Last time I did anything on the job with C++ was about 8 years ago. Here’s what I learned. It may still be relevant.
C++14 was alright, but still wasn’t everything you need. The language has improved a lot since, so take this with a grain of salt. We had to use Boost to really make the most of things and avoid stupid memory management problems through use of smart (ref-counted) pointers. The overhead was worth it.
C++ relies heavily on idioms for good code quality that can only be learned from a book and/or the community. “RAII” is a good example here. The language itself is simply too flexible and low-level to force that kind of behavior on you. To make matters worse, idiomatic practices wind up adding substantial weight to manual code review, since there’s no other way to enforce them or check for their absence.
I wound up writing a post-processor to make sense of template errors since it had a habit of completely exploding any template use to the fullest possible expression expansion; it was like typedefs didn’t exist. My tool replaced common patterns with expressions that more closely resembled our sourcecode^1^. This helped a lot with understanding what was actually going wrong. At the same time, it was ridiculous that was even necessary.
A team style guide is a hard must with C++. The language spec is so mindbogglingly huge that no two “C++ programmers” possess the same experience with the language. Yes, their skillsets will overlap, but the non-overlapping areas can be quite large and have profound ramifications on coding preferences. This is why my team got into serious disagreements with style and approach without one: there was no tie-breaker to end disagreement. We eventually adopted one after a lot of lost effort and hurt feelings.
Coding C++ is less like having a conversation with the target CPU and more like a conversation with the compiler. Templates, const, constexpr, inline, volatile, are all about steering the compiler to generate the code you want. As a consequence, you spend a lot more of your time troubleshooting code generation and compilation errors than with other languages.
At some point you will need valgrind or at least a really good IDE that’s dialed in for your process and target platform. Letting the rest of the team get away without these tools will negatively impact the team’s ability to fix serious problems.
C++ assumes that CPU performance and memory management are your biggest problems. You absolutely have to be aware of stack allocation, heap allocation, copies, copy-free, references, pointers, and v-tables, which are needed to navigate the nuances of code generation and how it impacts run-time and memory.
Multithreading in C++14 was made approachable through Boost and some primitives built on top of pthreads. Deadlocks and races were a programmer problem; the language has nothing to help you here. My recommendation: take a page from Go’s book. Use a really good threadsafe mutable queue, copy (no references/pointers) everything into it, and use it for moving mutable state between threads until performance benchmarks tell you to do otherwise.
Test-driven design and DevOps best-practice is needed to make any C++ project of scale manageable. I cannot stress this enough. Use every automated quality gate you can to catch errors before live/integration testing, as using valgrind and other in-situ tools can be painful (if not impossible).
1 - I borrowed this idea from working on J2EE apps, of all places, where stack traces get so huge/deep that there are plugins designed to filter out method calls (sometimes, entire libraries) that are just noise. The idea of post-processing errors just kind of stuck after that - it’s just more data, after all.
I once had a bug in a C# program I wrote. It made a HTTP request and if the user agent was left to default (whatever that was), the server just gave back an empty string as a reply. I took way to long until I understood what was going on and I kept chasing async, thinking I had messed it up some how.
For those of you who’ve never experienced the joy of PowerBuilder, this could often happen in their IDE due to debug mode actually altering the state of some variables.
More specifically, if you watched a variable or property then it would be initialised to a default value by the debugger if it didn’t already exist, so any errors that were happening due to null values/references would just magically stop.
Another fun one that made debugging difficult, “local” scoping is shared between multiple instances of the same event. So if you had, say, a mouse move event that fired ten times as the cursor transited a row and in that event you set something like integer li_current_x = xpos the most recent assignment would quash the value of li_current_x in every instance of that event that was currently executing.
My favourite thing ever seen in source code was a comment that read “this code is temporary” with developer initials and a date that was at the time about 5 years ago.
Followed by another dated comment from about 3 years later that read “Temporary my ass”
I’ve gone back and forth between a common bashrc file in my Dropbox folder that is symlinked to ~/.bashrc on my devices, and one that is imported from a regular bashrc instead, and recently it ended up in a state where it accidentally tried to do both, resulting in an endless loop. I discovered this on my Pop!_OS PC, which reacted to this situation by crashing on login lol. What??
programmerhumor
Active
This magazine is from a federated server and may be incomplete. Browse more on the original instance.