I am sort-of giving up the dependency injection idea, at least for the first pass, as the end result could be achieved by calling the function that returns the fixture.

In your example I wonder what kind of fixture could be such common feature without making the tests accidentally impacting each other. First thought that $fixture_A must be some resource that is not changed during the tests or that we can be sure that they change it in an independent way. (eg. in the case of a tempdir this could mean one of them creates file1.txt the other one creates file2.txt.)

However this sound a bit risky as the tests might accidentally step on each others toes.

However this way it is the decision of the author of the test.

{ 
   my  $fixture_A = temdir(); # fixture in closure
   
   sub test1  {
       my $fixture_B = get_fixture_b();
       # uses fixture A and B
   }

   sub test2  {
       my $fixture_C = get_fixture_c();
      # uses fixture A and B
   }

}
[download]