There are a couple subtleties to your question that I think are important:
- You're comparing an interpreted language (Python) to a compiled language (C++).
- Most scientific and engineering software is developed with a heavy Linux (and UNIX) bias, and is not usually known for cross-platform compatibility or great user support (big libraries, of course, excepted)
If you really want to use a C++ library, you probably want something that has a CMake-based build system. PLplot is one such library that looks reputable and has a C++ interface (unfortunately, the others that seem to come up are ROOT and MathGL...).
However, given the points above, if using Cygwin or MinGW is going to drive you into a murderous rage, I'd stick with Python or a visualization program you could install from a binary (like Paraview or VisIt, for instance; other programs could also work). Here's why:
Interpreted languages are convenient for tweaking visualizations
The big advantage of doing your visualization in C++ is that you can append it to whatever C++ code you might be using for simulation work. That said, if you have to compile your code after every modification you make, it's going to make tweaking your visualizations that much more time-consuming. With an interpreted language, once the data is loaded, you can post-process and tweak visualizations on the command line. If you plot raw data and decide it might look better normalized, no problem; a few lines of code, and you can bring up a new plot in a minute or two without having to recompile.
Compiling scientific software on Windows is a pain
...because much of the time, whoever wrote it didn't use a cross-platform build system like CMake, so you either have to slog through and hack together a Makefile that works, or use Cygwin or MinGW to compile everything and hope that the software doesn't rely on Linux libraries (because after all, Cygwin is a POSIX compatibility layer). If a project uses CMake, it probably means that the developers have thought about these issues, and CMake will output the necessary build scripts to compile software with Visual Studio, which will make your life less painful. Otherwise, it's probably not worth the time and effort.
If you can install the visualization software you need from a binary, you don't have to worry about grappling with Visual Studio. There are at least two Python distributions I can think of that will supply Matplotlib (Continuum's Anaconda and Enthought Canopy) with a binary install of Python on multiple platforms. Paraview and VisIt are both powerful scientific visualization packages that, again, can be installed from binaries (they can also be built from source on Windows).
I don't think there is a "standard C++ way" to plot data
I haven't heard of a "standard C++ way" to plot data (it doesn't mean it doesn't exist); when I talk to data scientists, a lot of them stress how convenient and important it is to be fluent in an interpreted language like Python, R, or MATLAB because visualization and post-processing can be developed rapidly in those languages. Standard practice for a lot of fields is to dump data to some text format (column-delimited, CSV, HDF5, VTK, etc.) and then read in the data using one of those interpreted languages or a dedicated visualization program (like VisIt, Paraview, and for 2-D plots, gnuplot).