Are there any resources you recommend? Or do personal projects give you enough problems? And how do you map solving these problems to patterns you can use in a future problem you’ll face?
Practice. A lot. Learn the whole stack (down to the hardware, understand failures modes and limitations of each layer).
Solve many different problems, make experiments and throw away code. Make connections between sub disciplines.
You have to have the attitude that if somebody understood this, then I can too, if nobody did, there's no good reason why I can't be the first one.
Fail a lot and learn. Think about every failure you had, but as a problem to solve not a comment on your value as a person/engineer. What could you have done differently? What were the signs that got you in that situation?
For tricky problems, what works for me is the obsess-and-let-go strategy. Work intensely on some problem for some time, if you make no progress, just let go of it, forget about it, do something else and perhaps your brain will connect the dots. Talk about the problem with other people. Explaining it and different points of view usually change your perspective enough that you are no longer stuck.
Also note that reading about something is not the same than doing it. You need to read and attempt to replicate, even if a toy version of the thing to better understand. Some things just take time an effort to seep in.
1. Be thorough. Be sure you fully understand the issue, instead of jumping to conclusions.
2. Keep asking 'why'. Make sure you understand what the problem (or rather the requested feature) is, instead of being able to parrot what someone else says the problem is. (Even if the conclusion ends up being the same, you need the gained comprehension to determine the right solution.)
3. Separation of concerns. Keep practicing on separating orthogonal concerns that are all seemingly relevant to a problem. Evaluate each concern separately. Drop any of the concerns that are irrelevant in solving the solution. This can save you huge amounts of time, enough to make a deadline you would otherwise never make. (This will also help you to reduce the amount of added complexity.)
Things got a lot better for me when I realized almost nobody gets to the best solution on the first try.
Cobble together an ugly and hacky solution that you can confirm solves the problem, then iterate and iterate until you clean it up well enough to ship it.
> You can't reliably deduced anything about what your current problem is, if the things it depends on are flaky.
Interesting, as a junior dev I always start thinking about the fun architecture stuff first and end up being slow to complete a project. I figure that deploying one place vs another is so different you want to deploy early and often for testing. Maybe I should just start getting the basic functionality and tests down before packing it up into docker containers and running it on kubernetes and all that?
Each part you add must be rock solid before you move on to the next that depends on it.
That doesn't mean you have to do everything... merely make it solid and reliable.
Say you plan to depend on a subsystem that is going to do something wonderful and flashy.... but you don't need it right now, you just need to build the place where it's going slot in....
Then don't put your sketch of the wonderful shiny thing into your dependency tree, put a mock that might do nothing, but does nothing utterly reliably.
If you have something flaky firing bullshit events and rubbish into your whole system while you're trying to get the rest going... you're just going to hurt yourself.
You can add a feature toggle or chicken bit to switch the new shiny/flaky stuff in when you want to contemplate it, and switch it out when you're working on the rest of the stuff.
There are a ton of stuff to write about this but so little time so I'll try to just give one thing that helped a lot (small effort, large payoff)
Be explicit with what phase of your "thinking" you're in. When faced with a problem, focus on giving lateral/divergent ideas and come up with a multitude of different solutions. Don't spend time thinking why/why not a solution is good, just note the solution itself and move on to finding another one.
Once this is done, start the next phase which is the convergent thinking. Sit down with all the different solutions and start compare the tradeofs between them. Eventually you'll reach either some middle ground of some solutions, or just one solution.
Doing this when collaborating with other engineers is also super helpful, as people won't feel like their ideas are being shut down as soon as they utter them, so more people can part of the whole process (if needed)
No one will ever successfully transfer the problem in their head to your head first try. You will need to ask a lot of questions to try and bring your understanding near their understanding.
Equally you need to actually understand why the problem you're solving is a problem. If someone wants the header on their website to be bigger, that's a very easy thing to do but there's always a reason why they want it bigger and knowing that will help you do it properly.
Also, always ask why whatever solution you came up with works. There are very few situations where the actual answer is "Invoke these magic runes in the right order and pray".
Spending the time to find out what the magic runes are and what they do will help you make sure that
a) The problem won't come back
b) Your solution is actually a good one
c) You can solve similar problems in the future.
Empathy, taking with people, iterating often with the clients inputs.
Realizing that technology and architecture ultimately doesn't matter that much for 99% of the use cases and sticking to proven technology to deliver value (postgres, rabbitmq, ansible, redis and django)
When coming across a problem, I type out all hypotheses on a notepad.
Microphone doesn't work on app? Is it this device specifically? Is it the file system? Faulty recording code? Null pointer somewhere?
Then I find the simplest, hackiest way to (dis)prove the hypothesis. Be sure to narrow down the test as far as possible.
What a lot of people do is just bash random things, often the simplest hypothesis, repeat that same thing for half an hour, do something else, then repeat what they tried earlier. Or in somewhat tougher problems like pathfinding, where each hypothesis takes an hour to test, a lot of people get intimidated and procrastinate.
Age. Without question, age has improved my problem solving abilities. I no longer expect to come up with an instant solution; instead I get to know the problem as best I can, then go and do something else. My subconscious will eventually deliver the optimal solution.
Interestingly, "eventually" usually means in the shower the next morning. From that I assume that I'm more receptive to flashes of inspiration while I'm relatively relaxed in the shower, and also that the real work of finding a solution happens while I'm asleep.
Couldn't agree more to this. Even having a small tea break or watching a youtube video for distraction helps a lot. I think its something about perspectives, we cannot focus fully on a given task and also look at it from a wider perspective or different angles at the same time, so taking a break and coming back gets focus on these angles which were earlier ignored. Kind of like "boxed thinking" which self help articles always talk about.
Throwing away assumptions and trusting nothing, where it makes sense to do so. I've been burnt too many times by library bugs, framework issues and the other layers between me and what I'm actually doing. When I first started out I used to think that browsers such as Chrome were infallible and any problems had to be with my code.
As a result of being burnt, I'm happy to give up on logging and trawling through code in favour of just taking a PCAP and finding out what's actually going on over the wire. Or stracing my app written in a high-level language which runs in a VM. Sometimes you just want to see the syscalls.
I'm also happy to go digging in the browser's source code. I think my favourite bug to diagnose manifested as a visual issue with menus in a frontend framework. The menus were styled with some CSS, nested inside a media query:
@media (hover: hover) {
These styles were only supposed to apply on desktop devices with a mouse pointer capable of hovering over HTML elements.
The rules seemed to apply on some OnePlus devices though, with just a touch screen as an input device.
Getting to the bottom of this involved creating a test page to reproduce the problem, reproducing it in multiple browsers, digging into Firefox for Android's source code (yay FOSS) to find out how it implemented the media query, writing an Android app to reproduce the underlying data problem and eventually working out that it was a problem in the phone's operating system.
To be honest, for me the point where you discover "this media query does not behave correctly on this device that I need to support" is where you stop digging and find a different solution for detecting pointing devices.
It's very clever that you managed to discover a bug in OnePlus's version of Android but then what?
> To be honest, for me the point where you discover "this media query does not behave correctly on this device that I need to support" is where you stop digging and find a different solution for detecting pointing devices.
It enabled me to go back to the customer and explain the problem and make the informed decision to not continue work on the problem. The fact is that the broken OnePlus devices (not all of them, and not all the time) represent a very small proportion of the devices we need to support.
I guess maybe this is a bad example because the web is stuffed full of "this mobile device a lot of people use is inexplicably broken" situations (hello Mobile Safari) that I would never consider digging deeper.
Yes, exactly! If I gave up on my language's debugger or compiler every time I was troubleshooting something because I didn't trust it, I'd waste an absolute ton of time. I think finding a balance is important, but also not being afraid to consider the possibility that libraries, browsers and tools that you generally don't consider flawed might have their own issues.
use proper logging with log levels.. just being able to switch on debug level and get a wealth of info is a big step in the right direction.. my debug lines have method names and line numbers.. I take this so seriously that I have my custom logging method for bash and python (I script a lot in both).. I drop a lot of info lines as I code and switch them debug once I'm finishing up..
confirming fundamentals is another tried and tested method..
other than logging and methodology, having the right tools is important.. I use pudb for my debugger in python.. and bash -xv for shell..
error reporting is also important.. I use 'trap' and set -E in bash to capture all errors, provide a lot of info and even email it to me..
in python I have exception hook.. I even monkey patched threading to report its uncatched errors to the main threads exception hook (I use python2.7 for work.. this is fixed in latest python3)
during my dev phase I always give more time to coming up with a solution vs with running with an idea.. this allows you to engineer a solution..
and lastly.. in my down time I think about the problems I couldn't solve at the desk.. this could be when I'm about to sleep.. when I'm playing pool.. etc.. I find that solutions come easier at these moments..
Somebody once told me "the arrows are more important than the boxes".
It made me realize that focusing on the flows of data between systems resulted in better overall designs. You can always swap out the contents of a box later on.
Talking through the problem out loud. Ideally with someone else who knows what you're on about, but anyone willing to listen will do.
If you can't find a listener just find a place where you can talk to yourself quietly.
Basically offer up a detailed commentary of the problem and what you have tried so far. The slow, methodical pacing of actually speaking out loud, as opposed to mile-a-minute thinking in my head seems to help my brain spot problems and come up with solutions.
Working on real world problems without seeking help from Google/StackOverflow, even if it means reinventing the wheel to make it fit for your particular problem, or even going through pages of documentation/code to find the purpose of a single function. It may sound like a waste of useful time, but it slowly makes you self-sufficient to deal with any kind of issues later on.
For me I would say it's fault finding. Releasing software is easy but finding the sometimes very obscure issues that crop up once released is always a challenge, especially when bug reports are not always the most detailed.
When dealing with some of these it's often necessary to delve deep into the internals of how something is working. I once spent 4-6 months working on a fix which turned out to be a stack corruption on a RTOS but was hard to pin down as the system usually took over 2 weeks to crash.
I now mostly deal with embedded software so it can often be with conflicts of hardware which have never been tested together and with unpredictable results. I've also dealt with finding faults in software I had to maintain across several platforms/databases/etc written in a few languages.
For the most part it's been fun or rewarding with some difficult times. Delving into the depths of a system does take some time but gives a wealth of knowledge that can usually be applied to another problem
Learn a little bayesian decision theory. Not because you need it for the solution of your problem, but because of what it tells you about the process of doing so.
I've got a tongue-in-cheek presentation entitled 'How to seem a Genius at Debugging with this 1 Weird Trick' based on decision theory. One day I must write it up properly as a blog post.
- In bayesian search theory a good (sometimes optimal) approach is to maximise information gathered per unit effort
- In debugging, a step can vary by several orders of magnitude in amount of effort required to check a theory (eg, from 'type ls' to 'spend several hours implementing a theorised fix/tracking framework/etc)
- Consequently, the probability of a hypothesis can be quite low and still be the optimal thing to check first, if the effort to test it is also low.
So, the 'one weird trick' is to look at all the evidence you can look at very quickly, even if the connection to the problem is tenuous. 95% of the time the tenuous ones don't pay off, but when it does, you're colleagues are like 'WTF? how did you think of that?' and as advertised, you look like a genius.
Big picture: experience is the main way to get better at this. That, plus watching what more experienced folks are doing.
In terms of smaller strategies:
I think stepping back and asking "What is the real problem that needs to get solved here?" is very often the best thing to keep in mind when I'm stuck. Along with "How important is this, really?" These kinds of questions help you to evaluate whether a particular technical roadblock is actually worth solving, or whether perhaps it is not worth the time investment because some other solution (or just walking away) would be a better choice.
Also, I don't know if you need to hear the clichés, but context switching is a huge and important strategy for problem solving — whether rubber duck debugging, talking to colleagues, switching to another medium for thinking (like a white board), or just going for a walk around the block — these are often good ways to get unblocked.
My workload has changed from writing new systems and features to making surgical strikes on existing codebases for performance & reliability gains. Building up an empirical practice around something I don't yet understand or control has been great for exercising new and different problem solving muscles.
- Never stop learning! Luckily, nowadays we have plenty of resources(HN, technical books, podcasts, etc...) we can use to improve Domain Knowledge on a subject, this is extremely important because it will create the hypothesis set we can query to solve problems later.
- Being resources constrained. Working with too many degrees of freedom is not only a common UX antipattern but also a way to kill creativity. If it's possible in your company, I strongly recommend doing technical customer support twice per month. Imho, this is a great setup to improve your problem-solving skills.
The wealth of knowledge, experience and advice here is pretty unique. I always learn some new way to think about a problem or new actionable things to try every time I come here.
Teaching beginners, correcting their mistakes and seeing all the weird, out of the box ways how they attempt to solve common problems, sometimes successfully, sometimes not.
Also, working on legacy systems, and seeing why and how the theoretical drawbacks of some approaches manifest in practice, and other theoretically discussed risks really don't.
Focusing on minimizing risk without sacrificing value. This is about balancing engineering costs to both build and maintain and ideal business outcomes. The hardest problem is not so much the how to build something but what is the least amount of work you can put forward to get the job done (with quality and maintainability).
all the normal pain points are there: proprietary standards, 16 standards, major tooling fights, bandaid fixes that “we’ll refactor later”, tech debt, etc.
But the architecture is really really visible. And when you’re done you have a bike!
I usually read other's code a lot. I personally believe, reading other's code (at your workplace, open-source, etc.) is better than practicing. I am not sure if it's true but it worked for me.
It helped me a lot. I can write hundreds of non-trivial but well tested lines of code which passes all tests on the first try and is fit for code review in a couple of hours. I wouldn't be able to do that if I hadn't practiced competitive programming a lot.
This skill wont solve all of your problems, but if you have a lot of theories you want to test then your ability to implement them quickly is invaluable, and testing a lot of your theories is how you learn most of the other important skills related to software engineering.
I feel like it helps me in reducing the number of passes before I ship something (so less errors/better tests). I also recently left the industry so I'm not sure if it'll help in my current position as a grad student as well.
I always like the quote that people in the photography community use - I think it applies just as well to software. "The best camera is the one that you actually use"
1. Look at prior art. Many problems have been solved before. It can be more fun to dive in and create your own solution, but looking for previous solutions first often saves time, results in a better solution, or may show you that you don't need to write the code in the first place.
2. Write down the problem you are trying to solve before solving it.
3. Solve the problem multiple times before committing to a final solution (if you have time). Unless the problem is familiar, your first solution will not be the best one.
4. Learn when to ask for help. This depends on the scale of the problem, but in general if you aren't making progress within half a day, consider asking someone else for their help and ideas.
7. Do programming puzzles. I've found it helps a lot with software maintenance and other problem solving at the micro level. (It doesn't generally help with big systems design because puzzles tend to be smaller in scope than that, but a lot of our work is maintaining existing code.) I find https://exercism.io/ the best for this because you usually complete puzzles in the same environment you'll use for daily work (your text editor/IDE of choice), rather than in a web-based environment with limitations.
Read a lot of code. Pick one of the most popular libraries in a language you like and commit to reading it, understanding how it was architected, and tweaking it.
Talk to other people who build “big systems”, whatever you decide that definition means. Ask them lots of questions. (Not the “Emacs or Vim?” kind. Instead: “When you wrote x, where did you start? How did you arrive at the solution you found? What would you change if you started again today?”)
Build systems, big and small, on your own and with others.
If you're not sure what to build, I like “Exercises for Programmers” by Brian Hogan:
Watching people code can be dull or exciting depending on your mindset. I try to dip into these channels occasionally and soak up ideas from them, then go away and make things. I don't generally learn much without writing code too, and look for technical books and courses with exercises and problems.
We need more of this, though. I wish there was a programming puzzle site full of problem statements of the type, “build an application that does x”, where solutions are ranked by the community and sorted by language. It would be such a helpful resource.
Solve many different problems, make experiments and throw away code. Make connections between sub disciplines.
You have to have the attitude that if somebody understood this, then I can too, if nobody did, there's no good reason why I can't be the first one.
Fail a lot and learn. Think about every failure you had, but as a problem to solve not a comment on your value as a person/engineer. What could you have done differently? What were the signs that got you in that situation?
For tricky problems, what works for me is the obsess-and-let-go strategy. Work intensely on some problem for some time, if you make no progress, just let go of it, forget about it, do something else and perhaps your brain will connect the dots. Talk about the problem with other people. Explaining it and different points of view usually change your perspective enough that you are no longer stuck.
Also note that reading about something is not the same than doing it. You need to read and attempt to replicate, even if a toy version of the thing to better understand. Some things just take time an effort to seep in.