Dekker After following your recommendation of passing std::unique_ptr by value instead of r-value, I are getting errors on most compilers related to the pimple object type being only a forward declaration. Any further suggestions or recommendations?

Sorry, I did not know about this problem. I've been trying to reproduce the errors at https://godbolt.org/z/cP7Pas5zT but so far, I don't see them there. You see, I just tried:

class ForwardDeclared;

class MyClass
{
void Func(std::unique_ptr<ForwardDeclared>);
};

Anyway, if there is no other solution, just revert my suggestion to pass std::unique_ptr "by value". 🤷

[blowekamp]

So this case is a bit more interesting. Here is the compilation error reproduced:
https://godbolt.org

If the G function's argument is changed to a r-value reference then the error is no longer there.

Thank you for your response.

[N-Dekker]

So this case is a bit more interesting. Here is the compilation error reproduced:
https://godbolt.org

It does not show up at my side. Can you please select Share >> Short link in the Godbolt site, and share the link? Or otherwise just copy-paste your code?

[blowekamp]

Correct link: https://godbolt.org/z/7fMaxG9bd

[N-Dekker]

Correct link: https://godbolt.org/z/7fMaxG9bd

I see, you pass the unique_ptr-to-an-incomplete-type forward from F to G, and in that case, it appears that it must be passed by && reference, instead of by value. This seems to break the rule that unique_ptr must be passed by value, indeed.

Looking at https://www.reddit.com/r/cpp_questions/comments/gpfij5/passing_unique_ptr_by_value_vs_by_reference/ they all say unique_ptr should be passed by value, but it looks like you proved them wrong!

I can still ask around to see if your case is indeed an exception to the rule. But in the mean time, do whatever you need to avoid those compiler errors! 👍

[N-Dekker]

Off-topic nit-pick: the C++ idiom I think this is PR about is officially called "Pimpl", either in upper or lower case, but without the letter "e". (See https://en.cppreference.com/w/cpp/language/pimpl or http://wiki.c2.com/?PimplIdiom) So you might still consider replacing any "Pimple" with "Pimpl".