Let's say that I have a binary that I am building, and I include a bunch of files that are never actually used, and do the subsequent linking to the libraries described by those include files? (again, these libraries are never used)
What are the negative consequences of this, beyond increased compile time?
In addition to what Sasha lists is maintenance cost. Will you be able to detect easily what is used and what is not used in the future when and if you chose to remove unused stuff?
In addition to compile time; Increased complexity, needless distraction while debugging, a maintenance overhead.
Apart from that, nothing.
If the libraries are never used, there should be no increase in size for the executable.
Depending on the exact linker, you might also notice that the global objects of your unused libraries still get constructed. This implies a memory overhead and increases startup costs.
I have never experienced any problems with linking a .lib file of which only a very small part is used. Only the code that is really used will be linked into the executable, and the linking time did not increase noticeably (with Visual Studio).
If you link to binaries and they get loaded at runtime, they may perform non-trivial initialization which can do anything from allocate a small amount of memory to consume scarce resources to alter the state of your module in ways you don't expect, and beyond.
You're better off getting rid of stuff you don't need, simply to eliminate a bunch of unknowns.
If the libraries you're including but not using aren't on the target system, it won't be able to compile even though they aren't necessary.
You mention an increase in the compilation time. From that I understand the libraries are statically linked, and not dynamically. In this case, it depends how the linker handles unused functions. If it ignores them, you will have mostly maintenance problems. If they’ll be included, the executable size will increase. Now, this is more significant than the place it takes on the hard drive. Large executables could run slower due to caching issues. If active code and non-active code are adjacent in the exe, they will be cached together, making the cache effectively smaller and less efficient.
VC2005 and above have an optimization called PGO, which orders the code within the executable in a way that ensures effective caching of code that is often used. I don’t know if g++ has a similar optimization, but it’s worth looking into that.
A little compilation here of the issues, wiki-edit it as necessary:
The main problem appears to be: Namespace Pollution This can cause problems in future debugging, version control, and increased future maintenance cost.
There will also be, at the minimum, minor Binary Bloat, as the function/class/namespace references will be maintained (In the symbol table?). Dynamic libraries should not greatly increase binary size(but they become a dependency for the binary to run?). Judging from the GNU C compiler, statically linked libraries should not be included in final binary if they are never referenced in the source. (Assumption based on the C compiler, may need to clarify/correct)
Also, depending on the nature of your libraries, global and static objects/variables may be instantiated, causing increased startup time and memory overhead.
Oh, and increased compile/linking time.
I find it frustrating when I edit a file in the source tree because some symbol that I'm working on appears in the source file (e.g. a function name, where I've just changed the prototype - or, sadly but more typically, just added the prototype to a header) so I need to check that the use is correct, or the compiler now tells me the use in that file is incorrect. So, I edit the file. Then I see a problem - what is this file doing? And it turns out that although the code is 'used' in the product, it really isn't actively used at all.
I found an occurrence of this problem on Monday. A file with 10,000+ lines of code invoked a function 'extern void add_remainder(void);' with an argument of 0. So, I went to fix it. Then I looked at the rest of the code...it turned out it was a development stub from about 15 years ago that had never been removed. Cleanly excising the code turned out to involve minor edits to more than half-a-dozen files - and I've not yet worked out whether it is safe to remove the enumeration constant from the middle of an enumeration in case. Temporarily, that is marked 'Unused/Obsolete - can it be removed safely?'.
That chunk of code has had zero cove coverage for the last 15 years - production, test, ... True, it's only a tiny part of a vast system - percentage-wise, it's less than a 1% blip on the chart. Still, it is extra wasted code.
Puzzling. Annoying. Depressingly common (I've logged, and fixed, at least half a dozen similar bugs this year so far).
And a waste of my time - and other developers' time. The file had been edited periodically over the years by other people doing what I was doing - a thorough job.
It could perhaps even fail to compile if the build tree isn't well maintained. if your'e compiling on embedded systems without swap space. The compiler can run out of memory while trying to compile a massive object file.
It happened at work to us recently.