fallthrough

[John E]

I'm building a project which needs to get built with VC2005 for various reasons. One of the files has a switch statement looking like this:-

		switch (argc) {
			case 9:
				if (types[8] == 'f') {
					linkid = (int) argv[8]->f;
				} else {
					linkid = argv[8]->i;
				}
			case 8:
				if (types[7] == 'f') {
					linkset = (int) argv[7]->f;
				} else {
					linkset = argv[7]->i;
				}
			case 7:
				if (types[6] == 'f') {
					port = (int) argv[6]->f;
				} else {
					port = argv[6]->i;
				}

				// etc
		}

Notice that there are no break; statements in between each case. Someone else builds the same project with gcc and he's added some [[fallthrough]] lines at the end of each case - i.e.

		switch (argc) {
			case 9:
				if (types[8] == 'f') {
					linkid = (int) argv[8]->f;
				} else {
					linkid = argv[8]->i;
				}
				[[fallthrough]];
			case 8:
				if (types[7] == 'f') {
					linkset = (int) argv[7]->f;
				} else {
					linkset = argv[7]->i;
				}
				[[fallthrough]];
			case 7:
				if (types[6] == 'f') {
					port = (int) argv[6]->f;
				} else {
					port = argv[6]->i;
				}
				[[fallthrough]];

				// etc
		}

I don't know if later versions of MSVC would accept this but VC2005 doesn't (giving me:- error C3409: empty attribute block is not allowed).

Obviously I could just comment out the lines but I just wondered if fallthrough is a valid keyword of some kind later versions of C++ ?

[wolle]

but I just wondered if fallthrough is a valid keyword of some kind later versions of C++ ?

It's a standard "attribute specifier sequence" in C++ 17,

https://en.cppreference.com/w/cpp/language/attributes

I had no idea this feature even existed but I cannot be the only one since no one knows all of C++ except Bjarne.

[2kaud]

Yes, there's lots of 'goodies' in c++17! Unfortunately, for pre c++17 compilers, you can't even define a null macro for [[fallthrough]] as an identifier can't start with a [ Doh!

For info about c++17 and MS, see https://blogs.msdn.microsoft.com/vcb...he-c-standard/

Also see http://en.cppreference.com/w/cpp/com....2B17_features for details of what's in the various c++ standards (including so far in the new c++20) and which compilers/versions support them.

[John E]

Thanks guys - it looks like C++17 is supported from VS2015 onwards so I guess I could change all instances of [[fallthrough]] to fall_through and then do something like this:-

#if _MSC_VER < 1900
#define fall_through
#else
#define fall_through [[fallthrough]]
#endif

[2kaud]

it looks like C++17 is supported from VS2015 onwards

Not quite. Only a quite small part of VS2017 is implemented in VS2015. To get the full c++17 implementation, you need the latest version of VS2017 (15.7). Even early versions of VS2017 didn't fully support c++17!

[[fallthrough]] is only implemented in VS2017.

[Arjay]

Failthroughs [sic] are difficult to maintain and debug. Is there no other way to structure the code?

[2kaud]

Failthroughs [sic] are difficult to maintain and debug. Is there no other way to structure the code?

But even using case fall-throughs, the code can be simplified as

switch (argc) {
    case 9: linkid = (types[8] == 'f') ? (int)argv[8]->f : argv[8]->i;
    case 8: linkset = (types[7] == 'f') ? (int)argv[7]->f : argv[7]->i;
    case 7: port = (types[6] == 'f') ? (int)argv[6]->f : argv[6]->i;

    // etc
}

which makes it much more obvious about the 'fall-through' - even without [[fallthrough]] being used. IMO actually using [[fallthrough]] in this simplified code would detract from the code readability.

[Arjay]

Years ago I wrote argument parsing classes for c++ and c#. An instance of the class describes the header, footer, and arguments. The end result is consistent, self documenting (via /?) cmd arguments.

Any cmd tools utilities I write use the same classes and it's easy to setup and one less thing for me to debug.

[wolle]

Failthroughs [sic] are difficult to maintain and debug. Is there no other way to structure the code?

One way would be to use a simple "state machine" implementation, like

	int state = args; // start state
	while (state > 0) {
		switch (state) {
			case 9:
				// do something in this state
				state = 8; // goto some other state
				break;
			case 8:
				// do something in this state
				state = 7; // goto some other state
				break;
			case 7:
				// do something in this state
				state = 0; // exit
				break;
			default: // error - state missing
				state = 0; // exit
				break;
		}
	}

Not much of a difference maybe but there's no fallthrough.

[2kaud]

if I'm reading the code right, you could consider something like

	const char types[] = "fzfe ffyf";

	int var1, var2, var3, var4, var5, var6, var7, var8, var9, var10;

	// Array of pointers to variables to be set
	int *pvar[] = {&var1, &var2, &var3, &var4, &var5, &var6, &var7, &var8, &var9, &var10};

	const size_t st = sizeof(types) / sizeof(types[0]);
	const size_t sa = sizeof(pvar) / sizeof(pvar[0]);

	static_assert(st == sa, "Sizes of the arrays not the same");

        // Set the required variables
	for (size_t a = std::min(argc, sa); a > 0; --a)
		*pvar[a - 1] = (types[a - 1] == 'f') ? (int)argv[a - 1]->f : argv[a - 1]->i;

which should be OK with VS2005?

But if I recall your other posts, you're just trying to get code given to you to work with VS?