This week's CI of GNU libunistring has a new test failure, not present last week, on native Windows (mingw, MSVC):
FAIL: test-openat ================= ../../tests/test-openat.c:94: assertion 'openat (dfd, ".", O_RDONLY) == STDIN_FILENO' failed FAIL test-openat.exe (exit status: 3) The only suspicious commit in this area is this one: > 2024-11-06 Paul Eggert <egg...@cs.ucla.edu> > > + save-cwd: don’t depend on fd-safer-flag, unistd-safer > + They aren’t needed in packages that use xstdopen > + * lib/save-cwd.c: Do not include unistd--.h. > + (GNULIB_FCNTL_SAFER): Do not define. > + (save_cwd): Do not use fd_safer_flag, as the package should use > + either fcntl-safer+unistd-safer or xstdopen if it cares about > + closed stdin. > + * modules/save-cwd (Depends-on): Remove fd-safer-flag, unistd-safer. The test, in test-openat.c:94, is reasonable IMO: It verifies that the openat() replacement for native Windows behaves the same as a POSIX-compliant openat(). libunistring does not use fcntl-safer, unistd-safer, xstdopen, but it includes the unit test since it uses openat() among the test dependencies. What do you think? Should some module dependency be added to the 'openat' module? Or to the 'save-cwd' module? Bruno
