OK, but why is it that ASDF's tests are vulnerable to oddities in the state of my working directories (as opposed to, e.g., only the state of the directories created by the test scripts?). And why is it that this manifests itself only when running inside make?
I don't know why it happens only in one case and not the other. How do you invoke the test, when not by Make?
And yes, I believe that ABCL gets confused when a real file has a wildcard character — which is indeed forbidden by the CLHS, if unescaped. ABCL needs to somehow escape wildcard characters.
OK, I suspect that the reason that these were failing for me with make, and not with run-tests is that the working directory for run-tests is asdf/test/ -- not a super-directory of doc/ -- whereas for make the working directory is asdf/, which is a superdirectory of doc.
If that's the case, then this represents us simply stubbing our toes on an ABCL pathname-handling bug.
This test (and test-utilities, see https:/ /bugs.launchpad .net/asdf/ +bug/1307569) are suffering some kind of state bleed-over through make.
When I run
make test l=abcl
test-encodings and test-utilities both fail.
When I do
run-tests abcl test-encodings. script
or the same for test-utilities, both tests pass. This suggests to me some oddity in the makefile.
run-tests abcl
also passes all tests, which tends to support my conjecture that the problem is with the makefile.