I forgot to point out that while having my revenge on the
PR116701, I incidentally discovery that
gfortran.fortran-torture contains tests with dg-directives.
Those are ignored. From "git grep dg-":
compile/pr66352.f90:! { dg-additional-options "-fprofile-generate" }
compile/pr85863.f:! { dg-do compile }
compile/pr85863.f:! { dg-additional-options "-ffast-math -ftree-vectorize" }
execute/st_function.f90: if (st5 ("01", "02") .ne. "01 02 ") STOP 5! {
dg-warning "Character length of actual argument shorter" }
execute/st_function_1.f90: call check (foo("Hello"), "Hello Wo") ! {
dg-warning "Character length of actual argument shorter" }
execute/where21.f90:! { dg-do run }
The question is what you'd like to do about that, if
anything. At the least, the directives are confusing and
misleading, not only by design but also their application
here. There's also missing coverage and possibly hidden
bugs. For example, the warnings aren't tested (and not
emitted), the options aren't applied and the profile
generation test isn't exercised as intended (and would need
gating for targets with profiling support).
The mechanism in the non-dg a.k.a "classic testsuite" is
tcl/dejagnu code in files with the suffix .x and the same
basename as the test, so for the first file, that'd be
compile/pr66352.x.
You won't see this in e.g. gcc.c-torture, because that
changed to actually use the dg infrastructure Some Years Ago
and is now just left in the old place: the preferred place
for new "torture" tests is gcc.dg/torture.
brgds, H-P
