[ses4j]

I disagree strongly with a couple of your points.

- No need to manually manage memory.

That isn't true at all. You know enough memory management to avoid memory issues in your example. C++11 didn't save you from needing to understand it -- you understand it so avoided needing it.

- Compile time with -O3 is roughly the same as Python VM startup and has to be done only once.

It still requires two separate steps. Also, compile gets slower as the program gets larger, which isn't nearly as true for Python.

[Pxtl]

Thinking it over, memory management isn't really a good argument here. Even if your program leaks, if we're talking about a job script? Who cares, the OS will clean it up when the process closes. And either way, stack allocation and references will do just fine for small jobs, you shouldn't need to be doing lots of pointer stuff for a little script.

There are many sane subsets of C++ that save us from worrying about memory. If you just keep everything on the stack or in an auto_ptr, you should do fine.

[pcwalton]

This is not true. auto_ptr (in C++11 unique_ptr) and stack discipline are not memory safe. There is no sane subset of C++ that is memory-safe. I can provide (and have provided—see my posting history) dozens of examples.

[ajross]

That's missing the point though. Broadly speaking, when people talk about languages with automatic storage management being "safer" they are not talking about a correctness proof of their memory handling. In fact some languages, perl among them, fail to be safe from leaks in all cases, yet no one flips out about it.

The point is practical: is the language as typically used subject to routine "accidental" memory leaks? That's surely true for C, and remains true for most C++ idioms used up until the last few years or so.

It's not true of the kind of RAII style being talked about in the linked article. In that style it's routine to write large projects that literally never call operator delete, and need to resort to an operator new only in rare circumstances (often for compatibility with older APIs).

Modern C++ when used at this level[1] really does have the same kind of casual robustness against leaks and free-memory issues that you expect to see from garbage collected environments. And it's not even hard.

[1] Which is not to say that all contemporary C++ can be written in this model. Obviously if you're doing syscall-level code you'll need to be touching memory (and probably the heap) directly. But that's sort of the point.

[pcwalton]

> The point is practical: is the language as typically used subject to routine "accidental" memory leaks? That's surely true for C, and remains true for most C++ idioms used up until the last few years or so.

If this were true, we would expect to see large C++ codebases without memory-related security vulnerabilities. But the security history of every large C++ codebase that I have seen or heard of says otherwise. I would love it to be true, but I don't think it's a tenable position that C++, even "modern" C++, is memory-safe in practice.

We can argue over whether the C++ deployed in practice is "real" modern C++, but I think that enters into no true Scotsman territory really quickly. The fact is that C++ is not memory-safe in theory and has not been shown to be memory-safe in practice. For example, I know of real security bugs in Firefox that were caused by issues that are not fixed by any "modern" C++ idioms.

[ajross]

> If this were true, we would expect to see large C++ codebases without memory-related security vulnerabilities.

OK, we're talking past each other. The linked article and my point was about C++'s suitability for achieving software quality in tasks that are traditionally done by "scripting" languages. Security analysis is an entirely different world, and I tend to agree that other languages have a head start there as far as memory safety.

But that said, "memory safety" is hardly a big contributor to the overall vulnerability list. C++ is much less used on web backends, and it's likewise true that almost no large web service codebase exists without non-memory-related security vulnerabilities. I don't know if there are any deployed Rust codebases of this size, but I'd expect them to have their share of whoppers too.

[pcwalton]

I agree with you that C++ is often "safe enough" for tasks that aren't security-critical: log processing or scientific computing, for example.