Hacker News Reader

Comment by tqaky | Hacker News Reader

I'm guessing you've never used Python for a serious project before, because your statement is incorrect. Aside from the fact that Python is memory safe, my experience has been that it is far easier to avoid logic errors in Python than in C/C++.

Equivalent Python code is shorter and less noisy than C/C++, and Python makes it easier to create succinct and simple to use abstractions. The result is that code is far easier to audit. Furthermore, my experience has been that it is much easier to create unit and integration tests in Python, which means that people actually do it and/or test more comprehensively.

My experience has been that if the performance cost of Python is no object then there is a correctness benefit over C/C++ for a disciplined programmer. I suggest you gain some more experience with both languages before making an offhand comparison.

In my experience it's much harder to write Python that won't crash than C/C++ that won't crash. With Python, rarely taken code paths can have very dumb errors in them (e.g. accidentally introducing a new variable with a different name than the variable you wanted or typoing a method name, or passing arguments with the wrong type) that would get caught by a C/C++ compiler but won't be caught in Python until the crash happens some months later. Typically in "reliable" code you're not going to do much dynamic allocation anyway, at least past the initialization stage (because memory fragmentation!) so memory errors aren't that common anyway.

Yes, probably Rust and Ada provide even more guarantees...

C/C++ memory errors are literally responsible for 70% of ALL critical security bugs. Most python "dumb errors" can be caught by linters. Memory is infinitely harder.

Source -- https://msrc-blog.microsoft.com/2019/07/16/a-proactive-appro...

The linter can't tell that the method doesn't exist until runtime, can it? If so, it wouldn't accept lots of valid python... Also, what linter should I use?

Either way, not all software needs to be secure to an adversary (since it's running behind several firewalls and if the attacker has shell access, then it's already game over).

And Python is implemented as a large C program.

66% python, 32% C - for cpython.

94% python, 5% C - for pypy.

cpython is the reference implementation and 32% means about 350,000 lines of code. My point stands.

So is glasgow haskell compiler or ocaml compiler - so what?

> ...there is a correctness benefit over C/C++ for a disciplined programmer.

Businesses want cheap programmers, not disciplined programmers. It's a great utopia if everyone just follows the best practices and does it perfectly all the time, but I guarantee that as margins become thin people writing, say, code for autonomous vehicles are doing the _very_ minimum to meet correctness requirements.

Genuinely curious (I don’t work with low level systems day-to-day), but why would C/C++ be better for safety?

I would imagine that the risk of bugs related to memory access and data racing would be pretty high, compared to a higher-level language or something like Rust that focuses on safety.

AFAIK (and I am not an expert) when they use C/C++ in automotive systems (in the automotive part, not elsewhere) they might tend to use deterministic programming (f.e. fixed-length arrays, no allocations, etc.) which eliminate most of the bugs.

But, again, I am not an expert and Toyota's sudden acceleration problem was said to be caused by bad C programming (said by some, not by others).

I thought it was caused by loose floormats.

Tooling and experience. In principle, there are other languages that might be better than C or C++ for safety-critical software. You know all the hoary jokes about the difference between theory and practice?

> You know all the hoary jokes about the difference between theory and practice?

Only in theory; do you have a practical example?

Can't a guy leave anything as an exercise to the reader anymore?

If you really want safety, use Rust or Ada.

Or Nim [1]..kind of an Ada with Lisp macros and more Pythonesque surface syntax.

All three are ahead-of-time compiled/less REPL friendly than Julia, though. Taking more than 100 milliseconds to compile can be a deal breaker for some, especially in exploratory data analysis/science settings where the mindset is "try out this idea..no wait, this one..oops, I forgot a -1" and so on. In my experience, it's unfortunately hard to get scientist developers onboard with "wait for this compile" workflows.

[1] https://nim-lang.org/

The Python-like syntax is a minus for me. I didn't care for semantic leading whitespace in ABC. I don't care for it in Python. I therefore doubt I'd care for it in Nim. If that's your thing, an Ada with Lisp-like macros sounds delightful.

It is definitely more possible in Nim than in Python to use parens () in many (but not all) places to be more like bracy languages (but with parens being the brace..).

I've actually become somewhat interested in Hy lately. Having a Lisp that targets the Python AST may come in handy. If there was a Nim facility for full Lisp or a very Lispy syntax that could be really nice.

https://github.com/hylang/hy

Even if you take, 100ms to compile, in practice, it takes about 2 seconds since it requires the user to type in another command and register that it has finished.

Is C/C++ really such a better choice in safety-critical systems? It's notorious for having all sorts of buffer overflows and memory issues on unexpected input.

MISRA C or MISRA C++ are used, enforcing much stricter guarantees than what the C or C++ specs provide.

IMO MISRA C, CERT C, and CERT C Secure should be standardized together into a compiler with a different language name. I'm aware there are other tools. If the compiler itself enforced everything to do with those standards and rejected violations as invalid code, we could have a much improved grounding for that language. There are already languages similar to a safer C, so we know there's a demand.

like dropping memory safety! oh, wait…

> tons of C++/C engineers needed ... educational background is irrelevant, but all must pass hardcore coding test.

https://twitter.com/elonmusk/status/1224182478501482497?lang...

Reply on news.ycombinator.com