> inefficient abstracted programming models where two years down the road
> you notice that some abstraction wasn't very efficient, but now all
> your code depends on all the nice object models around it, and you
> cannot fix it without rewriting your app.
The opposite can also be said. By re-inventing a simple data structure, you can end up 2 years later in the same position where your structure is inefficient (list vs. array, etc) and you end up having to rewrite your app. While implicit operator overload was probably a C++ mistake, it does allow to replace some kind of structures with other one without changing the code relying on them.
Other than that, I kind of agree that C++ is kind of terrible. So is perl, php, js or any other languages that send a barrage of features to the programmer without managing to educate them on their impact on performance and memory structure. Most C++ devs know about vTables and that's about it. I would also argue that C isn't a very good language either. It may be simpler to "get" what actually happen when you run your code, but has sub par code organization, total lack of "pointer flow" tracking, causing huge security traps and has an evil reliance on casting for any kind of structure/callback abstraction. All of those things could be "fixed" in new languages, but beside Rust, it seem there is no leverage for proper system level languages these days. OOP is probably a better choice for business programming anyway, but that's not even half of the market.
> While implicit operator overload was probably a C++ mistake
are you talking about the implicit conversion operator or operator overloading in C++?
For the former, it has been vastly improved in C++11 with the addition of explicit conversion operator. For the later, some very important capabilities of C++ rely on operator overloading namely the assignment operator (copy/move assignment) and the function operator (allows what C++ calls functors). Another useful thing with operator overloading is allowing the creation of generic functions that work on existing types (int, float, ..) and user defined types, for example std::accumulate.
The main issue I have with them is "brave" libraries developers who think it is a great idea to overload ->, . and delete. This can create issues where the code you read doesn't actually do what it look to do. I know about smart pointers and all, but really, those "features" are misused.
Scripting languages are not meant to be fast. For speed, we have C. A script spends most of its time in its underlying C components anyway. Therefore, there is not even a need for these scripting languages to be fast.
My comment wasn't about fast "scripting" languages, for whatever your definition of that is. I don't think stronger typing and memory constraint enforcement have anything to do with scripting languages. In my opinion, C lack them by design, not be necessity. A language could, in theory enforce them without runtime penalty and without losing too much flexibility (such as array aliasing).
C++11/14 "auto" is an example where someone can have the equivalent of a void* with compile time type checking and "no" runtime penalty. Yes, multiple specialized version of the method will be created and increase the binary size and potential cache fault, but much of them will (ok, should...) be eliminated/combined again during LTO. A slightly incompatible fork of C could also remove array to pointer demotion, sizeof() would then be consistent and potential overflow would be possible to catch using branch analysis or -fsanitize=.
In C++11, auto just says, "make the type of this thing the same(ish) as the type of its initializer". There is no potential for code bloat, even in theory. In C++14, auto can be used for generic lambdas, but even those are almost always called with a single set of argument types, so again no code bloat even in theory.
(Template code bloat itself is nearly mythical. I've seen exactly one occurrence of it causing a problem, in 8 years of Standard Library maintenance.)
My biggest problems with C++ (one of my favorite languages)
1. ABI issues.
2. Template Error messages.
3. Templates need be defined in header files (generally).
4. Object Slicing.
5. Cross platform Unicode/ASCII string encoding.
I understand the reasons behind these, none the less they are really annoying.
Any sufficiently complicated C program contains an ad hoc, informally-specified, bug-ridden, slow implementation of half of C++.
For example:
1) Macros for data structures instead of templates
2) Virtual functions emulated with function pointers
3) goto fail instead of exceptions
What's a left-right rule? Arrow binds tighter than deref, just like multiply binds tighter than plus. That's * (x->a->c)=5, just like how a * b * c + d * e * f works. C and C++ have some confusing precedence (famously, bitwise), but this is not one of them.
Yeah, it's alway funny to see the reaction of fresh-from-the-mill CS-academics when they have to code in an MISRA regulated project for the first time.
Around half of them quits after a month because they are unable to write 100 lines of working code without some babysitter (aka. garbage collector) cleaning up behind them.
C++ started out as a great 'next generation C', but in the last years it became a playfield for CS-academics, just adding one useless esotheric feature after each other.
If you ask Bjarne, I'm sure he'd have a position that would be based on his understanding. Same as asking Linux about Git leads him to answer: it is elegant and simple.
I think the question is, what makes reality - an individual opinion, or the mass-consensus? This is always a delicate question to ask.
> Unfortunately, I haven't had any time inte the last few days to code, nor read
mail. I'm assuming that there is no point in me finishing the patch and that git
will go with the strbuf solution? /Lukas
Everything I've ever wanted to be able to do with C++, I can do with Lua. I see it as the obvious progression from C - though I still find C a highly useful language. Putting Lua on top of whatever needs to be low-level, written in C, provides the next-generation toolset that I find so very productive.
The fact that C++ is vast is well known. If a language supports some arcane things then it will end up in the code (even if its very less percentage wise). Because its arcane, most of the developers will not be able to deal with it effectively. Example of arcane: overloading comma operator.
The opposite can also be said. By re-inventing a simple data structure, you can end up 2 years later in the same position where your structure is inefficient (list vs. array, etc) and you end up having to rewrite your app. While implicit operator overload was probably a C++ mistake, it does allow to replace some kind of structures with other one without changing the code relying on them.
Other than that, I kind of agree that C++ is kind of terrible. So is perl, php, js or any other languages that send a barrage of features to the programmer without managing to educate them on their impact on performance and memory structure. Most C++ devs know about vTables and that's about it. I would also argue that C isn't a very good language either. It may be simpler to "get" what actually happen when you run your code, but has sub par code organization, total lack of "pointer flow" tracking, causing huge security traps and has an evil reliance on casting for any kind of structure/callback abstraction. All of those things could be "fixed" in new languages, but beside Rust, it seem there is no leverage for proper system level languages these days. OOP is probably a better choice for business programming anyway, but that's not even half of the market.