C++Now video tidbits

You know that post on hidden friends that I was going to write back in April? Well, Anthony Williams just ate my lunch! (“The Power of Hidden Friends in C++”, June 2019.)


The other day I watched Matthew Fleming’s C++Now 2019 talk, “An Alternate Smart Pointer Hierarchy.” He starts with a shout-out to little old me! Thanks. I’m sorry I wasn’t in the audience. (I was at Gašper Ažman’s “Points of Order” at the time.)

The highlights of Matthew’s talk include:

  • @9:10 — Matthew quotes Louis Brandy (“Curiously Recurring C++ Bugs at Facebook,” CppCon 2017): “Is shared_ptr thread-safe? If you have to ask, the answer is no.” Hear hear. (See also: “If you can’t spot the sucker, the sucker is you.”) I have another similar mantra, which I will probably blog about at some point: “What does volatile do? It makes reads and writes really happen. If you don’t know what the phrase really happen means in the context of your computer, then you shouldn’t be using volatile.”

  • @35:50 — If you =delete a constructor to improve your error messages, you must make the deleted constructor explicit; otherwise it will pop in and ambiguate your overload resolution at funny times! Matthew doesn’t show an example in the talk, but here’s one.

  • @50:55 — Working around C++’s lack of covariant return types for smart-pointer return types (as opposed to native-pointer return types).


I’ve also now watched Conor Hoekstra’s extremely prize-winning talk “Algorithm Intuition.” It’s worth the watch! Personally, I would take many of his stylistic recommendations with a huge grain of salt, but it’s a very entertaining and informative session.

I think it might not be coincidental that Jonathan Boccara released two blog posts on std::is_permutation (1, 2) not too long after Conor’s talk hit YouTube. That’s one of the algorithms that Conor claimed one could implement in terms of std::reduce (a.k.a. “fold”). I think what that would look like is something like this:

template<class FwdIt1, class FwdIt2>
bool is_permutation(FwdIt1 first1, FwdIt1 last1,
                    FwdIt2 first2, FwdIt2 last2)
{
    return std::reduce(
        first1, last1,
        true,
        [&](bool acc, const auto& value) {
            if (acc) {
                if (auto ct = std::count(first2, last2, value)) {
                    return ct == std::count(first1, last1, value);
                }
            }
            return false;
        }
    );
}

which is a really terrible \(O(n^3)\) algorithm, despite my clever attempts at optimization. This is one of my pet peeves about many of the “higher-level” STL algorithms: their computational complexity is always optimizing for something, but it may not be for the thing you really care about. std::is_permutation optimizes for memory usage at the expense of running time. std::sort optimizes for number of comparisons, at the expense of number of move-constructions. That kind of thing.

However, it is still pretty cool that both GCC and Clang are able to constant-fold this entire function away at compile time.


I missed “Algorithm Intuition” at C++Now because I was planning to go see Ryan Dougherty’s “Experiences in Teaching Modern C++ to Beginners” instead. And then I missed Ryan’s talk because I had just finished my own “Trivially Relocatable” talk and ended up in a deep hallway discussion that ended up consuming the whole time slot. But I caught Ryan’s talk on YouTube, and I guess I didn’t miss too much — it feels kind of light on takeaways, and what takeaways there are mostly just reinforce my preconceived notions. For example:

  • Yes, we still have to teach pointers and arrays. Students are going to see them in real life, so they have to know what they mean and what they do. (And how C++ arrays are different from Java arrays!)

  • Students mostly learn by example, and by question-and-answer.

  • Students with no prior programming experience will omit curly braces and end up with confusing bugs, unless you set an immaculate example. Curly-brace all your control flow structures, people, even in slide code!

I thought it was counterintuitive that Ryan prompts his students (non–CS majors, all) to watch CppCon talks and give reports on them. I would expect the vast majority of CppCon talks to be way over the head of the average CS 101 student. (But maybe less so at CppCon 2019, thanks to its Back to Basics track!)

Posted 2019-07-02