Hacker News new | comments | ask | show | jobs | submit login
The Worst Programming Interview Question (nomachetejuggling.com)
113 points by talhof8 on June 27, 2014 | hide | past | web | favorite | 184 comments

Hypothetically, if when asked this question in an interview I were to reply "I'd look up the canonical solution on the internet and just implement that. I have better things to keep in my brain than linked-list algorithms, and there's no way I'm going to beat decades of phd work off the top of my head", what would happen?

Obviously it depends on the job I was interviewing for. Let's assume I'm not angling for a position as lead on a distributed load balancing layer or something.

I suppose if they're looking to see my problem solving workflow then I'd look like kindof a dick. But then, as the article points out, so would they.

That's probably fine if you're going for a junior position where your job is to bang out code for designs that others have come up with.

But if you go up a few steps then suddenly the problems might not be that clearly defined and the set of possible solutions can be huge. A person who is constantly relying on Google to come up with solutions is probably not going to be that useful in design discussions.

I'm not defending the particular question, but on-your-feet problem solving is a real and valuable skill.

I agree: the hard mental work is in recognizing that a fuzzy, client-provided problem description is merely a disguised form of a well understood problem -- ie, getting it to the point where you can google easy solutions at all.

But then in that case, the problem as given is still bad, because it gives that part away! If you want to test this ability, you shouldn't frame the problem in terms of linked lists. As another poster here suggested, you should ask something like "The customer gave a list of everyone and who they report to. How would you sanity check the hierarchy?"

That then requires the candidate to identify cycles as something they'd want to look for, and then this problem as a special case of "find a cycle in a linked list".

Oh I agree. That's exactly the stuff I keep in my brain instead of linked list algorithms, which have a mathematically proven optimal solution. Remembering them when I don't use them regularly would be a waste of brainpower which could be better used for solutions to business/UX/architecture problems, which one cannot solve with pure logic.

I am not so sure. It's not just about solving the problem, but sometimes it's also the willingness and mentality to tackle something that's outside your comfort zone, as well as the approaches.

For example, given the loop in linked lists question, even if a candidate is not able to solve it completely, a reasonable approach would be to break the general case to more specific, trivial cases, such as: what if the loop has known length, what if the loop starts at a known location, etc. Interview questions are not binary decisions, you can learn a lot from a candidate just by seeing how they respond to a question.

(Just to be clear, as this is a permanent and public record, I'm perfectly capable of doing this kind of thing. It's not really an unreasonable question to ask. I'm just being facetious because it's friday afternoon.)

A great response would be to just say "hold on", whip out your phone and Google the answer, show it to them and finish up with "How long did it take your other candidates to complete that task? next..."

Point being, you're demonstrating initiative and, that old cliche, thinking outside the box. Think Will Smith in Men In Black.

Some interviewers might be impressed with your lateral thinking, others might be annoyed at your attempt to be a 'smart ass' and avoid the intention of the question. But that's always the problem with these type of interview questions, it depends so much on the expectations of the interviewer.

I think it would be funnier to silently transcribe the code from the phone to the whiteboard, and then just sit back down and look at them while you wait for a response.

My response would be, "OK, walk me through the code, show me how and why it works, or if it doesn't (code from teh Internets is buggy), fix it." Anything higher than a junior position requires maintaining code as much as, or more than, writing it.

Maybe this points to a better class of interview questions: the interviewer supplies the broken algorithm, the interviewee supplies the bugfix.

"Anything higher than a junior position requires maintaining code as much, or more than, writing it."

I think that applies to junior positions, too.

You're correct, but generally when critical thing X is broken and you need it working for the customer, you give it to your more experienced devs, not the college intern.

True, but when I'm in a more senior position, we're less often in a situation where critical thing X is broken.

"oh, you can use google. Surely none of the other candidates could think of that"

The professionals use StackOverflow.

If an interviewee had the stones to try that in an interview, I'd probably have to at least give them a golf clap.

I did something like this recently. The question was more around increasing efficiency of an algorithm though, and on top of what you said, the other part of my answer was more like "I wouldn't waste my time as long as performance was acceptable until it became an actual problem at larger scale. I'd move on to shipping other features."

It felt that they were interviewing for a computer scientist despite advertising for a software engineer.

Needless to say, we were mutually not a good fit.

One day interviewers will realize this. Caching into your mind what's one stackoverflow away can often be an invalid optimization.

>> Obviously it depends on the job I was interviewing for.

In my experience, it often does not depend on the job. Interviewers ask such things even if the job does not require it because ... ultimately because they are not good interviewers. If you get asked the question, the best bet is to just answer it or else the interview goes downhill from that point on.

But that solves the problem no? its like when I needed to limit a tool I wore to so many api requests in an hour I knew I needed a queuing algorithm but I don't keep that in my head a couple of quick google queries and a quick look on CPAN problem solved.

I tried that once and was told "we don't use freeware here".

ah so what do you use instead of TCP/IP then :-)

This question has an "a-ha" moment, but it is not required to solve the problem adequately. I wholeheartedly disagree with this statement:

> People like to pose puzzlers to “see how people think” but that’s nonsense in the case of puzzler questions. You can’t reason your way through a puzzler, that’s why it’s a puzzler

There are plenty of ways to solve this problem. They're not all optimal, of course, but you don't ask these questions to get the optimal solution, you ask these questions to understand how the person reasons about them.

This more falls on the interviewer than the candidate. It's upon the interviewer to encourage a dialogue about the problem and talk aloud through it.

Yes, some qualified candidates are bad at this. I'm sorry to hear that, but it can be practiced. Being good at interviewing can be learned. It's not terribly difficult. My number one pro-tip? Never do this:

> If you don’t have the a-ha moment in the interview, you won’t get it, and a good chunk of your brain will be devoted to thinking “oh shit I’m blowing this interview” rather than focusing on the question at hand.

Never think that you're blowing the interview. I got a job at Google and I know for a fact that I didn't get the optimal solution to some of my questions. It's not about the solution, it's about the process.

Google and others have largely given up on these types of brain teasers.


This is... debatable. It depends on what you consider a puzzler. I will tell you that I was asked questions similar to the one referenced in the parent article. It is a programming question that can be answered precisely with code, and that code can then be analyzed for time and memory considerations.

I was not asked silly questions like "how many golf balls could you fit in a school bus" or "how do you move a mountain 3 feet to the side". These are the types of questions he explicitly references in the article.

edit: I will also add that there are differences between eng and non-eng interviews. I'm not familiar with the non-eng process really.

I used to have a math puzzler that I'd ask. Almost nobody got it right. One guy got close, and we ultimately hired him. He asked for the full answer, and I gave it to him. He told me I was wrong. I pointed out that I had asked this question at least 50 times, and knew it in and out, and that he should reconsider. Then he sent me a spreadsheet proving that I was wrong. Whoops.

You can't tease us with such a question and not tell us what it is! Or perhaps you're still using it and don't want to reveal it publicly? :(

It's a question that checks intuition on probability and the law of large #s. I don't want to give too much more since I do pull it out on the very rare instance where someone is very bold about their stats knowledge.

To your credit, you are among the very few who would admit this kind of error and you likely took it gracefully.

But how many times did you ask this question without actually verifying that YOU could answer it? And what does that mean about your process for generating and grading interview questions? And why did it have to be the interviewee who told you?

Maybe interviewees should be more confident, but maybe interviewers should be a little less self-confident. This is a huge cultural problem for the industry, but it's right in our blind spot.

And we have to ask why there are less women in tech since it became dominated by erroneously self-confident, high-testosterone personalities who think that hiring people like themselves is a meritocracy?

Without getting into too many specifics (on rare occasions I pull it out again) it was a question of statistics and law of large numbers intuition. I was checking for how directionally correct people would be on the subject, with 3 parts of the question to see how deep they knew it. More about intuition than actual computation. There was a subtle error on how aggressively to apply the law of large #s.

How many times had I asked it? 50? I had a couple folks answer it like I thought it should be answered. He was the only who nailed it enough to catch the error too.

The reality is interview performance is very different than work performance. Interview questions test interviewing ability more than anything else.

Over time I've realized that the only reliable predictor of success is seeing people's work results first hand when they don't think they are interviewing. The danger is this too breeds issues of familiarity and lack of diversity, because one only hires from their narrow circle of the world.

This makes it very tough to make college hires. The only things I've seen that worked are "High enough" GPA (with "High enough" being higher on non-technical majors) and some sign that the student had to significantly multi-task (lots of extra curriculars, research, or working through school). Most 21 year olds don't have interview training, so it's a very bad way to pick them.

So, if the guy "got close", but then proved that you were wrong, was his answer was actually right, but that you thought it was wrong?

Yes. Much to my chagrin. But I was glad we hired him because he was smart and thorough, and only rubbed my nose in it once or twice.

I was in the position of that candidate before, and the interviewer refused to (do the analog of) letting me set up the spreadsheet, then vetoed me from the rest of the interviews.

What was the puzzle?

Also interested in finding out what the puzzle was.

Indeed, a "silly puzzler" is "any question I couldn't solve when I tried".

> Yes, some qualified candidates are bad at this. I'm sorry to hear that, but it can be practiced. Being good at interviewing can be learned.

Do you interview people to discover if they are good on the job, or just to discover if they are good at interviewing?

What a great solution. "Just don't feel anxious."

Obviously if someone is lacking in confidence, is anxious or is a humble person, they should just learn not to be that person any more so they can get a Google job like you, the paragon of success.

Otherwise, they DESERVE to be judged as unqualified. That certainly follows.

Ha! I am hardly the paragon of success. You judge my words too harshly. I am simply pointing out that, yes, companies conduct interviews by asking coding questions. Yes, some of the questions are hard and don't have obvious optimal solutions. Don't panic just because you don't get 100%.

I've never been asked such a question, provided a sub-optimal answer, and been allowed to continue. Every single time, the next question then is 'Can you make it <faster|better|etc.>?' Maybe, but by this point, I already know this is a terrible place to work.

I've been in this exact same situation. And I was offered positions. I guess that's my point - I didn't always find the perfect solution. I was pressed to continue to think about the problem until whatever allotted time was up. I wasn't [entirely] judged based on my final solution; I was judged on my ability to reason about the problem and demonstrate competency in my coding abilities.

Places that encourage me to find the best answer are good.

> Never think that you're blowing the interview.

How do you that? Because that's all I ever think about once I mess up during interviews, and it's all downhill from there.

>> once I mess up during interviews, and it's all downhill from there

I have witnessed this. I was asked to write code for LRU cache [1]. I started out wrong, thinking that LRU caches "Least Recently Used" items (in other words, most recently used need to be discarded). It turns out that LRU cache "discards" LRU items. I wonder if the name is actually a misnomer.

I caught this extremely quickly though on the very first hint and then tried to confirm the definition multiple times, but the interviewer had mental blocks by that time, nodding his head without even listening to what I was asking.

[1] http://en.wikipedia.org/wiki/LRU_cache

If you mess up, say so: "Oh, that's completely wrong. Huh. Ok, let's fix it." Again, these interviews are more about seeing how you reason about a problem. People make mistakes all the time - wrong assumptions, incomplete solutions, etc. What matters is what you do once you realize there's a problem.

I don't have any trouble admitting I made a mistake when I notice it, it's recovering from it. At that point I think about how I messed up, what I should have done, and how bad the rest of the interview will go.

In every day life, this is not the thought process I go through if I make a mistake. But in an interview the stakes are higher, you'll be quickly discarded if you don't do well.

In theory, yes.

In practice, you may lose even if the interviewer is wrong. I have. At some big-name companies.

Just for example, an interviewer at Apple was adamant that frequency can never be negative. Whatever happens to Fourier transform that integrates frequency from negative infinity to positive infinity.

I have more examples. :-)

This may very well be, but there's not much you can do about it at that point but continue to try to fix the problem. If you can't be amicable about it with the interviewer, then either they don't want to work you or you don't want to work with them (probably both). Might as well make the best of it.

Yeah, I think I once lost out on a Google interview in part because the interviewer lacked the math to follow my reasoning (also, they were attempting to get me to a different solution by applying criteria that my existing solution already adhered to better than what they wanted).

It's a bad question if you expect a quick, explicit answer. It's somewhat less bad if you're looking to have a conversation about how to solve problems the candidate hasn't seen before.

I've used an arguably "puzzle" question that was presented to me in an interview with some success -- the missionaries and cannibals problem. There are three missionaries and three cannibals that you must get across a river. The boat holds only you and one other person. The cannibals on either bank must never outnumber the missionaries on that bank. It's a good question to see if a candidate understands how to apply some basic computer science concepts. It can easily lead to a 30 minute discussion at the white board, with no need to get working code.

My favorite question is completely different. "What do you like best about your favorite programming language and what would you change about it if you could?" Anyone who doesn't see any flaws in their favorite language, particularly if it's Java, gets politely shown the door.

The problem with using this as a meta-question, is that someone who knows the answer may treat it as such.

That is, the person who comes to realize the canonical way of solving it may truly be brilliant and having been hinted and helped along the way came up with the solution, or he might have already known it and just played a metagame, assuming that's what you wanted.

Basically, you have four possible outcomes:

People who have never heard of it, don't come up with the canonical solution.

People who have heard of it and just blurt out the answer.

People who have never heard of it, but who come to realize it over the course of the interview.

People who have heard of it and who play it as a game, working towards it and then suddenly having an epiphany.

The first is interesting. The second you hopefully dispense with the question as being useful. The third would be a great hire, but is indistinguishable from the fourth, who you've learned nothing about except they know how to play your interview process. Which may or may not be the criteria you're looking for.

I was asked a puzzle question that I hadn't heard before. I answered it directly almost immediately. In a later interview with the CEO he asked if I had heard the problem before because the interviewers were sure I had with such a fast answer. I said I had not and just left it at that. Playing the whole thing back in my head, I think this cost me the job for "lying" to them. I might have been able to convince them on how quickly I got the problem if I had taken this as a questioning of my integrity.

"There are 128 chairs on a ski lift. You get on at the bottom, and ride it all the way to the top. On your trip up, how many do you pass on their way down?"

There may have been some distracting info of distances and speeds in there to make get you going in the wrong direction.

My experience with puzzlers is that average people like to feel superior for knowing the answers to a puzzle that someone else doesn't know. The people who can solve these easily are thrilled to find a puzzle they don't know the answer to. Then there is the group who can make a good puzzle.

Something like happened to me once, someone asked me a riddle (it was: "you have a polynom of coefficients that are positive integers. I can give you the values of this polynom at integer points of your choice. Can you get all the coefficients by asking only two values?"). Somehow I got it very quickly, and was thinking in my head "oh man I didn't know I was that clever", but the guy thought I already knew the answer.

I have said 72 assuming constant speed and equidistant spacing of the chairs- though as that seems to easy

Wouldn't the answer be 127 of them? You're passing the one behind you as you get on at the bottom, and the one in front of you as you get on at the top.

I well my view if you walk to the base of the lift half on the chairs are on the up side leg half on the down you get on the one right at the bottom. so you pass all the chairs on the downward leg

Actually this is a simplification as the chairs slow down and bunch up at the bottom and top to make getting on and of simpler

> Actually this is a simplification as the chairs slow down and bunch up at the bottom and top to make getting on and of simpler

Yeah, this is the problem with the question that I noticed as well.

If you don't know what a ski lift is (grew up in the south and didn't have money for vacations as a kid), you might not even know how ski lifts work. In which case maybe you guess that the chairs are fixed and give the 127 answer. Ironically, you guessed wrong and got the correct answer.

If you do know how ski lifts work and assume the interview does as well, then you just look like an idiot.

I guess the lesson is to state all your assumptions (and also, don't ask these questions).

I think I just had an "A-Ha" moment. It took me about 5 seconds to think it through. Good question.

Disagree: it's a terrible question. The answer (all of them) is something you either see or don't, and neither tells you much more than one bit of information about the candidate.

I'm not convinced; it at least tells you if the interviewee will rush to an answer that appears correct and say "All of them", or if they'll consider it a little more carefully and say "The rest of them". Or if they'll ask for a strict definition of what it means to pass a chair on the way up, to clarify whether or not you count the chair that you're riding on as one that you pass.

I don't think he meant in terms of an interview question, just a good puzzle.

Someone having a strong tool preference and thinking their choice of tool is good enough that it has no important downsides for their usage (and being aware of how to steer around pitfalls) is not automatically an incompetent programmer. Maybe it tells you that they chose their tool thoughtfully, it is a good tool for the usage, and they learned the tool well, and are enthusiastic rather than negative about it.

Most people who think they could improve a language don't have any background in programming languages, don't have a good nose for what would actually be an improvement, and would actually screw it up badly. Having declined to shave that yak for years hardly means that someone is incompetent. Are you interviewing people to create new general-purpose programming languages? Do you often build new programming languages as part of projects? If so, you must miss deadlines a lot.

It's incredible to me how interviewing is in the stone age: some guy gets assigned to interview people because it needs to be done, and how hard could it be anyway? Then he Googles questions to ask, which are more or less arbitrary, and grades them more or less arbitrarily. And the same guy who asks is the one who judges the quality of the question, which is tied up with ego and confirmation bias. Over time, he adjusts to his own taste, but is accountable to nothing else. You get a process which, at best, selects people who are like the interviewer, because the interviewer assumes he is good, so other good people have backgrounds and opinions (and demographics) like his.

It's the same process that 10-year-old boys at a skate park might use to judge whether a new boy is a "poseur." It depends as much on whether the judges think the new boy is like them or dresses correctly or whatever as on any meaningful evidence of ability.

> My favorite question is completely different. "What do you like best about your favorite programming language and what would you change about it if you could?" Anyone who doesn't see any flaws in their favorite language, particularly if it's Java, gets politely shown the door.

What about the poor people who've found a language that they actually like, and understand the tradeoffs that were made? I suspect I recently failed an interview for not having a good answer to that question, but really, what am I supposed to say? If I thought there were significant problems with my favourite language, I'd find a better one.

I once voted -1 on someone for not having a good answer to this one a language that he had used for years. I regretted the vote later, as he turned out to be excellent.

Having said that, it does surprise me that people have a hard time with this. Certainly I hope that you like your chosen language and understand the tradeoffs being made, but that is the point of the question really. Surely you can reason about these tradeoffs and have a discussion in this area? If your language is Java and you talk about how you are ok with their decision on erasure because of the legacy concerns, that's a good answer. I may (or may not) agree, but it's still a perfectly good answer. :)

OTOH, it is strange to me that someone would have worked in a language for 10+ years and literally can't come up with anything that they dislike (even understanding the reasons for those limitations). No matter how good something is, it is hard to imagine working with anything (or anyone) for ten years without coming up with a few pain points.

> If your language is Java and you talk about how you are ok with their decision on erasure because of the legacy concerns, that's a good answer. I may (or may not) agree, but it's still a perfectly good answer.

I wouldn't feel comfortable answering "what would you change about X" with something I actually wouldn't change about X (though maybe I should).

(FWIW my language was Scala, and while I've only been using it for ~5 years there's very little I would change. The answer I gave was to use HLists rather than Tuples in the standard library, but that feels like a nitpick more than anything else)

I don't ask for significant problems, just anything about the language that they would change if they could. A common answer when discussing Java is verbosity, for example.

Generics and type erasure.

When I interview for data scientist positions I ask your favorite question, but with "programming language" replaced by "statistical analysis package". (There's overlap between the two sets; their intersection includes Python, with the SciPy stack being what it is now.) I honestly don't care what package they name; it's more a way to get them talking about how they use whatever package they respond with, which reveals quite a bit.

Or, at least, I think it does. I'd love to have data on which interview questions actually elicit signal-containing answers...

You must be stating the missionaries and cannibals problem incompletely. If only one person can fit on the boat at a time (which is essentially the case because you didn't say the person operating the boat is a missionary or cannibal, and you are implying that they must be operating the boat), it's provably impossible to solve. Wikipedia says you may have 1 or 2 people on the boat (as opposed to 0 or 1), which makes it relatively easy to solve as soon as you realize you can take people back that have already crossed.

"The boat holds only you and one other person." That's 1 or 2, not 0 or 1.

The correct solution requires that you get out of the boat and put two cannibals on it. The statement that the boat holds "only you" and one other person very strongly implies that is not allowed.

Ahh, fair enough. I blame coffee for not working instantly. :-)

I really like the questions about programming languages. What's their favorite, just favorite language, features, etc. If the candidate says they've only ever used one, or they don't have any features they dis/like, that's a flag. If they point out a feature, you can use it to dive into a deeper discussion about something they say they know about.

+1 to this. The objective is not to see the person solve the question immediately, but to observe their problem solving process.

Wikipedia [1] describes the cannibals and missionaries problem differently than you do. Particularly, there is no "you" who transports them, there are only 3 Cs and 3 Ms.


I don't really have a favourite. I just go with the one that fits in with the team and if we need to choose one, try and find one that is a good match for the problem domain.

Will that do ?

but now anyone who's watched Fargo (tv) will be able to work it out

I appreciate all the technical discussion of this question in the article and here in the comments, but I think the question is just a symptom.

The root issue here is the other interviewer and the apparent interview process.

>'In the post-interview pow-wow, all of the engineers who’d interviewed him gave him the thumbs up, except my interview partner except my interview partner, who refused to hire him...'

Yes, yes, optimizing for false negatives over false positives is all the rage, but this just sounds broken - wasting the efforts of a full panel only to accept an arbitrary single-question veto.

>'..on the grounds that he completely flubbed this question, and "any engineer worth his salt should be able to answer it."'

Good engineers don't necessarily make good interviewers. Outright excluding someone based on any single question is just horribly self-satisfied behavior.

If you look at the structure of their pow-wow rather than its labeled purpose, you see that its real purpose is not to find a competent engineer but just to build a consensus. E.g. so nobody complains they wouldn't have hired the candidate.

It's the same principle which puts multiple men on a firing squad.

Good way to hire a camel when you need a mustang - your optimising for mediocrity.

Everyone does know the joke about a camel being a horse designed by a committee?

And should not have the chair/leader of the committee over ruled the outlier anyway.

You should absolutely get a camel over a mustang.

Even though they look like they're designed by committee, camels are an example of form following function. They're very well suited for their environment, and although they're stubborn, make good work animals.

A mustang on the other hand, is a feral horse. They're harder to domesticate because they've known freedom, and they've adapted to living in the wild by being smaller and more timid than horses selectively bred for doing work. If you really do need a horse, there are few cases where you need that horse to be a mustang.

If someone is hiring you because they think you're a mustang, you better hope they're caught up in the romanticism of cowboy stories instead of them being desperate for cheap labor or meat.

WARNING contains Sarcasam WARNING*

Of course that why history is full literally full of elite heavy camel units who turned the tide in so many battles and why Alexander's companions rode camels and of course the charge of the heavy camel brigade in the Crimea.

Yes assuming the is the USA I am surprised that HR do not insist that all interviewers have done some basic training just to cover the company's ass legaly

Certainly at British Telecom all people who sat on interview boards had to have done a two or three day residential course.

> With all of those hundreds and hundreds of minds, this problem remained open for 12 years.

A very good point. Some people just forget that while solution can look simple when it's already known, actually finding it can be extremely hard and more akin to invention breakthrough. So such question is more similar to asking to prove some difficult theorem really, than to anything else.

This article raises, sort of incidentally, one of the issues I've noticed recently teaching intro CS. Every CS2 class teaches linked lists, right? (Except for the ones that hit them in CS1, of course.) But: when was the last time you had to roll your own linked list?

Once, not that long ago, it was important to cover linked lists (and other data structures) because you'd have to be writing them in order to implement a program of any significant size. Now, though, all the common ones are already implemented for you (and to be honest, we don't even use LinkedList all that often---array-based lists are better in most real use cases).

So now their role in (my) CS2 is chiefly as setup. I still need to cover linked lists because they establish concepts that will be important when we get to trees (links) and to more general graphs (cycles and the lack thereof), as well as in operating systems (continuation inodes, memory allocation) and other areas.

But implementing an honest-to-god plain-old linked list holding ints or strings or whatever? Who does that anymore?

This is a question I see asked all the time and it's absolutely mind-boggling to me that you could ask it. Who do you think writes the linked-list libraries you use? Do you think the libraries just magically appear when someone wishes them up? Do you think we should use the current set of libraries forever, calling them "good enough", and limiting ourselves to current programming languages for all of history? That's what will happen if we no longer teach people to write these things; or else (more realistically) we leave it up to them to make the same mistakes and waste time solving the same problems which were solved decades ago, because no one ever bothered to teach them the answers. Give it a few years and we'll have no one left who can program in anything but javascript; no one left to write the browsers! --- or so I worry when I read HN sometimes.

You'll notice I'm not suggesting we stop teaching them. But what you describe are specialised applications; my point is that where linked lists were once something that every programmer, large and small, could reliably expect to have to write repeatedly, they're now something that are only directly relevant to, say, library maintainers.

I think an unreasonable question (ie, of a practical situation in which no one will ever find themselves) deserves an equally unreasonable answer:

Write a program that starts at the head of the list and in a loop follows the "next" pointer on the current list entry, halting when the pointer is null. Run the program and wait. If the program never ends, there's a cycle in the list.

Then just come up with an algorithm that proves if the program will halt or not. Easy peasy.

I have had this problem in real life. I had a graph database, and I needed to find the end of a path in the graph. If the the path looped back on itself, my program would not terminate.

This occurred to me as well: the graph data model is one of the few areas in which you might practically need to address this problem. Again, however, a lot of these issues specific to the graph data model are open research questions and not really appropriate to an engineering interview.

Finding the loop in a graph is a much more valid scenario and any candidate aware of traversal algorithms will have a much better chance of dealing with it than finding loop in a LL.

What if you gave the answer immediately after asking the question and then judged the candidate by how much their face lit up?

But why? One puzzling thing about our industry is our obsession with finding suitable proxy indicators for programming ability.

But we can test for programming ability directly.

Oh, I agree. I just thought it was interesting to think that a bad question might be improved by giving its answer away.

Disagree that this problem never arises if you're not implementing a linked list data structure.

Here's a file a customer sent you that describes their organizational hierarchy:

  ID    Name            Manager
  1     Alice           4
  2     Bob             1
  3     Charlie         8
  4     Dave            -
  5     Elise           3
  6     Fred            2
  7     Gemma           1
  8     Henry           5
Is this a valid hierarchical tree? Okay, it's not exactly the same problem, but it's certainly similar...

Technically in the context of an org hierarchy that would be detecting cycles in a directed graph, not a linked list cycle. I don't think the algorithm to detect cycles in a linked list applies to general graph problems, and the problem is much harder for graph cycles.

It's not a generic directed graph either, though - each node only has one outbound link. The sequence of parents for any given node is a linked list, and a cycle in that list is invalid. You certainly don't need to run tortoise and hare for every node path looking for cycles. There are other approaches, clearly - reachability analysis starting from each root node, for example.

But if I were to ask you to write a routine to find the root node for a given node based on that data, then ask you to guard that routine against finding yourself in an infinite loop or stack overflow, tortoise and hare would be a legitimate answer.

Just trying to make it clear that just because the question asks about linked lists, doesn't mean it's always about abstract data structures.

The article seems to imply that the turtle and hare algorithm is only valid solution for this problem, which is certainly not the case.

This problem can be also solved by building big set of already seen list nodes or adding bit to list nodes themselves. Whis is both something that everybody should be able to come up with and implement.

I think that this is in fact quite good interview question because while there is one non-obvious perfect answer, it also has some sub-optimal solutions that are pretty obvious and simple to implement.

You're right, though I think he's talking about the weird type of interviewer who expects candidates to know obscure, non-intuitive optimal solutions to these problems.

I once met an interviewer who liked to use the "reverse a string" question, except he was expecting candidates to know the XOR trick. Suffice to say few, if anyone, ever got past his "filter".

Primadonna interviewers who use interviews as ego-strokes are a problem in our industry. It's where the stupid "why are manhole covers round" questions came from, and it's where these hyper-optimized algorithms questions come from.

Yeah - the article built it up and built it up, but when the question finally came up my thought was "wait, that's it?!"

If the interviewer asked the question and rejected anyone that didn't immediately answer with the tortoise/hare approach that's the fault of the interviewer, not the question.

Agree. The anti-pattern here is expecting a specific answer for a question. The goal should be, "how does this candidate think through problems?" If they arrive at a logic solution and can critique it, then I would consider the question completed.

Some questions are better than others here, specifically ones that have many correct implementations and varying levels of optimal-ness. I'm a big fan of asking for Fibonacci implementations, for instance.

Since this question can have different correct implementations, then I would not consider it a bad question to ask.

It's upon the interviewer to be open to different answers, evaluate them fairly and ask smart follow-up questions. "Ok, so what's the complexity of your answer? Do you think it's possible to do better?" or "Compare your answer with this implementation".

This was my first thought about how to solve it. It provably works in finite time. This ought to be a satisfactory answer in an interview, particularly if the candidate points out that it's unlikely to be The Optimal Solution. After which point, the appropriate follow-up question from the interviewers would be to ask what the time complexity of the algorithm would be in the worst case.

'Between 1955 and 1967, the problem of “how do we de­ter­mine if there is a cycle in a linked list without mod­i­fy­ing the list” was an open problem. Meaning, any number of PhD can­di­dates in Math­e­mat­ics or Com­puter Science could have written about it as part of their dis­ser­ta­tion. With all of those hun­dreds and hun­dreds of minds, this problem re­mained open for 12 years.'

Were there any computer science graduate programs between 1955 and 1967?

You're not dealing with "hundreds and hundreds of minds", you're dealing with those minds who 1) heard about linked lists, 2) heard about or realized there was a cycle detection problem, and 3) devoted any energy to that problem as opposed to the myriad other problems available to a mathematics or electrical engineering graduate student. All of this has to have happened between 1955 and when the problem was solved in advance of starting to write the paper on it, in advance of publication of the paper in 1967. The paper also presumably included a proof of correctness that isn't being asked for in the interview.

The interviewee has the further advantage that they know there is an answer in the space of the constraints the interviewer has drawn. As they suggest less optimal solutions, those constraints get tighter, and the answer gets easier to find. If they have been through an undergraduate computer science course, they probably also have more familiarity with linked lists than most of the people attempting to solve the problem before 1967.

None of which is to say that failing to find the answer deserves disqualification - that's probably still unreasonable - but the picture is nowhere near as absurd as presented.

I disagree with the author that this is a "puzzle question" that requires an "a-ha" moment. Quite the opposite, it is an easily solvable algorithmic question (maybe with a little guidance after the candidate suggest suboptimal solution). Also, the good thing from the candidate point of view is that it is the sort of questions you can prepare for.

>sort of questions you can prepare for.

The point is why do you need to prepare for and regurgitate these silly answers that you will almost assuredly never ever ever use in your day to day job?

+1 to that. Most programming jobs are about integrating a bunch of different technologies together to solve a business problem. In 20 years I've never written a linked-list or sorting algorithm for work yet interviews are increasingly narrowly scoped to algorithm questions. I've found that these questions are usually asked by young people because they are fresh out of school and have been led to believe that knowledge of algorithms is the most important consideration. Give them a decade of creating products based on css, html, node.js, nosql, yessql, etc and they will come to realize that programming is more about understanding what a customer needs and finding the most efficient way of delivering a solution. Ability to work with other, lead a team, write clearly, explain to customers the tradeoffs and costs of different solutions are actually way more important than knowing how to find a cycle in a linked list.

But how do I test all of that in 30min or less? Do I have to actually get to know the person, their work history and their strengths? Does that mean I have to read the resume before I enter the room?

Well one can prepare by memorizing a ton of algorithms and hope to get lucky and get asked the questions about those they happened to study or by actually studying up on what's relevant to the job, which random algorithms that can be looked up on Google most certainly is not. It's impossible to do both--it's difficult enough concentrating on one area.

I know which engineer I'd hire. Then again, I wouldn't ask such useless questions in the first place.

Did you miss the part where it took 12 years to come up with this solution?

Has anyone actually answered this question on the spot without help and without knowing it? Probably not without help, but then how do we judge the amount help given to an interviewer?

I have been going through my own crises of interview question. My go to question for interviews which I have asked of about half a dozen candidates so far, wasn't completely flubbed only once. The only people I know who gave the good answers besides myself were my boss and our intern from MIT during mock interviews. I have been grading the candidates not on being able to write the solution in code like I intended, but on being able to come up independently with bits needed for the solutions. This has been quite unsatisfactory. How do I know if my question is too hard, or if something else is wrong, like prescreening of candidates?

Yes, I did figure it out. Actually, the answer I came up with was slightly different: it was pretty clear that I had to move the first pointer to get inside the cycle. Once I was inside, I thought the second pointer would need to move a lot to find the cycle, rather than just going two hops. So I proposed to advance the first pointer by n, and the second one by exp(n). Maybe it would go faster, who knows.

By the way, this was the question at my first interview ever, so I do remember it quite well.

PS: I interviewed many candidates since, and I would never ask such a question.

Can you tell us what that question is? Not details, but the general idea.

Me, I'd ask them this:

1. Write a function that takes a string of hex digits and converts them to a long (or equivalent in the language). Keep code minimal.

2. The opposite. Keep code minimal.

3. Write a function that reverses a UTF-8 encoded string (provide a good reference to the UTF-8 format), keeping combining characters in the reversed string in the same order they were in the original. I.e. aüç should be çüa, not ¸c¨ua. Assume the string of bytes is a proper utf-8 string, that only combining characters have to be order-preserved and that the only combining characters are the ones with unicode character codes in these ranges: (give them some standard ranges). Keep code minimal.

Anybody who can do these three things should be able to learn most any other tech stack you throw at them. If on the other hand you're looking for people with specific expertise with a given branch of computer science, then you need a different set of tests. The central question is - do you need a developer or a mathematician specialising in discrete mathematics. Those are two different sets of people, though they tend to overlap.

I have. I had help, but the process was just like any other problem. State constraints. Look for a violation of constraints. Sort of like proof by contradiction.

I can't remember the actual hints. I blitzed through 3 or 4 naive solutions first. My favorite was using tagged pointers to mark visited nodes. Eventually through a combination of hints and exploration I got to the idea of using two cursors. At that point it was a simple constraint about which cursor would be in front of the other and if there was a violation of that constraint it could only be because of a cycle.

It made for a good interview, but there was good chemistry between myself and my interviewer. Any question would have worked in reality.

I instantly knew, with no pre-knowledge, of two of the sub-optimal solutions presented in this thread that require O(n) storage.

The optimal one, no but I didn't try any further than the solutions I had.

Yes, I had it during 1st phone screen with Microsoft, coding it in their interviewing system and I figured out the two pointer version without a prior knowledge.

> In this case, the generally “correct” answer is a tortoise-and-hare algorithm, where you have two pointers at the head of your linked list, one traverses the list two nodes at a time, and one traverses the list one node at a time; if ever the pointers point to the same node, you have a cycle somewhere.

I'm an extreme novice to programming. I don't fully understand this.

1) How can you traverse a linked list two nodes at a time? Is this just short-hand for "go to the next node, then go to the node after that, and call that one operation"?

2) If you can detect when two pointers are pointing to the same address, couldn't you solve the problem by traversing the list one node at a time, storing the addresses of each node in an array, and checking to see if the array contains any duplicate values at intervals?

1) Yes, that's exactly it. node := node.next.next

2) Yes, you could, and that is a valid way to discover a loop. However, consider that your linked list is already a store of the contents of the list. Rather than copying values out, you may as well use the list itself.

If you could assume that the loop contains all members of the list, then you could detect the loop by iterating through the list and checking to see if the node is the same as the head of the list. Unfortunately, the loop doesn't always point back to the head of the list, so you may have something like this:

  1 -> 2 -> 3
       |    |
       ^    \/
       5 <- 4
By maintaining two node references, and advancing one of them twice as quickly as the other, you are effectively doing the exact same check, but eliminating the problem that the loop may not include the head. In both tests, the separation between the two referenced nodes increases by one every time, but only in the second test is the "head" reference guaranteed to eventually be in the loop.

With regards to (2), this is one way of solving the problem. However, your solution uses O(n) memory (where n is the size of the linked list), and the tortoise-and-hare algorithm uses only O(1) memory. Here's how it works: http://en.wikipedia.org/wiki/Cycle_detection#Tortoise_and_ha...

unless the interviewer plans to move the goal posts, there was no specified memory restriction in the original question.

Complexity optimization is a generally useful skill.

"moving the goal posts" is called "every month at work, when you finish 1 task and then move on to another task, instead of going home and getting paid to rest on your laurels"

moving goal posts, is "I asked you for x, you provided me with x, but I really wanted y, so now I'm going to tell you you're wrong"

You want someone to answer a programming question within a particular set of boundaries, you set those boundaries.

I don't think it has to be "wrong". If I were asking this question, and I was provided with "store them in a hash", I would say something along the lines of "that's a correct solution" and then add additional constraints.

"How you adapt in the face of additional constraints" is very much relevant to programming.

The "tortoise" pointer does a standard traversal - one at a time. The "hare" pointer goes from one node to the next _and then the next_ - i.e. two at a time. If the hare hits a null node, there's no loop. If the hare hits a cycle, it will eventually run back into the tortoise and there's a loop.

I must have learned this algorithm in school, but completely forgot it. My naive implementation was 2), but this is the first explanation on this page that made Floyd’s cycle-finding algorithm click for me:) Now I can't unknow it - thanks for the clarity and concision!

1) yes. 2) is ok (with a hash table instead of an array to have linear execution time), but doesn't run in constant space unlike 1), or you could mark the nodes when you visit them.

1) One loop, two pointers. Each iteration increments each pointer by a different margin.

2) Yes you can do this but not in all languages.

Had that exact question given to me before, but it was in the form of a closed book "pre-screening" written test.

I simply put down my pen, walked out of the room and told the recruiter if this company was hiring based on these type of questions I would be a terrible fit :)

They were likely hiring to fill a PHP dev roll writing CRUD web forms too, right?

I've been places with tough interviews and incompetent developers and the opposite as well. One does not necessarily correlate with the other.

I would love to find the definitive resource for preparing for interviews like this. Besides answers like "just be awesome", is there such a book or page?

In the past I've searched for various compendiums of algorithm interview questions and saw the cycle detection one come up. I knew I would never be able to solve such a problem on the spot. This blog post initially put it in perspective for me, that its pretty much a BS academic question that you have to already know the answer to - until I read the comments here, full of totally awesome alpha SV nerds who are saying "I instantly got the optimal answer without any foreknowledge". Now I don't know what to think.

After a 10 minute phone screener I now rely on a paid pair-programming session to identify good candidates. This is what you're going to hire them to do so why waste your time asking questions that at best form a poor approximation of their skills?

If I do not know any better, the first thought that would come to my mind would be: "Wow, this company has really messed-up code base, and people here cannot even get linked-list done right. Do I really want to work here?"

I had this in an interview (back '96). Came up with a linear time constant space solution: create a new node, push it on the front, reverse the list, check if the 1st node is the same one you pushed, if so it has a loop, otherwise it doesn't. The downside is you need to reverse again to get the original list.

Another answer is to mark nodes using the low bits of the next pointer. Also linear time, but also requiring clean up. I can see how not knowing c would cause people to think this question was somewhat unfair, but I didn't see it that way.

How would you reverse a linked list with a loop?

If your list is laid-out a contiguously in memory as structs with both next and prev ptrs, you can visit each member using ptr arith and switch the values without actually walking the list using next and prev. Ditto if there's an underlying data-structure involved such as an array.

How could you have a linked list with a loop in contiguous memory? Or in an array?

You allocate an array of objects that contain pointers, then link them together. It does not have all of the typical benefits of a linked list (in particular, there's a hard max size and may require a fixed size, depending on just what you're doing to it here) but it does allow O(1) reordering elements for a list-order traversal, while also allowing a linear traversal through memory.

If the linked list has a loop, but you don't know it, how do you figure out the length of the list to allocate enough space for your array? How do you copy the list into the array without traversing the list? How do you traverse the list without hitting the loop and looping infinitely?

If you know that all the elements in the loop are already in the array then traversing it is easy. Of course, if you have an upper bound for the length of the list, there's an easier check...

I don't think interview must entirely be related to the practical world. I mean having a very hard and interesting practical problem is the best of both world but not necessarily the only way to go. If you want to know if the candidate can really build things you can check out his github/resume. I think the 45-min interview is just to determine how smart the candidate is. If you want to determine if someone is smart you put them in a very hard situation (not necessarily a practical one) and see how they reason about it.

Tom Forsyth listed one of the reasons--when a storage is explicitly expressed. The other two that I can think of are for external linkage and non-constant expressions on the rhs. In the end, you really don't know without some introspection. It's up to the compiler. External linkage is one that will force storage, but in general const will require internal linkage and fold expressions accordingly (at least for primitives).

Edit: I would never expect anyone to actually remember or need to know that for an interview. It would be definitely something you would re-discover/discover because you had to run into it.

I completely disagree that this question is irrelevant. I can imagine this exact situation in compiler design where you have dependent types that can't have infinite loops (in pseudo-code where forward declarations are not needed or implied):

type a = b

type b = a

If a program has a structure like this with cyclic dependencies, then the compiler must be aware of this and break. (At least, I can imagine a situation where this could be a desired attribute.)

You are mixing up cyclical linked lists with loops in a graph. The reason why the question is bad stems from the fact that it's specifically about linked lists.

I could definitely be mistaken, but I don't see a difference between a linked list and a graph. They are two ways to represent the same thing. That being said, if he asked me to find cycles in a linked list, I might determine the adjacency matrix and check for loops:


This is at least very easy to reason about. However, if you are still right my thought process is just very wrong.

"I could definitely be mistaken, but I don't see a difference between a linked list and a graph."

A linked list is a particular type of DAG (or at least DG). The relevant distinction between linked list and the general case is that a linked list restricts nodes to (zero or) one outbound edge. This is relevant to the problem at hand, because you only have one tortoise and one hare and so they can only follow one path.

Right. As a graph theorist might put it, a linked list is a directed graph with a vertex-disjoint path cover consisting of exactly one path. I.e., linked lists are specialized graphs. The tortoise-and-hare algorithm can be generalized to detect cycles in any connected graph, as long as a consistent traversal order is used.

"The tortoise-and-hare algorithm can be generalized to detect cycles in any connected graph, as long as a consistent traversal order is used."

But not in O(1) space, right?

I think it depends on how the graph is represented. With the "standard" representations, it would be O(n) space in the worst case, where n is the number of nodes in the graph. You'd still only need the 2 pointers, but the traversal itself would need to also remember which paths it has not yet visited. The worst case is when the number of paths is much larger than the length of any particular path. The best case (apart from the null graph) is equivalent to the linked list, where there is only one path in the cover.

I can imagine representing the graph as a list of paths, in which case the tortoise-and-hare algorithm is O(1) space, though the graph itself would be (potentially much) larger. Virtually every other operation on such a graph would take a performance hit, too.

There may be other factors I haven't thought of, but you're right that the generalized algorithm could not be as space-efficient as the special algorithm.

It's worth noting, though, that the original question poised in the article did not place an O(1) restriction on space.

"It's worth noting, though, that the original question poised in the article did not place an O(1) restriction on space."

For sure.

Ok, that was poorly worded. What I mean is that a linked list can be thought of as a restrictive graph. I'm just saying I'd prefer to think of it as a graph. I could still be wrong. Is that kosher?

That's certainly kosher (though, nitpicking terminology, I'd say "restricted" rather than "restrictive").

That doesn't get around the original objection, though, which is that the restriction has to be exploited to get the algorithm in question, and thus the algorithm cannot be applied to the general case of graphs.

Having said that, certainly the ability to make use of restrictions provided by your domain is hugely valuable, and this should serve as a relatively familiar case.

A graph can have branches, but a linked list is linear.

The page is down, so here is the cached version: http://webcache.googleusercontent.com/search?q=cache:rwL5dDE...

A not quite serious solution [1]:

   Let m = size_of_current_process / sizeof(list_node) + 3
   Let np = pointer or reference to first node in list
   while (np != null)
       if (m-- == 0)
          return HAS_CYCLE
       np = np->next
   return NO_CYCLE

[1] this approach might actually be acceptable in some circumstances

I don't want to complain but one of the major culprits who conducts these kinds of interviews is Amazon. After going through their interview process, I'm of the opinion that they basically want you to memorize careercup, geeksforgeeks and leedcode and just spit it out in chunks at appropriate times during the interview.

"...'Write a func­tion that can detect a cycle in a linked list'...You can’t reason your way through [that question]...all you’re doing is testing if someone has seen your par­tic­u­lar puzzler before or not."

I think I'll take a pass on interview advice from this character.

Once in an interview I was asked to implement a synchronization mechanism between processes using shared memory. It turns out this is the petersons algorithm. I mean if I could solve this, I would be Peterson :) heck, even smarter because I would do it in 20 mins.

I'm no fan of this particular interview question, but the article takes it too far. You can figure out tortoise-and-hare algorithm by reasoning (I'm pretty sure I did the first time I've seen the problem - although not in an interview setting). The key is to recognize that there's a very limited number of things you can do given all the constraints: moving pointers is pretty much the only tool you have.

If it indeed took 12 years for the algorithm to appear, it's likely because finding loops in a linked list isn't a very common problem, not because the algorithm itself is that hard.

Nor is the 12 years thing particularly good reasoning... By that thought process, it took humans hundreds/thousands/millions of years to develop a compiler, so how can I be expected to develop one over the course of 4 years in college? Or even over the course of my life?

Not exactly a valid analogy, because his start date isn't "the beginning of human history", rather it's the development of the linked list data structure, which is the first point at which humanity possessed sufficient knowledge and tools to figure out the cycle-detection algorithm.

Given that we didn't have the knowledge or tools necessary to make a compiler hundreds/thousands/millions of years ago, you'd have to bump up your start date considerably.

Sure, I stretched that too far. But hopefully the point is made. Then should we expect no students (or anyone) to be able to build a compiler (or os kernel) in any amount of time given the internet and whatnot? While I may be articulating this poorly, I think my fundamental issue with his point is solid.

Because over the course of 4 years in college you're provided with resources that explicitly teach you how to develop a compiler, a scenario which is entirely different from what happens in an interview.

I am the guy who asks that question on every interview. You can hate on me now! Let me clarify. I don't expect the candidate to get to the "a-ha" moment at all. I make my expectations very clear upfront so that the other side wouldn't panic. However, I expect to see how the candidate would approach the problem with the right attitude so I could get some insight of his/hers thinking patterns. I personally wouldn’t mind if an interviewer asked me something in the lines of "how would you build a nuclear plant?", although I have no knowledge on the matter.

Uh we did this in our first year CS curriculum...you are telling me your interview candidates can't even do what a first year CS freshmen can?

Are you referring to tortoise-hare algorithm? I refuse to believe CS freshmen regularly come up with this on their own. Maybe a few percent, everyone else is googling or relying on huge hints from the lecturer.

If you were referring to a more trivial algorithm (e.g. linear space), I think you misinterpreted what the interviewer was expecting and would not have gotten the job.

It's also worth noting -- esp if you're referring to the TH algorithm -- that CS curricula are not standardized and that there are probably dozens of "clever trick" algorithms you haven't seen and couldn't come up with on the spot (e.g. can you initialize an array in constant time?), but someone else thinks is trivial because they happened to see it in a lecture/read it online.

My solution would have been: Simply have a wrapper around the linked list that keeps count of the elements (+1 for insert, -1 for delete, initialized with 0). Traverse the list. If your traversal takes more steps than there are elements, you can stop it and conclude there's a cycle. I wonder how the interviewer would value this answer...

If I was interviewing you, I would point out that you're redefined the question: the question gives you a standard linked list, and you state that you want a wrapper-api around it. I would then ask you to try to find a solution that works for a bare bones linked list.

I would make a note that you did find one type of solution, of course.

edit: I would probably also ask why you created an api that even allows cycles, being as that you are defining control over adding and removing elements. Further, if you stated that your api explicitly allows cycles, I would ask why you don't simply mark them at insertion time and flag them, making the "detection" algorithm O(1).

The naive way to do it is just create a hashset that stores pointer to next, walk the linked list, and if you get a collision you have a cycle. The problem is that it requires O(n) memory, where the tortoise and hare method only needs O(1).

As others noted you can't know how many elements there are supposed to be in the list. But you can guesstimate.

1. Take a guess about how much memory the program has allocated so far. If you can't, go with 2^48 = 281474976710656 bytes - the maximum you can do on a 64bit pc today.

2. Divide by the minimum size of a list entry.

3. Traverse the list until the end or until you traverse more elements than there ever could be.

In general though, as noted, you'll know if you're making a list that could possibly contain cycles due to exposing an interface that allows the creation of cycles. In that case, given that you want to be able to efficiently detect cycles, you'd make a special linked list implementation that does keep internal count of its size.

That's a great idea. I had another approach that assumed the entry addresses were aligned to 32/64 bit words in memory as structure's by default are. Then the next pointers will never have the ones or twos bit place set and can be used as a visited marker[1]. This gives a O(n) with a much smaller constant factor than the hair and tortoise.


If you argued that many programming languages already provide list.size() or equivalent (more so if I told you "in a language of your choice" before), I'd swallow my pride and accept it.

Then I'd suggest you try the other way, but it's a perfectly good answer.


The problem is, in Java and probably many others, that method computes the size in linear time by traversing the list. The javadocs for java.util.LinkedList note that this was a decision taken so that splitting a list at an arbitrary node, etc. could be a constant-time operation.

What all algorithm and idiotic interview questions (like moving Mt. Fiji) are great for is the interviewee himself, not the interviewer. They are a huge red-light indicator that there is something wrong with the engineering culture and department at the company you're interviewing. While there are exceptions, in my experience, there is such a direct correlation between the two, that questions like this have led me to walking out of interviews (politely of course). Why continue if you know you won't take that horrible position? Obviously, if you're hired to write algorithms, then it might be a different story, but most programmers are simply not writing new algorithms. Ever.

Isn't writing algorithms almost the definition of programming?

I think "coming up with interesting algorithms" was the intended meaning.

In which case, no, everyday programming is mostly not about designing algorithms. Otherwise, programmers would spend most of their time with journals and books, with a few days of programming each month.

Interesting is fairly subjective but ultimately you are designing processes to solve problems. If you cannot demonstrate ability to solve a simple, well trodden problem how will you be able to solve a new one?

> Interesting is fairly subjective

Interesting is a social construction. As such, I would define interesting as worthy of publication in a first or second-tier conference at some point in the history of the field.

> If you cannot demonstrate ability to solve a simple, well trodden problem how will you be able to solve a new one?

0. I've seen many well-respected researchers who have solved truly interesting, new-to-humanity problems stumble on well-known problems others consider easy and/or fundamental. In fact, this is probably true for every single researcher. And doubly so for practitioners. One person's fundamental technique is another's blind spot.

1. Simple != easy to come up with. Often exactly the opposite! Take the TH algorithm as an example. It's a very simple algorithm. Small. Simple. Optimal. Awesome. Not trivial to discover, however.

2. Well-trodden -> unless the algorithm is absolutely fundamental and taught everywhere (e.g. sorting), you've given up on testing problem solving skills. You're just testing whether the person's intro/DS/algo professor covered this particular algorithm (and maybe their memory). Or, more likely, rather their random interview studying happened to hit your pet problem. Since curriculum is not anywhere close to uniform in CS -- even among top schools -- this is just silly.

3. Simple problems don't always have (simple) solutions. Create a fair voting scheme.

It depends on why you are asking the question. Are you literally demanding the candidate spit out the optimal answer immediately? If they can they it means they can remember their algorithms class which is a good signal.

If not, can they solve it with a naive method and then make reasonable attempts to iterate on that, even if they never hit an optimal solution?

It is a little like a high end restaurant asking a trainee chef to make an omelette, which is apparently common in the culinary industry.

> Are you literally demanding the candidate spit out the optimal answer immediately?

I think the author of the article is indicting situations where this is the case. Perhaps he could have been more explicit up front, but reading between the lines this is the situation he's complaining about.

> If not, can they solve it with a naive method and then make reasonable attempts to iterate on that, even if they never hit an optimal solution?

If I were asking the question, absolutely.

> It is a little like a high end restaurant asking a trainee chef to make an omelette, which is apparently common in the culinary industry.

Are there omlettes that are "optimal" (very awesome in some sense)? I honestly don't know (and if so, man, I want one!)

Can you detail a little the kinds of things you associate with stupid questions that make you want to walk out?

Come on, this is like the easiest problem you could have gotten during Google's, Microsoft's or Facebook's phone screening interview. If you can't figure it out in 10 minutes, then you should forget about working there.

I miss the old times when these puzzlers were asked during interviews. My theory is that you can recognize an excellent place of work with a great future when they actually ask you these sorts of questions as it attracts certain kind of excellent and creative people. I could see the "interestingness" of companies being shifted as they abandoned these questions, first Microsoft (picked up by Google), then Google (picked up by Facebook), like they were passing some kind of "greatness" token (MS->G->FB).

I was told at Google that I solved two problems like this during an interview that nobody solved before me in 15 minutes (one led to a super complicated math formula and dynamic programming, the other I solved optimally but differently to their solution, and provided an optimality proof based on discrete calculus). It was the greatest and most challenging interview in my life. I miss it, all interviews nowadays are just boring as hell :-/

What did you exactly prove during this interview? That you know to write readable code? That you are know to code in team? Did you write any test case during this inteview? Any docs?

I proved that my solution had optimal time complexity using discrete calculus, for a highly non-trivial problem that requires to come up with unexpected ideas under huge stress.

They saw my own open-space real-time software 3D rendering engine with texturing and shading (C++/Asm) at SourceForge already, they saw my own SQL server (many students at my first university were ripping it off for their own class projects in the next years so it was a mistake to make it open) as well as other projects I did prior to that interview so they were familiar with my coding style and quality.

Also, before that while working at SUN I did some work on a test suite with 10,000s of tests, many of them for SUN's own compliance with one of their standards, making sure it was executing on a distributed cluster every day for every branch they had in order to ship the best quality to customers. Many of these tests were for obscure cluster failure cases using custom bleeding-edge test frameworks. So the answer is no, I didn't write any test at Google interview, it wasn't required.

How can you prove you can code in a team during an one-on-one interview? At best you can give hints that you are willing to discuss stuff and change your progress on input from the interviewer and explore if there is any sort of chemistry between you two, which is about all you can do.

I really hate this kind of nitpicking, averaging everyone to a "normal" level, trying to find something wrong in someone because she or he goes against (IMO) lazy typical stereotype. Enjoy your mediocrity! Are there still hackers on HN?

I hope your answers are meant to be a mockery of such interviewers and their class pets.


Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact