Christopher Wright wrote:
However, if a patch exists, there is only one excuse for not including
it: lack of testing.
And there is one huge reason that nobody submits
additional test cases to your DMD test suite -- you've never released
it, or even specified the required format.
That's because of its uncertain copyright status. It's a collection of
everything that went wrong in the past, from a wide variety of sources.
As to its format, it is designed to be run by a shell script. Each
source file is expected to compile and run without error, error means
terminating with an exception or a non-zero return from main().
In other words, the test suite is designed to not require manual
inspection of results, and it halts on the first failure.
A typical test case would look like:
/***************************************************/
template A()
{
static T foo(T)(T t) { return 7 + t; }
}
struct Bar
{
mixin A!() a;
}
void test2()
{
auto i = A!().foo(1);
assert(i == 8);
i = Bar.a.foo!(int)(2);
assert(i == 9);
i = Bar.a.foo(3);
assert(i == 10);
}
/*****************************************************/
along with a main() that calls the testNN() functions. The shell script
I run is with a custom shell language
http://www.digitalmars.com/ctg/shell.html
The custom nature of it is its ability to keep rerunning a group of
commands with different combinations of arguments. This is pretty handy
for testing with the combinatorial explosion of compiler switches.
