> Radix sort looks fast, with its O(n)O(n) worst-case time complexity. But, if you're using it to sort binary numbers, then there's a hidden constant factor that's usually 32 or 64 (depending on how many bits your numbers are). That's often way bigger than O(\lg(n))O(lg(n)), meaning radix sort tends to be slow in practice.

The constant factor in radix sort isn't the number of bits, it's the number of digits. Since one typically sorts on a computer using 8-bit 'digits', that's a constant factor of 4 or 8 (not 32 or 64).

Of course in the real world for problems where sorting is actually the bottleneck and not just something you want not to kill your app's performance, you end up with things like timsort that destroy all these in practice.

I think it's important to mention that counting sort and radix sort are not "true" comparison-based sorts: they have additional requirements on the input data.

Who is still asking candidates to talk about sorting algorithms?

It is just trivia and has little to no bearing on if the candidate can actually do the job (unless the job relates to the runtime of sorting algorithms, of course).

Original author here, happy to answer questions about data structures, algorithms, and coding interviews!

Curious: did people notice that you can click on each algorithm and click the blue button to get a detailed write up of how it works? (Or is that too hard to find?)

There is also sorting networks. If you have to sort short list of items, it's probably the best solution around.

https://en.wikipedia.org/wiki/Sorting_network

The best real sorting algorithms are often hybrid.

Would be nice to see mention of parallel sorting algorithms. These have basically linear speedup and you can do them on GPUs.

Obvious disclaimer that you should never care about your sorting algorithm choice until you need to. Performance is more often IO or memory related than compute. Tech interviews are daft, etc.

Best case for heapsort is actually O(n). Build heap always takes O(n). Then e.g. when all keys are the same, the max-heapify call will take O(1) instead of O(logn).

feature request: make O(k) bright red, like O(n^2). As it stands, it looks nice and pleasant like O(1) but it's honkin' terrifying.

There are two situations where I wouldn't harass my coworkers for asking sort related questions.

One, I've asked. Not implementing sort, but implementing complex compare() functions for sortable values. There are many things I can accept as flaws in a coworker but inability to deal with 3 conditionals at once isn't one of them. And I've had people (interviewed or worse, worked with) who can't sort by last name, first name, age.

The other is the "I gave you code now tell me what's wrong with it" answer but that's not writing a sort algorithm, that's detecting that someone needs a better one.

Personally, I'd be all too happy if we stop interviewing people expecting answers that would never pass a code review. I'd say it's hypocritical but it's not even that. It's just wrong headed, and gives bad information to the candidate.

I see your point, and usually these work best if the interviewer can prove that such a thing is needed. For instance, many jobs might require sorting of data that cannot fit on disk (think for instance of loading a massive file that you want to parse, and sorting might help you in processing it), in which case knowing merge sort (the disk variety) can help. So yes, the onus is on the interviewer I think.

I'm less worried about understanding of sorting algorithms than I am about understanding of patterns. With any team at scale my hardest challenge is about patterns. More importantly why a pattern is used. Efficiency of a sorting algorithm can be refactored. Is the implementation of the sorting composable? I'll hire without question someone who understands patterns and composablilty over someone who can site an ideal solution. Seriously at what cost does readability over complexity really mean anything to consumer value. I understand that in very limited scope computational efficiency is imperative but let's be pragmatic and say good programming and understanding of best practices trump perfect solutions.

I think the intention is that the job will relate to the running time of other algorithms and that sorting things is a toy problem you can use to test whether they know anything about that subject in general.

There may be better ways to assess that, though. For example, you could ask them to give one or two examples each of algorithms that are O(1), O(log N), O(N), O(N log N), and O(N^2), and O(something worse than N^2).

Sorting is such a foundational problem in computer science that pretty much every job will end up relying on the runtime of sorting algorithms. Not every job requires being able to write a bulletproof dual-partition QuickSort, of course, but knowing complexity bounds is important, even if you’re just calling your standard library’s implementation: it places fundamental bounds on what things you can improve.

Disagree. If you think that, what do you ask? "How would you Google or Stack overflow for the answer?" That's hardly going to provide any useful signal. At the end of the day you need to ask something that shows evidence there candidate knows something about efficiency tradeoffs.

Some people like that sort of thing. Mostly the "academic" types. The memorizers. There are people who got through their higher education by using memory instead of wit. I know some of those people, I went to school with some of them, they were always studying and ramming as much "data" into their brains as possible... while I was getting through with minimum of effort by just coding a lot and trying to think about things in my own way.

The memorizers in this industry often get jobs higher up the tree and therefore they are highly represented in the hiring practices.

Anyways, I completely agree with you. In my opinion it is essential to KNOW that you DON'T KNOW (i.e. you don't remember how exactly an RB tree works, you just know there is such a thing). Everything else is just a quick Google session away.