[shaftway]

I'm in the interviewer pool @ Google.

> practice writing syntactically correct code on a whiteboard

This probably differs from interviewer to interviewer as to how strictly it's adhered to, but it's not really a hard and fast rule. I'm sure there are some interviewers that will ding you on a forgotten semicolon, but I suspect that most would not.

Personally I look for code that isn't so far from syntactically correct that it's clear you are trying to BS me. I'll even accept pseudocode for the most part. But I've had candidates that try to make up language features, and that doesn't fly with me.

> practice solving problems with a 30 minute timer

I only give my candidates 30 minutes. The whole interview is 45, I spend 5 minutes introducing myself and setting up expectations, 30 on the question, and 10 on answering their questions (after all, they're also interviewing us).

You can tell pretty early whether they're on a solid trajectory, and I'll offer the occasional hint to keep someone on track, or ask tangential questions if they're doing well on time. Not finishing isn't a deal killer, provided you had a solid approach and weren't just running in circles. But a good candidate will finish in about 25 minutes and we can spend some time talking about alternate approaches. Sometimes I'll show them the optional approach and see how that conversation goes.

Nine times out of ten a candidate scores low because they overlooked an infinite loop or code would crash on boundary conditions and candidate wasn't able to realize that even with hints.

[ummonk]

>Nine times out of ten a candidate scores low because they overlooked an infinite loop or code would crash on boundary conditions and candidate wasn't able to realize that even with hints.

I guess this depends on how you run the interview but one of the things that frustrates me about whiteboard interviews that focus on the coding rather than the design is that I have to step through test cases (especially edge cases) like this manually, which is tedious and not at all how you do things in real life, where you just run the test cases and see what happens instead of having to run your own code on a whiteboard.

[shaftway]

My question is about walking around a data structure. Most candidates choose to represent it with arrays.

I agree with stepping through code being lame. If a candidate has tried to do some sort of boundary condition check and mentions a test that'd catch it, then I'll give them a pass. If a candidate just blows past the code without any attempt whatsoever to check that they're in bounds on an array, then I'll ding them for that. You have no idea how often I see code like...

    Node neighbor = data[x+1][y];

... without any check at all to see if x+1 is in bounds for data.

The infinite loop in my question is fairly obvious (because it involves walking around a data structure iteratively). Candidates either see it right away and handle it as they solve, notice it halfway through and crowbar it awkwardly in, or don't notice it until asked to walk through a specific test case, or don't even notice it while walking through the test case.

Ultimately, if the code looks correct to me, or even close (I mess up occasionally) then I'll just ask what cases they would test.

[blumomo]

There are however cases where

    Node neighbor = data[x+1][y];

does not require an index boundary check when you know that 'x + 1 < length(data)' for any 'x'. This could be due to the iteration of 'data' excluding the last element or any other condition that's always hold.

[seventhtiger]

I'm like you in that I used to code "against the IDE and test cases" by basically rapidly iterating and kind of refining as I go along. Always fast to compile and see what happens.

Learning to write code on paper for exams and on white boards for interviews actually improved my skills a lot. Maybe not everyone coded like I did, "against constraints", but having to slow down and take into account the constraints mentally led to making better code.

[CodeMage]

> But I've had candidates that try to make up language features, and that doesn't fly with me.

This piqued my curiosity. Can you give an example?

[justrudd]

I'm still floored by how many candidates I've talked to that assert with great confidence that the local variables they declare will still be there with the same values when they call the function recursively.

And most recently, when iterating over a string's characters the underlying string methods KNOW that the string is being iterated and will pick up at the current iteration point. For example, you've got the string "5432112345". And via iteration, you're currently pointing at the first "2" at index 3. The candidate asserted that if you call "indexOf('2')", it would return 6 because "indexOf" knows that it is iterating and should start 1 beyond where it is pointing at.

And my favorite - once had a candidate that misspelled a function name while coding in Ruby. I didn't ding them for it. But I pointed it out because it just bugged me. And the candidate then swore to me that Ruby will automatically call the correct method if there was only 1 candidate based on the misspelling. You know when you do things like "git inti" and it says "did you mean init?". The candidate swore that Ruby would just call "init" for you because it knew what you wanted.

[perturbation]

https://github.com/yuki24/did_you_mean#installation :

    Ruby 2.3 and later ships with this gem and it will automatically be required when a Ruby process starts up. No special setup is required.

It doesn't call the method for you, but it does do the did-you-mean automatically if you misspell and it's close enough.

[justrudd]

Maybe this what they meant? I dunno. I can see this being helpful in irb/pry.

In the end, we did end up making an offer to this candidate. They did well on everything we asked them. This was the only "brain fart" they had.

[fjp]

I once passed a whiteboard interview just calling random made-up operations on generic java arrays, like Array.flatten().

I was relatively new to programming and had been practicing in Java but didn't know how to execute a lot of map/filter/reduce operations off the top of my head like that. I did, however, know that my interviewer was a Python programmer that probably didn't know much about Java language features.

[seanmcdirmid]

I thought you were allowed to assume functions you needed in the interest of a modular solution. Array.flatten is an obvious “assume I have this, I would write it anyways.”

I think most decent interviews will give you a pass on an enhanced standard library.

[Zelizz]

When doing an interview at Google in C, I asked if I could assume I had a hashtable implementation with so-and-so interface, and the interviewer said no ¯\_(ツ)_/¯

[yoz-y]

This depends on the question but I assume that in this case the interviewer wanted a different solution than one using a hash table.

[seanmcdirmid]

Ouch. Well, most hashtables in C are custom, I guess.

[sanderjd]

If you can explain what it is supposed to do, I don't care a whit if you make up a method that probably exists in a standard library somewhere (Guava / Apache Commons, or wherever). If it's slightly more obscure, you might have to write pseudocode to show you know how it might be implemented.

[shaftway]

This one candidate had been allowed to skip the phone screen, and phone screens would likely have filtered them out. Candidate used Java and claimed to be deeply familiar with it. Obviously this isn't exact, but the result looked something like this:

    public isTheFooBarred(class Node { int[][] positions, int size } node, int x, int y, Set visited = new HashSet(node.size)) {
      // ...
    }

The first parameter to the method had an inlined class definition, and the last parameter was optional. We discussed this and the candidate claimed that this was legal vanilla Java. In the case of the last parameter, the candidate claimed that the compiler generated every possible combination of methods including and excluding each of the optional parameters (so, 2^n methods).

I really tried to give the candidate the benefit of the doubt, and asked if there was some sort of annotation processor or code pre-processor that they were using, but they were adamant that it was plain old vanilla Java. I'm pretty sure that they were using some in-house built thing, but their solution had significant problems, so it wasn't the only reason I was against the hire.

[emmelaich]

I've had people interview for a Java job who didn't realise strings were immutable.

The question asked was a very short multiple choice; only two of the answers were possible under immutability.

[sanderjd]

That doesn't sound like a good question at. It sounds like you're looking for "Java programmers" instead of solid engineers.

When I interviewed for my current job using Java, I had been programming in Ruby and JS for the better part of a decade and had to refresh my Java syntax fairly quickly. I know I made some dumb syntax mistakes in my phone interview, like instantiating collections totally incorrectly. I distinctly remember one of my in-person interviewers saying "well, in Java it's boolean, not bool, but sure...". More semantically, I may very well have messed up mutability of collections and primitive conversions and boxing and such.

Enough of my interviewers saw through all of this that I got the job. Now people on my team come to me and say, "hey you're a java guy right?" and ask me questions about this stuff. I wasn't a java expert when I interviewed, but now I am, because that's what my job required, so I learned it. That's what my interviewers were looking for, to the company's benefit.

[emmelaich]

You're right that it is not disqualifying even if the Java expertise is essential for the job and the candidate describes themselves as a Java expert. This was a single short question among many though. And it's pretty fundamental to Java methinks.

BTW, they still got hired and proceeded to be a solid productivity reducer and all round waste of time.

[vips7L]

> Nine times out of ten a candidate scores low because they overlooked an infinite loop or code would crash on boundary conditions

Why does this even matter? They're writing the code on a whiteboard, without being able to compile or debug, in 20 minutes. Does Google really expect code to be correct and production ready in 20 minutes?

[robocat]

> ... and candidate wasn't able to realize that even with hints.

He said why. Why did you misquote?

[vips7L]

That is why they were "dinged".

Not why those things matter in whiteboard coding.

[shaftway]

The standard solution to my problem is on the order of 12 lines of code. I don't expect it to be perfect or "production ready", but the infinite loop should stick out like a sore thumb. I do expect candidates to demonstrate that they recognized that problem, and to at least make some attempt to check that they didn't overrun the bounds of an array in the 30 minutes they have.