shared_ptr(&w, _1)

Today I saw some code (from circa 2012) that looked like this:

Widget::ptr 
Widget::instance() {
    static Widget w;
    static Widget::ptr p(&w, boost::lambda::_1);
    return p;
}

As is relatively common in “early modern” C++ code, Widget::ptr is a typedef for boost::shared_ptr<Widget>.

My first thought was that this was a broken attempt to make a shared_ptr that called w when the last copy went out of scope. (“Broken,” because p itself has static storage duration and so the last copy won’t go out of scope until the very end of the program.) I had to put the code into Godbolt Compiler Explorer to figure out what was actually happening.


For those readers who haven’t seen Boost.Lambda before, it’s a bunch of template metaprogramming with the ultimate effect that you can write something like

using namespace boost::lambda;

std::copy_if(
    src.begin(), src.end(), dst.begin(),
    (minVal < _1) || (_1 < maxVal)
);

std::for_each(
    v.begin(), v.end(),
    std::cout << _1 << "\n"
);

as opposed to the pre-modern approach of writing your own functor class from scratch, or the modern approach of writing

std::copy_if(
    src.begin(), src.end(), dst.begin(),
    [](const auto& x) { return (minVal < x) || (x < maxVal); }
);

std::for_each(
    v.begin(), v.end(),
    [](const auto& x) { std::cout << x << "\n"; }
);

(Could I have written auto&& x in both cases to achieve the same effect? I sure could.)


So, if boost::lambda::_1 + 1 is a Boost Lambda function with the meaning “Take whatever arguments you receive and return the first one plus 1,” then boost::lambda::_1 on its own is a Boost Lambda function with the meaning “Take whatever arguments you receive and return the first one with no modification.”

In other words, this is a function with no side effects at all.

So our original code snippet:

Widget::ptr 
Widget::instance() {
    static Widget w;
    static Widget::ptr p(&w, boost::lambda::_1);
    return p;
}

is equivalent to

Widget::ptr 
Widget::instance() {
    static Widget w;
    static Widget::ptr p(&w, [](auto&& p) { return p; });
    return p;
}

which is equivalent in this context to the following snippet — the snippet with which I replaced the original code, eliminating the Boost.Lambda dependency —

Widget::ptr 
Widget::instance() {
    static Widget w;
    static Widget::ptr p(&w, [](auto *) {});
    return p;
}

That is, the original snippet is a perfectly correct (but “clever”) way of writing a function that gives back a pointer to the w singleton instance, as a smart pointer, but a smart pointer whose deleter does nothing. (Which is as it should be, because we never want to delete w.)

TLDR: boost::lambda::_1 is an “early modern C++” idiom meaning “a no-op lambda.”

Posted 2019-02-06