Reversing the technical interview (2017) | Hacker News Reader

Reversing the technical interview (2017)

aphyr.com

158 pointsbryanrasmussen35 comments

There is another one that is new and not linked into the series.

https://aphyr.com/posts/353-rewriting-the-technical-intervie...

Ahh. Hexing the Technical Interview is still my favourite, but that was brilliant.

I absolutely loved the introduction on this one, and the realization that parenthesis were simply too far too the right in C-likes was mind blowing for me.

The more I program professionally, the less patience I have for clever code in a work setting. I used to love it, but after the nth time it caused me real, physical pain while oncall, I'm no longer a fan.

Clever code doesn't help me in a hundred million++ line repo of code maintained by thousands of people. It is actively detrimental.

Still fun to pick apart for leisure.

Or to write about.

cleverness is closely related to wit, and brevity is the soul of wit. A sufficient amount of cleverness often allows you to say in a few lines what would otherwise be said in 100s of lines - for example switch statements or numerous ifs going through each possibility with not much difference in how each branching is handled. I find the verbose listing of each possibility gives me physical pain, so I support cleverness in such situations.

Until the spec changes.

The way I get around this is having a dsl no one but me sees that compiles to $lang of choice that everyone does.

Once it produced 20k lines of code in an afternoon. I didn't have to do any work for the next month and still was the top performer by commits.

That definition of "top performer" bears scrutiny. Code is nearly always a liability, not an asset. 20k LoC, generated by a secret tool, sounds more like grounds for firing than promotion.

Or were you making a joke?

Have you worked at big brain dead corp inc?

Yep. But I never used being surrounded by nonsense as an excuse not to try to make things better.

I was getting paid $1,800 per day for three months to do whatever they told me. Does it pay more to do things your way?

If I was motivated by money I’d be on an entirely different career path.

That's nice for you. What would that path be?

Investment quant, usually. If a person wants to truly maximize dollar per hour.

Yes. In more ways than one, too.

100x this. "Clever" code that takes mere mortals 10x as much time to wrap their heads around than just doing the boring straight-down simple thing are one of the tell-tale signs of smart but inexperienced devs. Hopefully they get the right feedback and use their smartness to write well-structured understandable and maintainable code when they are more senior. Failing that thay tend to grow up into the bitter grumpy dude who just things everybody is stupid.

+1 This is why Haskell & co will never be a mainstream language.

This "clever code" is a bit misnomer for this kind of phenomena isn't it? Even overly-complex code is not an accurate term for that because overly-complex code can be caused by an overly-complex valid spec.

"Stupidly out-of-spec code" or "Stupidly prematurely abstracted code" are more fitting to describe those kind of code.

I personally feel uncomfortable calling them clever code because they are not clever at all so that they are causing physical pain.

I'm fortunate to have a complete "creative-control" over a code several times. I frequently tell people to stop writing code that we don't need for the moment. Most of them are anticipation of reuses.

The next post is even more mindbending: https://aphyr.com/posts/341-hexing-the-technical-interview

My favorite line so far (still reading it):

    “If it were meant to be illegal,” you remind him sagely, “Sun Microsystems would have made it unrepresentable.”

The way I think of that trick is it's simply translating struct.field to struct(field).

You can also use it to represent sum types and then pattern match on them. For instance, Python doesn't have a case statement. We could define an Either type as:

    def Left(value):
        return lambda Left, Right: Left(value)
    def Right(value):
        return lambda Left, Right: Right(value)

    Left("failed, so sad")(Left=complain, Right=fh.write)
    Right(b"great data")(Left=complain, Right=fh.write)

You can combine the techniques to get a list type without relying on nil:

    def Cons(head, tail):
        cons = lambda field: {"head": head, "tail": tail}[field]
        return lambda Cons, Empty: Cons(cons)
    def Empty():
        return lambda Cons, Empty: Empty()

    my_list = Cons('a', Cons('b', Cons('c', Empty())))

    def foldr(func, zero, list):
        return list(
            Empty=lambda: zero,
            Cons=lambda cons: func(cons('head'), foldr(func, zero, cons('tail')))
        )

Ive worked for many different companies for the last 20 years. I started as a Jr front end dev / designer, transitioned to a backend team, was doing "fullstack" before it was a real term, eventually was considered a sr engineer, then lead and now I run the tech for an early stage start up (vp of eng, but my title is irrelevant because Im actually also filling the roles of CTO and lead eng).

I despise whiteboard interviews and the hiring process in general and vowed to myself that I would do my own thing if ever given the chance.

I am fully in charge of the hiring process here. What I look for are good personalities, a passion for technology, and the ability to read code in the stack we use. I show the candidate a script that we run in prod which has all comments removed and ask them to walk me through it. Massive bonus points if the candidate is active on the internet (streams code sessions on twitch, contributes to open source projects, has side projects etc).

No brain teasers, traditional whiteboard challeneges, live coding sessions or take home projects.

Did you by any chance post on the wrong thread?

There’s three great reads like this. previous: https://news.ycombinator.com/item?id=13886626

And I really wish I were good enough at, like, lambda calculus and category theory and whatever other black magic is in there to understand them!

It's...not exactly lambda calculus or category theory. It's a dramatic flattening of the layers we think of as natural in our programming world. Of course we write source code describing a class that is compiled into an object file which is loaded into a JVM. So just write the bytes that makes the JVM have a class directly into its memory. And on the other side, that anything obeying the same mathematical axioms can be treated as the same, be damned how it's implemented. Those two taken together can make it seem like witchcraft until you realize that all the conventions of our craft are just that: conventions.

Not for this one.

Absolutely beautiful writing!!

The interviewer asked the right question at the end.

Sorry

edit: Interesting post from a humor point of view (not my cup of tea TBH).

Hmm... Methinks your missing sense of humor would disqualify you from the same 99%.

Oh come on

"So I was a holding an interview and my interviewee was a witch who used Clojure to write a low level implementation of a list rather than using the one in the stdlib. And at the end of the interview she just flew off out of the window on her broomstick! Rude! Apart from anything else, when she was concentrating hard, black snow started appearing in the room, I just don't think that would work in the office. Hard no from me!"

Probably, but the 1% of places would be amazing and worth missing out on the other 99%.

Just in case you didn't know, Aphyr is responsible for https://aphyr.com/tags/jepsen.

You know, the testing tool that managed to show that pretty much every distributed system out there didn't live up to its claims about how it tolerated a variety of failure conditions.

If you're not able to appreciate him, that's your problem. Not his.

Reply on news.ycombinator.com