Python vs C++

[Christian Gollwitzer]

Am 21.08.14 14:54, schrieb David Palao:
> I consider myself a python programmer, although C++ was one of the
> first languages I learned (not really deeply and long time ago).
>
> Now I decided to retake C++, to broaden my view of the business.
> However, as I progress in learning C++, I cannot take out of my head
> one question
>
>   Why to use C++ instead of python?

You are asking in the wrong group; ask this in comp.lang.c++ Here most 
of the programmers know Python well and maybe some C++

> It is not ranting against C++. I was/am looking for small-medium
> projects to exercise my C++ skills. But I'm interested in a "genuine"
> C++ project: some task where C++ is really THE language (and where
> python is actually a bad ab initio choice).

The truth is, that both languages have a fairly large overlap. C++ spans 
a very wide gap, from tight low-level loops and direct memory access 
"close to the metal" up to composing a GUI application from ready-made 
building blocks. There is a variety of programming paradigmata, like 
functional (so-so), object oriented (quite good), generic (very good), 
imperative (shiny = the C subset). Graphically

C++:
metal |----------------------------------|

Python:
metal         |------------------------------------|


If your application lives within the overlap region, and a lot of them 
do, the choice is a matter of taste. I'm not even convinced that the 
development time is significantly lower in Python within this overlap. 
It becomes different at both ends: The more you go to the higher level, 
the more will Python outperform C++ in terms of development costs, but 
conversely C++ will win if you go closer to the metal. Then, as a Python 
programmer, you will find yourself rewriting parts of the application in 
Cython, trying PyPy, numpy, finally embedding C...

A few arguments outside of "what I can do with it" have already been given:

* For deployment, it is nice to compile and link a self-contained small 
binary. Try that with matplotlib - pyinstaller gets me a 100MB 
executable containing OS libraries, while the linker in C++ is able to 
only link what is needed.

* Access to the hardware: C structs, raw pointers, unboxed datatypes are 
needed in a painless and efficient way if you want to write, say,  a 
middle-level device driver

* performance is not only speed, but also memory usage. In most cases a 
C++ program will consume less memory

* static type checking can find bugs at compile-time. C++ is easier to 
compile (though on of the harder languages for compiler implementers), 
and therefore many tools exist for static analysis.

> The usual argument in favour of C++ (when comparing to python) is
> performance. But I'm convinced that, in general, the right approach is
> "python-profiling-(extension/numpy/Cython/...)".

Well, that's Ousterhouts dichotomy.

	Christian

	Christian