What questions should an expert in STL be expected to answer, in an interview

Question

I was looking at a job posting recently and one of the requirements was that a person be a 9/10 in their knowledge of STL.

When I judge my skills, to me a 10 is someone that writes advanced books on the subject, such as Jon Skeet (C#), John Resig (JavaScript) or Martin Odersky (Scala).

So, a 9/10 is basically a 10, so I am not certain what would be expected at that level.

An example of some questions would be found at: http://discuss.joelonsoftware.com/default.asp?joel.3.414500.47

Obviously some coding will be needed, but should everything be expected to be memorized, as there is quite a bit in STL.

In some cases Boost libraries extend STL, so should it be expected that I would be using Boost also, as I may sometimes confuse which function came from which of the two libraries.

I am trying to get an idea if I can answer questions that would be expected of a STL expert, though it is odd that being a C++ expert wasn't a requirement.

UPDATE

After reflecting on the answers to my question it appears that what they may be looking for is someone that can see the limits of STL and extend the library, which is something I haven't done. I am used to thinking within the limits of what STL and Boost give me and staying within the lines. I may need to start looking at whether that has been too limiting and see if I can go outside the box. I hope they don't mean a 9 as Google does. :)

Answer 1

Well, you could walk into the interview and say "I noticed that your posting asked for someone knowledgeable in STL but that term is sometimes used to mean: (1) C++ standard library; (2) the library Stepanov designed at HP; (3) the parts of [1] based on [2]; (4) specific vendor implementations of either [1], [2], or [3]; (5) the underlying principles of [2]. As such, the term is highly ambiguous, and must be used with extreme caution. If you meant [1] and insist on abbreviating, "stdlib" is a far better choice."*

Honestly though since it is a library it is somewhat finite and probably not composeable to nauseating infinitum like a language proper. So I would say any question that had them use some of the stdlib algorithms would be effective to see if they knew them well.

With iterators being an integral part of the stdlib I would also maybe ask them to "Put all the integers in a vector to standard out." I would expect something like:

// thanks to onebyone
std::copy(vec.begin(), vec.end(), std::ostream_iterator<int>(std::cout, " ");

If they write something like the following they probably aren't very familiar with iterators:

for(int i = 0; i < vec.size(); ++i)
    std::cout << vec[i];

Also one interesting thing to look for is if they do using namespace std at the top of their code file. Ask them why, and if they don't say something along the lines of "I use that for short demo code only" or if they put it in a header file, thank them for coming in and send them out the door.

Another aspect of the stdlib is it's heavy use of templates, the person should have a good understanding of basic template programing for type substitution. Maybe ask them to "Write a function that will write all of items of any stdlib container to standard out". I would expect to see something like:

template<typename InputIter>
void Output(InputIter it, InputIter end) {
    while(it != end)
        std::cout << *it++;
}

These are probably not 9/10 questions but interesting ones I think a 2-3/10 should know.

One 9/10 difficulty I would say is to write a derived iostream properly without using the boost stream base classes. But there is probably quite a difference between using the stdlib and extending it...

*(thanks to nolyc on freenode ##C++ for the quote)

Answer 2

Funny -- I don't consider myself a 9/10 in STL (I used to be, but I'm a bit rusty now), and I do fully agree with @joshperry's important terminological distinguo (I've often been on record as berating the abuse of STL to mean "the parts of the C++ standard library that were originally inspired by SGI's STL"!-), yet I consider his example code less than "optimally STL-ish". I mean, for the given task "Put all the integers in a vector to standard out.", why would anyone ever code, as @joshperry suggests,

for(std::vector<int>::iterator it = intVect.begin(); it != intVect.end(); ++i)
    std::cout << *it;

rather than the obvious:

std::copy(intVect.begin(), intVect.end(), std::ostream_iterator<int>(std::cout));

or the like?! To me, that would kind of suggest they don't know about std::ostream_iterator -- especially if they're supposed to be showing off their STL knowledge, why wouldn't they flaunt it?-)

At my current employer, to help candidates self-rate about competence in a technology, we provide a helpful guide -- "10: I invented that technology; 9: I wrote THE book about it" and so on down. So, for example, I'd be a 9/10 in Python -- only my colleague and friend Guido can fairly claim a 10/10 there. STL is an interesting case: while Stepanov drove the design, my colleague Matt Austern did the first implementation and wrote "the" book about it, too (this one) -- so I think he'd get to claim, if not a 10, a 9.5. By that standard, I could be somewhere between 7 and 8 if I could take an hour to refresh (custom allocators and traits are always tricky, or at least that's how I recall them!-).

So, if you're probing somebody who claims a 9, grill them over the really hard parts such as custom allocators and traits -- presumably they wouldn't miss a beat on all the containers, algorithms, and special iterators, so don't waste much interview time on those (which would be key if you were probing for a 7 or 7.5). Maybe ask them to give a real-life example where they used custom traits and/or allocators, and code all the details of the implementation as well as a few sample uses.

BTW, if you're the one needing to cram on C++'s standard library at an advanced level, I'm told by knowledgeable and non-rusty friends that Josuttis' book nowadays is even more useful than my friend Matt's (unfortunately, I've never read Josuttis in depth, so I can't confirm or deny that - I do see the book has five stars on Amazon, which is impressive;-).

Answer 3

It is just a dumb requirement for a job. When hiring you want a good great programmer first, specific knowledge second.

It would be reasonable, at this day and age, to expect knowledge/familiarity/etc with the STL. But unless the job is to reimplement the STL, you don't need 9/10. Even if that is the job, you still just need a great programmer that has lots of experience with templates (making not just using).

For example, for all the answers to "output the integers of a vector", probably the exact same code is generated. Only the version that has been templated to handle any container of any items shows a hint of 'great' vs good (just a hint). ie the ability to abstract.

Anyhow, just go for it. Be ready to use the STL to help solve other problems. Nothing more.

(In fact, for most of the interviews I've been in, the requirement was to NOT use the STL. ie - write a function that reverses a string. My first answer is that there is probably something in the std lib that would do that. Then they say, right, of course, but what if you had to write it yourself...)

Answer 4

I should preface this by noting that I think the same criteria should be applied not only to STL (regardless of which definition you prefer for that), but to many other types of things as well.

From my perspective, simply knowing the existing STL components and being able to apply them well probably should not qualify as a 9/10. Rather, I'd consider that level about 7/10. 8/10 is when the person is able to extend the STL by providing new components that follow its philosophy and fit with existing components naturally and easily.

By 9/10, I'd expect to see somebody who can not only provide new components, but is able to improve some of the existing ones, such as Boost::bind. For 10/10, I'd expect to see this go beyond the rather ad hoc, localized improvements of a 9/10, and moving toward looking at a more architectural level, such as using ranges instead of individual iterators. For a concrete example, consider the difference between Boost's ranges and Andrei Alexandrescu's ideas for ranges. Boost's ranges are handy, useful and convenient, but they change what you type, not how you think. Andre's version of ranges is much more all-encompassing -- an architectural solution that changes how you design and think about code, not just how you type it.

Answer 5

9/10 is quite subjective. I have been asked good questions about STL. These are examples:

• When should you use a deque vs a vector (knowledge of how they are internally implemented is helpful)
• Recognize STL code that uses invalid references, or can end up using invalid references.
• Implement simple operations on different containers, and know where and when to use std::algorithm vs member functions of a container.