Apart from the usual brain-teaser and whiteboard coding, we gave candidates 20 minutes to browse through a small database project in eclipse. Afterwards, I checked whether they could explain the data structure of that database.
The outcome of this test was a much better predictor for competence than any other technical test. I can whole-heartedly recommend it.
Sometimes you realise that was a mistake :)
int WINAPI WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance, LPSTR lpszCmdLine, int nCmdShow);
Note the hInstance and hPrevInstance bits. Back in the day Windows loaded the code for a Windows application's .EXE once into memory and then maintain several pointers to a separate data section for each running instance of the application. The hInstance and hPrevInstance are actually pointers to such datasections. One of the reasons for this was so that the program needed to only do certain kinds of initialization once; if future instances needed that data, they could copy it into themselves from the previous instance (remember, no protected memory!) with the GetInstanceData function. In addition, it was a simple way of telling if you were already running; window classes, for instance, were registered once per program, not once per instance.Of course in Win32, instances of a program are their own processes with their own address spaces, so you can't do the kinds of clever tricks you could in Win16 to save time and memory. And hPrevInstance is always NULL. Win16 programming is full of hacks like this that made sense in their day, but are counterproductive now, though their vestiges remain on in Win32.
On the other hand, if the job requires a lot of algos and low level code, this will be a good test to gauge an engineer's thought process and mental capability.
We tried sending the sample to them the day before, but there wasn't a strong correlation between doing that and good responses from the candidates. Basically, if they got it, they got it. And if they didn't, they didn't.
mysql_query('
INSERT INTO `products` (`name`, `descriptions`, `price`)
VALUES ( '.$_GET["name"].', '.$_GET["description"].', '.$_GET["price"].');
');
Then he would ask, "how do you feel about that?" He claimed this question weeded out a lot of potential candidates.I don't know if you can use this kind of test on fresh-out-of-college hires, because a lot of professors don't teach security, and there may be plenty of other bad habits to break anyway.
But on any experienced hire -- even someone one year out of school -- it's perfect for screening out the codemonkeys.
Every time I opened Management Studio for SQL Server 2005+, it took forever and caused massive disk churning, compared to the Win32 version for SQL Server 2000 and before, which didn't engage in abstract disk churn and memory gobbling.
I wonder if something went wrong with the native image generation during setup? It may be trying to JIT it every time on startup. That could be quite slow and result in a lot of disk activity.
I know, technically they can cheat and google API calls or show it to their friends, but I bet with a little prying in the interview you could uncover that.
Interestingly this is why I believe VCs let founders of start-ups get do partial cashing out in later rounds. In the later rounds when you are shooting to build a billion dollar business, it's not optimal for the company if you as CEO is thinking about your everyday finances or paying past debt.
The really interesting thing about the quote is that it is from 1986 but could have been stated the same way in 1996, 2006 and likely in 2016 with the same "10 year back" qualifier and nobody would bat an eyelash. Systems are way more powerful but we're also pushing a LOT more data (in terms of depth of the data, breadth of the data, and types of data).
To be fair though, these days a lot of people see "spin up 25 more AWS instances" as a perfectly valid way of solving a bottleneck when compared to months of developer time whereas back in 1986 the decision would be a no-brainer in the other direction given that the equivalent of spinning up 25 AWS instances today would be economically infeasible back then for almost any commercial entity (even if you scaled back the amount of processing needed to fit in the context of that era).
Which is the whole point if the quote. And the "back in my day" is 100% true in the field of programming. Ever heard of Moore's law? Ever tried to program with 8K RAM?
>We still run into performance issues with things like bulk updates and computation times (MapReduce / distributed computing) quite frequently.
Yes, we still run into "performance issues". But with MILLION times the data they ran into performance issues with.
Heck, we can now do a whole computer of the time interpreted in a web browser in Javascript and still get acceptable speed to play games in it.
Perspective man.
edit: but yes, it's amazing.
You've just fully booted linux in your browser. That Linux sees 16 MB RAM for itself. If you are a web programmer, do you even try to see how much memory the browser spent for this, or any other page you click to?
Do you really think developers these days are just as good at optimizing for limited system resources as they were in 1986?
"We still run into..." Who? Maybe you - but most developers aren't running into map/reduce problems every day. And when they do, they certainly aren't coding a solution from scratch.
Yes boxes are faster and easier to spin up. The 'downside' to that is that many website on the internet are blazingly fast: Google, Amazon, Dropbox, etc. Because they are so fast (or perceptibly fast via tricks), the rest of the web is held to that higher standard. Twenty years ago, things taking a long time on a computer screen with no updates was acceptable. Today it really isn't because there are N other websites that are selling your exact product or experience.
Have you ever programmed a embedded system? Every worked on a 8-bit micro controller?
There are hard, very hard problems that come up which have to solved in with really limited resources.
My point is that those problems aren't any harder. They're just different. (in many cases it's because the toolchains stink by modern standards--embedded systems tend to just crash with no helpful debugging information) BG and co stood on the shoulders of the guys who made their chips 'just work' just as we stand on their shoulders. To complain about how 'kids have it so easy these days' just makes you sound crotchety.
Not really. The problems are made a lot harder by finite memory being more in play. Even with a good tool-chain when you actually have more data then you have RAM then you'll run into problems that are completely unique to an embedded world.
Code shrinks are very hard.
And --
"I still think that one of the finest tests of programming ability is to hand the programmer about 30 pages of code and see how quickly he can read through and understand it."
Does any company run "code comprehension" tests for hiring? Any thoughts?