Hacker News Reader

Comment by waynecochran | Hacker News Reader

I’ve been writing C++ code since it’s inception in the 1980’s. In the last two weeks I’ve written close to 1000 lines and no raw ptrs… in fact no smart ptrs either… not even one c-style array. No worries about leaks or memory safety. It doesn’t look anything like C++ code I wrote even 10 years ago.

View on HN

I'm curious and would love to take a look. Any chance any of that code is available somewhere? My c++, little that I write, mostly looks like C and I'd love to learn something new.

Most of its in a private GitHub repo but I think I can open a piece of it… here is part of it ... a header only library for computing closest pairs...

https://github.com/wcochran/closest-pairs/blob/main/closest-...

Why is Eigen/Dense included if you never refer to anything Eigen and just take it as a duck-typed template parameter?

True.. good eye .. the operations on type P are very Eigen specific, but another type could easily provide those.

After giving your code a quick skim, I cannot see any mistakes in your vector indexing.

However, your code is full of vector accesses without bounds checking: vec[i] instead of vec.at(i).

That's just as "unsafe" as C, so I recommend not bragging about how far you've come.

Your code is full of what is essentially unprotected C-style

  *(addr + i) = ...

  ... = *(addr + i)

You've been writing C++ since the 1980's, but your code is full of raw pointer math, and yet... "No worries about memory safety"?

I have a feeling you'll be reaching for Valgrind or address sanitizer someday soon, just like the C programmers.

Not bragging at all.. it was just introspection on how different my code looks now… maybe I should find some older code that is fraught was dangerous constructs. In fact, early versions of this were segfaulting for the very reasons you mention.

Thank you for being honest.

This safe, "no worries" C++ was segfaulting. You've managed to make it correct today but it's still unsafe.

Someday one of your coworkers is going to break it and this code will be trashing the heap. Mysterious memory corruption failures are your destiny.

That said, Happy Halloween!

Your comments come across as pretty mean-spirited. Let me rewrite one line for you. You wrote:

> Someday one of your coworkers is going to break it and this code will be trashing the heap. Mysterious memory corruption failures are your destiny.

You could have written something like, "It's possible this code will break in the future and you won't even realize the cause, because it will appear as a mysterious memory corruption failure."

See how I left out a word like "destiny" and took out the whole bit about coworkers? Yeah, much nicer and less personalized.

And then I said "Happy Halloween".

The point was that being a C++ programmer is like being part of a horror movie.

I say that as a person who uses C++ in my day job. It's a nightmare.

I have a love/hate relationship with C++ ... probably like most C++ programers.

I have no love for C++.

The point of the post by my fellow HN person was to give you feedback on your communication style.

You obviously have interesting things to say and opinion to share with a fair amount of passion, but the tone detracts from the message, which I find a pity.

Why are people so delicate today?

We're getting so soft and so fragile, how does it end?

“Hard times create strong men, strong men create good times, good times create weak men, and weak men create hard times." (G. Michael Hopf)

It’s not about people being delicate, its about what you value more: saying whatever you feel like saying, or actually getting your point across. The more inflammatory your tone, the less likely it is that you’ll change anyone’s mind, because they’ll be more likely to dismiss outright whatever you’re saying.

This quote doesn’t exactly make you look like the badass you think you are.. and all over C++ array access. Does that sound like a topic that warrants escalation to confrontational language to you?

The most dangerous thing I do is index a `std::vector` for which I can add range checking -- which I only do when debugging -- I am not willing to pay for that at release time. I am not performing any address math that can not be statically or else dynamically checked. No explicit memory allocation anywhere -- C++ does all the memory management for me. This is not true of older code I have written.

Well, there is always a option to enable bounds checking in operator[](), all major compilers provide defines or compiler switches for that, and I would advise anyone that cannot prove otherwise via a profiler to do just that.

Uh no, vec[i] may be unprotected in release mode but all three major standard library implementations, MS STL, libc++ and libstdc++ have bound checking in debug mode (along with many other sanity assertions, such as correct sorting predicates, etc) ; this way you get the (horrendous) bound-checked cost when developing and can ensure decent performance in release.

Who really uses debug builds in modern C++? They are just too slow to be usable.

Debug doesn't have to mean unoptimized.

I do as far as is possible all of my development under my stdlib's debug mode with -fsanitize=address and -fsanitize=undefined. If it's slow like this on my 2018 workstation, it'll be slow in release mode on my users's 2011 dual core laptops and on raspberry pis.

but if you need a bit more speed you can still build with

    -O2 -D_GLIBCXX_DEBUG=1 -D_GLIBCXX_DEBUG_PEDANTIC=1 # for libstdc++
    -O2 -D_LIBCPP_DEBUG=1 # for libc++
    /O2 /D_ITERATOR_DEBUG_LEVEL=1 # for msvc

this way you get compiler optimizations but the assertions are still there

They compile a lot faster.

I do! How do you use a debugger otherwise?

A) I do B) plenty of “debug” checks can be enabled in a “release” build, e.g. just don’t define NDEBUG

I always use debug builds while testing. They are of course much slower, but far from unusable - and I am mostly writing audio code

Plenty of people, not everyone is doing AAA at 120 FPS.

You could have raised any of those technical poitns without coming off as an asshole.

To me the code looks like very reasonable c++, and quite modern. I suspect the comment about memory safety may have been related to leaning heavily on things like iota, copy, partial_sort, lower_bound, etc as opposed to doing those things “by hand”.

I think it's much better to double-check your i's than to use `vector.at(i)`. Being unchecked at runtime is the benefit.

Much appreciated, thank you!

Most Google C++ code is written like that, check out their repos on GitHub

A project that my team works on might give you an impression: https://github.com/arbor-sim/arbor

That is some hard core science simulation...

If you consider thread safety an aspect of memory safety, I still see those issues pop up in modern c++. And memory leaks in terms of ever growing maps still happen. But the only times I have seen use after free or similar in modern code is where there was a bit that wasn't so modern. Is there some list for clang tidy modernize that will treat using anything other than value semantics as a warning?

He's just talking about using std::vector everywhere.

That and several other std types + algorithms…

But std::vector has been around since C++98 (and in the STL even before then). What's changed?

A lot. Reading Scott Meyer’s Effective Modern C++ makes me realize how different things are. Just providing a type doesn’t do it… it’s the whole ecosystem around it.

it was such a PITA to use in 1998 ...

I was called inexperienced because I was saying pointers are not 100% necessary in C++ anymore.

I'm not even certain if using stl containers always use the heap.

I'm a bit confused at times, since I avoid using oop and inheritance, I prefer data oriented, I just write functions and avoid side effects as often as possible.

What are the new language features that let you avoid using smart pointers?

Reply on news.ycombinator.com