Generating the tests to run in build.rs created problems for tooling. For example, cargo fmt, was ignoring the test_*.rs files and needed to be passed these files manually to be formatted. Now we simply use the feature mechanism to decide which tests to run.
Cstr::from_bytes_with_nul needs input bytes null terminated. Current
version does not include the last null byte, hence
Cstr::from_bytes_with_nul will panic with error 'FromBytesWithNulError {
kind: NotNulTerminated }'
Some tests failed when run using Docker because they assumed the
user would never be root. This is more of a band-aid solution.
An actual fix would be to test see if something like these tests
were to succeed when the user is root.
`test_chmod_ugoa` and `test_chmod_many_options` both change umask, which
is global state. Since tests run concurrently, this might lead to
a situation where one of the tests changes umask to a value that screws
another test's checks. To prevent this, we introduce a mutex that should
be held by any test that changes umask.
Unfortunately, there's no way to hide umask behind this mutex and
enforce its usage: programmers will have to maintain the discipline
themselves.
Add install utility skeleton source, based on
mv, including the getopts setup mirroring
GNU's `man install` documentation. Also
add a single test and build system code.
The main motivation is to move toward running those tests for a specific
target, that is, if a test won't run on Windows, then we shouldn't build
it. This was previously the default behavior and prevented a successful
run on AppVeyor.
I borrowed this pattern from the tests in the Cargo project.
