Some Boost.Test notes

These are just a couple of notes for some neat tips and tricks I’ve discovered over the years when using Boost.Test. They may not be all that useful to everybody else but they’re the ones I tend to forget about and then end up rediscovering. I’m using most of these with recent versions of Boost and these were tested with 1.54.

  • You can nest test suites. I don’t use this feature and was surprised to discover it, but it makes a lot of sense if you’re dealing with large test suites that can be easily broken down into smaller logical chunks.
  • If you arrange the tests in test suites (which you should anyway), you can run a single test suite by using -run_test=TestSuite. You can also run a single test inside a test suit or a nested suit by giving the appropriate path to -run_test, like so -run_test=TestSuite/TestName
  • Fixtures are normal C++ structs/classes so you can derive from them to create more specialized fixtures. For example, in a set of test suite used to test a fairly large set of COM APIs, I use a single fixture to initialize COM, then derive fixtures to set up the tests for each interface. This cuts down a lot on duplicated initialization code.
  • Boost.Test supports fixtures. Make lots of use of them to set up the environment for each of your tests instead of dumping a lot of the initialization code into each test case, which makes the test cases smaller, easier to read and you can follow the “let the first call in your test be the function you’re testing” mantra.
  • Members of the fixtures can be used directly in your unit tests
  • You can use BOOST_CHECK/BOOST_REQUIRE macros in the fixtures as well, although I would advise to make fairly minimal use of that. It does help if you are trying to test what I call “flaky” pieces of code.

Make sure you’re using the right command processor when running Incredibuild

Quick hack/warning for those using an alternative command line processor like TCC and also use Xoreax’ Incredibuild for distributed builds. Incredibuild is awesome, by the way, and if you have a larger C++ project that takes a long time to build, you should use it. And no, I’m not getting paid or receive free stuff for writing that.

However, if you have to start your Visual Studio instance from the command line because you need to set some environment variables first, or because of your general awesomeness, make sure you’re starting it from a stock Windows shell. Either the standard Windows command line (cmd.exe) or PowerShell will do nicely, thank you. If you start VS from TCC and have a couple of build tasks that spawn out to the shell, Incredibuild wants to shell out into TCC to run these tasks and the shelled out task don’t seem to return control to Incredibuild again. Yes, I was too lazy to investigate further as the method described above works.

Visual Studio 2012 C++ Variadic Template support in the runtime library

As VS2012’s C++ compiler doesn’t support “true” variadic templates, the new runtime library classes that use variadic templates are implemented using macro magic behind the scenes. In order to get the “variadic” templates to accept more than the default of five parameters, you’ll have to set _VARIADIC_MAX to the desired maximum number of parameters (between five and ten).

For more information, see the “faux variadics” section of this blog post on MSDN.

If you’re using boost::variant, you need to have a look at Boost 1.53

I was profiling some code a while ago that makes extensive use of boost::variant. One of the lessons from the profiler run was that boost variants appear to be fairly expensive to construct and copy.

As of 1.53, variants support rvalue constructors and rvalue assignment operators. My initial measurements suggest that when used with types that are “move enabled”, there is a benefit in upgrading to this version of boost variant, both in performance and memory consumption.

I don’t want to see another “using namespace xxx;” in a header file ever again

There, I’ve said it. No tiptoeing around.

As a senior developer/team lead, I get involved in hiring new team members and in certain cases also help out other teams with interviewing people. As part of the interview process, candidates are usually asked to write code, so I review a lot of code submissions. One trend I noticed with recent C++ code submissions is that the first like I
encounter in *any* header file is

using namespace std;

If we happen to review the code submission using our code review system (a practice I’d highly recommend), the above line is usually followed by a comment along the lines of “Timo is not going to like this”. And they’re right, I don’t.

So, why am I convinced that this is really, really bad practice despite a ton of (maybe not very good) C++ textbooks containing the above piece of code verbatim? Let’s for a moment review what the above statement does. Basically, it pulls the whole contents of the namespace “std” (or any other namespace that the author used the using statement
for) into the current namespace with no exceptions. And I mean *anything*, not just the one or two classes/typedefs/templates you’re trying to use. Now, the reason that namespaces were introduced in the first place is to improve modularization and reduce the chances of naming conflicts. It basically allows you to write the following
code and ensure that the compiler picks the correct implementations:

std::vector<std::string> names;
my_cool_reimplementation::vector<our_internal_stuff::string> othernames;

Now assume that we’re trying to reduce the amount of typing and put the above using statement in the code (or worse, use both namespaces) and write the following:

vector<string> names;
vector<our_internal_stuff::string> othernames;

If the author of the code is very lucky, the correct implementation of vector will be picked, at least initially. And some time down the road, you’ll encounter strange compiler errors. Good luck finding those – I’ve been in situations where it took days to track down this sort of problem. That’s a heck of a timewaster that you just got for saving to
type five characters.

Also, if you’re putting the using statement into the header file, you’re aggravating the problem because the conflicts you will run into sooner or later will be in a module far, far away for no apparent reason until you find out that three layers down, one include file happens to include the file that contains the using directive and suddenly polluted
whichever namespace the file contained with the whole contents of namespace std.

So why is using namespace std; found in so many textbooks? My theory is that it does reduces visual clutter and also helps with the layout of the book. In a dead tree book, you only have very limited amounts of space so you need to make the most of it. Code examples are generally fairly trivial – well, they have to be if you want the get the point across. Neither of these really hold true when it comes to production code in the age of 24″ monitors and text editors that can handle more than 60-80 characters per line (try it, it works!). So, don’t do it.

So, what can you do if you absolutely, positively have to use a using declaration in a header file? There are other ways to reduce the impact of having to do so – you can use one of them, or all of them in various combinations.

First, you can simply use a typedef. I would suggest that this is good practise anyway even if I don’t always follow my own advice. Using a typedef actually has two benefits – it makes a typename more readable and it documents the author’s intent if a well chosen name has been used. Compare the following declarations:

std::map<std::string, long> clientLocations;

typedef std::map<std::string, long> ClientNameToZip;
ClientNameToZip clientLocations;

The second declaration – even though it spans two lines – is much more self documenting than the first one and it gets rid of the namespace fuzz at the same time.

Another option is to limit the scope of the using statement in two ways – only “using” the names of the symbols you need to use, for example:

using std::string;

Again, just throwing this statement into a header file is almost as bad an idea as putting in “using namespace”, so you should limit its visibility by making use of C++ scopes to ensure that your using declaration really only affects the parts of the code that need to see your using declaration in the first place. For example, you could limit the scope to a class declaration:

namespace bar
  struct zzz

Unfortunately what you then can't do is simply pull in the namespace into a class definition in a header file:

class foo
  using namespace bar;

  zzz m_snooze; // Doesn'tpull in bar::zzz, would be nice though

What you can do if the design of your code allows for it is to pull the ‘zzz’ namespace into the namespace that contains class foo, like so:

namespace yxxzzz
  using namespace zzz;

  class foo
    zzz m_snooze;

This style has the advantage that you’re “only” cluttering a single namespace with your using directive rather than the whole global namespace. At least this way, you are limiting possible naming conflicts to a single namespace. But it’s still a headache if you are using the, err, using directive in a header file this way.

Another way to limit the scope of a using directive to a single function, for example like this:

void temp()
  using namespace std;
  string test = "fooBar";

This works in both header and implementation files.

Actually, I would write the above code like this, because there is no good reason (but plenty of bad ones) to pull in the whole std namespace if you only need part of it.

void temp()
  using std::string;
  string test = "fooBar";

In either case, you are restricting the visibility of a using directive to the part of the code that requires it rather than throwing it right into everybody’s way. The bigger your projects get, the more important it is to ensure proper modularisation and minimise unintended, harmful side effects.

Useful collection of Qt debug visualizers for Visual Studio

I had to reinstall VS2010 at work and because I clearly didn’t think this all the way through, forgot to save my autoexp.dat file before removing the old installation. And of course I didn’t realise what had happened until I had to dig deeper into some Qt GUI code that wasn’t quite working as expected, and of course I was prompted with the raw data.

Fortunately a quick search on Google led me to this page Human Machine Teaming Lab | Knowledge / Qt that contains a very comprehensive set of visualisers. I’d highly recommend them if you’re doing any sort of work with the Qt libraries. Just keep in mind that the Qt visualisers are for Visual Studio 2008 and 2010, so they’re anything but guaranteed to work with newer versions.

How to view undecorated DLL-exported C++ symbols in Visual Studio 2010

Yes, it’s one of those “note to self” posts, but I keep forgetting how to do it.

As the first step, you run dumpbin /EXPORTS and redirect the output into a file because the utility that unmangles the names (undname.exe) doesn’t appear to be able to take piped input via stdin. Then, run undname <filename>, with <filename> being the file that contains the exported symbols.

At least that way the symbols become mostly readable.