C++ – yet more Questions and Answers

I am continuing to mine the rich vein of questions from my online C++ lectures, which I started two weeks ago. I am always pleased to receive questions about any aspect of embedded software by comment, email or through social media.

Another three questions to consider today …

I just wondering how the compiler and linker optimize C++ code. How do we compare the this optimization in different tool chains?

For the most part, a C++ compiler will just do the same kind of optimizations that all compilers do. Of course, if it is designed for embedded use, there will be fine control over those optimizations. A compiler does have the possibility to add some “hints” to help the linker. A linker, which is designed to optimize C++ utilization can do a number of things, for example:

• elimination of redundant code where template instantiation has been done on a per translation unit basis
• removal of unused out of line copies of inline functions
• elimination of redundant copies of a base class subobject, when a class is based on multiple classes with a common base
• on ARM EHABI, the exception handler table entries can be combined for adjacent functions whose unwind opcodes are the same; this can save quite a bit of space because most functions have very simple unwind opcode sequences, and a lot of them end up being the same

What would be the difference between C++ and C# ?

The C syntax has been the basis for many other programming languages. In the late 1980s and 90s, there were at least three object oriented languages that were based on C. C++ was just one of those. Another one which has survived is Objective-C. More recently Java and Python are examples of the syntax being “overloaded”. C# is yet another, but differs because it is vendor [Microsoft] specific. C# has almost no relevance to embedded developers, unless, I suppose, they are using Windows Embedded. Much the same comments may be applied to Apple’s newly announced Swift language. C# is a “managed language” [a term Microsoft made up], and C++ is not. Managed languages are compiled down to bytecodes [much like Java], and then JITed at runtime to hardware specific instructions.

Do you use the C++ template features and library a lot in your code? Sometimes programmers worry about code bloat.

This question raises a number of issues. Personally, I have no objection to using templates, as they are a legitimate “shorthand” way to code sometimes. There is a perception that they can lead to code bloat. This is for two reasons:

1. A function template, for example, may be quite a small piece of source code, but may yield a great many different instantiations, resulting in a lot of binary code. This is not “bloat” because there is no redundancy. I would probably prefer to do explicit function overloading so that the scale of the coding is visible.
2. As mentioned in the first question above, templates are normally instantiated in each translation unit, which could result in redundant code, unless the linker is capable of deduplicating [optimizing] it.

I am wary of template libraries, as many are not optimized for or compatible with the needs of embedded software.
Once again, I acknowledge the input from my colleague Jon Roelofs on this blog posting.