The skills just aren’t there. I run interviews and struggle to find anyone, anywhere, that can actually program. English proficiency is also hard to come by. This wouldn’t be an issue in most of Canada, but I’ve yet to see a Canadian remote applicant.
Remote may bring down wages for workers located in HCOL, but it’s been great for me as I don’t have the ability to relocate.
And yet... I run in to many who still get hired. I've got some colleagues working at shops where people struggling with the concept of a front-end UI that calls a paginated API endpoint. Get page 1, then get page 2 when you need it, etc. They keep filing 'bugs' and someone just bumps up the 'max per page' from 50->80->100->120, etc. The team lead stopped allowing those PRs to go through, and others complained loudly enough that the PR blocking is overridden. Because... "pagination is hard". But these people are still employed. The thought is the team lead is being 'unreasonable' by 'demanding' pagination.
I regularly hear horror stories like this, and wonder "how do people like this get hired?" and also "why can so many of my skilled colleagues not get hired?" The gatekeeping and inability to evaluate skills, and perhaps a reticence to hire above your own skill level... all of these seem to be creating more stressors on an already bad job market.
It is not just about the absolute scarcity of resource, it is about the ratio between bad and good. Needle in a haystack problem.
Indeed it is and always will be. But the issue is that companies don't need good devs. They need "make the thing work" devs. And if they can make the thing work for long enough to generate revenue, it's all that matters. Driving customer value down while at the same time depressing wages, and telling both parties to just deal with it. It's become the standard operating model for all large companies now.
Almost no one passes.
Would you like to give me a programming interview?
It have made hireing so convuluted since it boils down to tabs versus spaces.
I'd rather argue that the methodology to hire is flawed generally and dice rolls are probably better then most processes. Especially the riddle and leetcode ones.
It's not "tabs vs spaces"; there are just many people employed as software engineers who 1) cannot translate an English (or other human language) description into working code, and 2) have no idea how software actually works. They know enough about the mechanics of programming to copy and paste code from Stackoverflow/ChatGPT and use git, but if they encounter a novel problem they are totally lost. Hiring them is like hiring someone who only knows how to drive to be your car mechanic.
If you've been in the industry for 5+ years, think hard and I'm sure you can at least think of one example of a coworker you've had that fits this description.
If it is this one, encountering at least one example of a coworker over 5+ years is pretty far away from the numbers implied. “ 199 out of 200 applicants for every programming job can't write code at all.” I am very skeptical of this.
Keep in mind your coworkers are the ones who passed the interview. How much hiring have you done? 199/200 of applicants not being able to code is probably an underestimate these days; now that it's so easy to apply for jobs with one click on LinkedIn, Indeed, etc. any job you post will get flooded with resumes from people who are completely unqualified and have literally never held a programming job or have any computer-related education. Can't blame people for optimism, I suppose?
If we jump to the next level of the funnel, to people who have passed an initial resume screen and are doing a phone or in-person interview, I estimate at least 60% of those candidates are still unqualified to hold a position as an actual software developer solving actual technical problems with code, in my personal experience.
What are you evaluating them on? Leetcode?
He comes dressed for it and with his own toolkit.
You seem to be testing for proficiency in slow NIH-tendencies basically, but since discomfort is a requirement, maybe that works for you and for them too.
But it's also then possible to make use of "one step up" by asking people for variations and seeing whether they can talk about tradeoffs (still at a very basic level, but still).
E.g. in a higher level language, if asked to reverse the order of the words you often get variations of split on space, reverse the resulting array, join. In which case you have an opening to ask them if they can do it lower level, in minimal space.
In which case, the "typical" solution becomes to reverse the string in place, then iterate and find the start of each word, and reverse that in place. Both reverse operations need at most two pointers and a temporary, and can be done with the same function if you take in a start and end pointer.
If you have that helper to reverse a given character range in place, reversing the order of the characters within each word without reversing the word order is just the same thing without reversing the whole string first.
One google interviewers ask me a question that starts with “implement google search” and ends with my implementing splitting a paragraph. Of course I failed the interview, but I have to think he has failed me way before we get to write any code.
I'd say about half of the candidates that interview are unfamiliar with the String#substring equivalent of their language. This is for a senior position, and everyone has multiple years of experience on their resume. It's... interesting.
Many people would say those candidates should be failed anyway due to X, Y, Z reasons. But I want to highlight that in those cases, the interviewers should realize they are assessing something else, and not programming skill.
I estimate that 80% of candidates could not accomplish this in an hour. 50% of them could not even get out of the gate.
array data = (array(int))random_string(100)/10;
int x = random(10); int y = random(10);
int seek = random(255);
int findgrid(array data, int x, int y, int seek){
for(int i=0;i<=2;i++){
for(int j=0;j<=2;j++){
write("%d:%d\n", data[x+i][y+j], data[x+i][y+j]==seek);
if(data[x+i][y+j]==seek){ return true; }
}
}
}
80%? hard to believe it's that bad.
grid = [
[1, 2, 3, 4, 5, 6, 7, 8, 9, 0],
[1, 2, 3, 4, 5, 6, 7, 8, 9, 0],
[1, 2, 3, 4, 5, 6, 7, 8, 9, 0],
[1, 2, 3, 4, 5, 6, 7, 8, 9, 0],
[1, 2, 3, 4, 5, 6, 7, 8, 9, 0],
[1, 2, 3, 4, 5, 6, 7, 8, 9, 0],
[1, 2, 3, 4, 5, 6, 7, 8, 9, 0],
[1, 2, 3, 4, 5, 6, 7, 8, 9, 0],
[1, 2, 3, 4, 5, 6, 7, 8, 9, 0],
[1, 2, 3, 4, 5, 6, 7, 8, 9, 0]
]
foo = function(x,y,num){
if (grid[x][y] == num) return true;
if (grid[x+1] && grid[x+1][y] == num) return true;
if (grid[x+2] && grid[x+2][y] == num) return true;
if (grid[y+1]){
if (grid[x][y + 1] == num) return true;
if (grid[x+1] && grid[x+1][y+1] == num) return true;
if (grid[x+2] && grid[x+2][y+1] == num) return true;
}
if (grid[y+2]){
if (grid[x][y+2] == num) return true;
if (grid[x+1] && grid[x+1][y+2] == num) return true;
if (grid[x+2] && grid[x+2][y+2] == num) return true;
}
return false;
}
alert(foo(1,2,3));
alert(foo(1,2,30));
https://jsfiddle.net/ycx159kg/
13 min later...
round2 = function(x,y,num){
return grid.slice(x,x+3).map(g=>g.slice(y,y+3)).flat().includes(num);
}I'm intrigued by the slice/map solution. It would have been perfectly acceptable if you could explain what the functions did. In fact, it would have led to very probing conversations: my usual interview partner and me had a deep interest in elegant solutions.
grid.slice(x,x+3)
If the second value in slice is larger than the length it slices to the end. It could be a 2x10 or 1x10 but lets pretend it is 3.
.map(g=>g.slice(y,y+3))
.flat()
.includes(num)
Also you seem to be bounds checking for 0s, but to do that in the first dimension you need null pointers to pad your row selection and a second set of 0s at the end of your rows.
It's interesting that both you and the person you replied to used arrays of arrays which is the very first mistake everyone makes in graphics programming when trying to make a 2D array or and image.
A much better way is to have a single contiguous array and use buf[y*width + x] to lookup into the one dimension buffer.
No, not at all. There are cache misses and simplicity, but think about when an 1k image is on the heap. You have to allocate a buffer of pointers for each row then you have do a thousand individual heap allocations to get all your rows.
Now think about an image with rgba channels. Technically this is a 3d image with one short dimension. Are you going to allocate a buffer of a thousand pointers, then off of each one allocate a buffer of a thousand more pointers so that you can ultimately do another 1 million heap allocations, all 16 bytes for each rgba pixel?
There is no upside and all downsides, which is why it is never ultimately done this way, though it is an almost universal mistake when people do multi-dimensional arrays for the first time.
Correct, it should be [y][x] indeed. I would have spotted that had I bothered to.
> Also you seem to be bounds checking for 0s, but to do that in the first dimension you need null pointers to pad your row selection and a second set of 0s at the end of your rows.
Not sure which is the first dimension but for an array of 10 if(arr[100]) returns false. Anything behind the && is ignored.
Looking at this again the really funny part is that it doesn't work if num = 0 since if(arr[100]) returns false if there is a zero at position 100.
> It's interesting that both you and the person you replied to used arrays of arrays which is the very first mistake everyone makes...
I copy that part from their solution but probably would have myself. I've been found out not being a graphics programmer. The experiment was a success.
Of course I've made all these mistakes to test if anyone was paying attention. (I always say that)
This is taking the first pointer and offsetting it by the length of 100 pointers. There are only 10 pointers in the array so it is going into other data.
If your pointers are 64 bit and your numbers are 32 bit, even if that is referencing the memory mapped data of the sub arrays it is going twice as far as the end of the array and returning 0 purely by chance of some other memory you're dealing with by accident.
A better way would be to keep things simple and bounds check that x is less than width and y is less than height while using the single buffer method.
The point here isn't to be too harsh with criticism, it's just that it's easy to see something as more trivial than it really is.
function foo(x,y,num){
for(i=y; i<y+3 && i<10; i++){
for(w=x; w<x+3 && w<10; w++){
if(grid[i*10+w] == num) return true
}
}
return false;
}
but that is an additional condition that was not in the original task. there was no mention of image programming.
since we are discussing interview examples here, leaving out such a detail and then calling people out on an error caused only by missing that would be considered deeply unfair towards the candidate.
my point is, that our choice in the solution is not interesting at all because it was part of the task. you are proposing a slightly different task, which is of course interesting too.
I'm not proposing a different task, I'm pointing out that you both thought it was trivial and both ended up doing arrays of arrays then having something that wouldn't work because of the complexities of that. You're right that someone could solve the problem with arrays of arrays, but no one did.
I think the problem is mostly trivial too, but before doing graphics programming and solving almost identical problems multiple times, I would have made most of the same mistakes.
The point is that the vast majority of candidates, including some with Master's degrees in CS, could not write even a pseudocode solution to this problem. It got so bad that I started doing a verbal fizzbuzz test during the phone screen so we didn't waste so much time.
You might say it's pseudo code but one of the original comments was talking about the low numbers of people who could actually get something to compile and solve a problem.
you are right about x and y. i was really tired :o
but it runs (at least as long as there is no overflow). here is a sample output:
> findgrid(data, y, x, seek); // there, fixed the x/y problem ;-)
160:0
165:0
65:0
174:0
23:0
253:0
229:0
130:1
(1) Result: 1
As per the question I've ignored error conditions, but using the simple expedient in this case of positioning the 3x3's top left within the first 8x8 not the full 10x10.
Example output is underneath.
// Make and fill the grid.
var r = new Random();
var grid = new int[10, 10];
for (var y = 0; y < 10; y++)
for (var x = 0; x < 10; x++)
grid[x, y] = r.Next(10);
// Draw the grid for reference.
Console.WriteLine();
Console.WriteLine(" 0 1 2 3 4 5 6 7 8 9");
Console.WriteLine();
for (var y = 0; y < 10; y++)
{
Console.Write($"{y} ");
for (var x = 0; x < 10; x++)
Console.Write(grid[x, y] + " ");
Console.WriteLine();
}
Console.WriteLine();
// Random starting point and target.
var tlx = r.Next(8);
var tly = r.Next(8);
var target = r.Next(10);
// Extract the 3x3 sub-grid and "[ ]" any matches.
Console.WriteLine($"Looking for {target} in 3x3 grid at {tlx},{tly}");
Console.WriteLine();
for (var dy = 0; dy < 3; dy++)
{
for (var dx = 0; dx < 3; dx++)
{
var ax = tlx + dx;
var ay = tly + dy;
var v = grid[ax, ay];
if (v == target) Console.Write($"[{v}]");
else Console.Write($" {v} ");
}
Console.WriteLine();
}
Console.WriteLine();
0 1 2 3 4 5 6 7 8 9
0 1 2 5 4 1 8 0 2 8 5
1 4 2 2 5 8 4 4 0 4 8
2 7 1 8 2 5 6 3 6 3 6
3 8 7 3 2 4 6 5 9 6 2
4 6 3 4 2 9 7 4 7 6 1
5 1 4 2 3 8 3 2 4 3 3
6 0 6 6 5 1 4 7 3 4 9
7 2 0 6 9 9 9 3 9 1 7
8 1 6 5 8 3 8 9 8 3 4
9 0 8 7 9 5 4 0 7 0 3
Looking for 5 in 3x3 grid at 2,6
6 [5] 1
6 9 9
[5] 8 3
Similarly there's the possible use of a ternary, less memory allocation, and tighter code, but again in this situation I'd go for readability.
There may also be edge cases not covered; this was a 10 minute hack (as it looked fun) and isn't exhaustively tested.