There are many constraint errors specified by the standard, which require us to print a diagnostic.
We should extend runtests to support tests to check these errors, then start adding test C files that violate constraints.
Perhaps this could be done by adding a .err file instead of a .qbe file, and the test runner would check that cc-qbe failed and stderr matched the contents of the .err file.
Should these tests go alongside the others, or in a separate directory?
By coincidence, I have just been working on this, I have a working prototype.
Here is an example test case:
/* PATTERN: incomplete PATTERN: c:12 */ int main() { { struct X {}; } struct X x; return 0; }
I have improved on my idea quite a lot and it seems to work well here are some examples...
/* t_output="hello world!" */ extern int printf(const char *, ...); int main() { printf("hello world!"); }/* t_rc=2 t_output="hi!" */ extern int printf(const char *, ...); int main() { printf("hi!"); return 2; }/* t_ccrc=1 t_ccouput () { exec grep "error" "$1" } */ int main() { retu 2; }The trick is just extracting sh hooks and evaling them, then you can do non trivial logic in your tests.
I have a parallel test running using redo, and it fall's back to a serial runner if redo is not installed.