I\'m interested in programming languages well suited for embedded programming. In particular:
Is it possible to program embedded systems in C++? Or is it better to u
Or is C++ OK only if some features of the language (e.g. RTTI, exceptions and templates) are excluded?
It's good to be thinking along these lines. Compile-time complexity is not a big deal, but runtime complexity has a resource cost.
C++ facilitates class/namespace modularity (e.g. method foo()
in more than one context) and instance modularity (member field bar
belonging to more than one object), both of which are a big advantage in software design. There are also features like const
, references, static casts, and templates, which can help enforce constraints and have little or no runtime cost.
I would not exclude templates. They're complex to think about and you need a compiler that handles them well, but the resource cost is almost all compile time -- what's going to "cost" you is the fact that each time you use a template with different class parameters, you produce a new set of code to instantiate member functions. But you would almost certainly have to do the same thing without templates. Furthermore, templates allow you to design and test libraries for general circumstances, in separate files that are instantiated at compile time rather than link time. Just to clarify that: templates allow you to have a file A.h that you test. Then you use it with file B.h or B.c to instantiate it at compile time. (A library would be linked in rather than compiled, and this makes it less flexible -- template methods can be optimized out so they do not incur a function call.) I've used templates in embedded systems to implement CRC code and fixed-point math: I can test the general code, put it in version control, and then reuse it multiple times by writing a simple class that derives from a template or has a template member field. The classic example of course is STL.
RTTI and exceptions: these add run-time complexity. I don't have a good idea of the resource cost but I expect RTTI would be fairly simple (just adds a type tag, costing extra space) whereas exceptions are probably beastly, involving stack unwinding.
virtual functions: I used to rule these out because of the memory + executiontime costs (minimal but still there), as well as the complexity of debugging, but they allow you to decouple objects from each other. If you don't use virtual functions, when an instance of one class (e.g. Foo) needs to execute code associated with an instance of another class (e.g. Bar), then the first class needs to know everything about the second (to compile Foo you need to have static linkage to all the methods in Bar) -- this adds a lot of tight coupling.
dynamic memory allocation: this is another big thing (that is in C as well), which we avoid like the plague at my company -- not only are there all sorts of errors that can arise, but the big runtime cost is the allocator/deallocator, and you've got to be willing and able to know what that cost is and accept it.
edit: I would love to use Java instead of C++ in the embedded world. Unfortunately my choices are limited and the runtime resource costs (code size, memory size, garbage-collecting time constraints) are too high in the space that I work in. My reason for using Java is less because of its runtime goodies and more for the fact that its software design is much cleaner, and the tools are much better (OMG! refactoring! woohoo!)... the key to me seems to lie with two things, which make C/C++ feel very clunky in comparison:
that everything is an object and all the methods are virtual, so you can lean heavily on the abstractions of interfaces.
the interface/implementation separation in Java is not this clunky ugly .c/.h
file splitting thing that makes compilers so slow. In constrast, I use Java in Eclipse and it automatically compiles the code right as I edit it. This is huge! I find most of my errors right away. In C/C++ I have to wait for a whole compile cycle.
Someday I hope there will be a language between C/C++ and Java that provides the advantages of Java for software development, without requiring the bells and whistles that make Java so attractive for desktop/server applications but unattractive in most of the embedded world.
It really boils down to what hardware platform you're operating on and hence what software platforms are open to you. For a lot of recent embedded kit - designed around a system-on-chip, a megabyte or two of RAM, a few devices - you really want a small operating system to manage the low level hardware while you concentrate on your application. Your choice of OS then constrains your available language set.
It's certainly possible to use C++ in the embedded space, but the full feature set of the language takes a lot of work to port correctly. For example, eCos is implemented in a mixture of C and what you might call the structural aspects of C++; support for RTTI, exceptions and the STL are lacking in the free version, though there are people working on this (and a commercial port available).
Similarly, it's possible to use Java - I know, I've ported a JVM to an embedded environment; it wasn't fun - though you usually get a cut-down Java language configuration, often something based on J2ME.