Random false failures in CI

[bkstein]

I have a test on my branch, which sometimes fails in CI and sometimes not, even if the code didn't change. See the last two actions:

• https://github.com/sfackler/rust-openssl/actions/runs/3427001004
• https://github.com/sfackler/rust-openssl/actions/runs/3427454223

The test never fails locally (that's why I can't debug it), so I assume, there is something unpredictable in the CI setup of this particular job (always windows-vcpkg). The test itself is simple and the error occurs before my new feature is called:

    #[test]
    fn asn1_string_from_asn1_type() {
        let null = null_mut();
        unsafe {
            // Create an ASN.1 type object
            let s = CString::new("PRINTABLESTRING:Hello Test").unwrap();
            cfg_if! {
                if #[cfg(any(ossl110, libressl280))] {
                    let s_ptr = s.as_ptr() as *const _;
                } else {
                    let s_ptr = s.as_ptr() as *mut _;
                }
            }
            let at: Asn1Type = cvt_p(ffi::ASN1_generate_v3(s_ptr, null))
                .map(|p| Asn1Type::from_ptr(p))
                .unwrap();
            assert_eq!(at.as_ref().typ(), Asn1TagValue::PRINTABLESTRING);
            // Get string content from Asn1Type
            let asn1stringref: &Asn1StringRef = Asn1StringRef::from_asn1type(at.as_ref()).unwrap();
            let osslstring: OpensslString = asn1stringref.as_utf8().unwrap();
            let string: &str = osslstring.as_ref();
            assert_eq!("Hello Test", string);
        }
    }

and it fails in OpenSSL which pretends not to know the tag PRINTABLESTRING:

---- asn1::tests::asn1_string_from_asn1_type stdout ----
thread 'asn1::tests::asn1_string_from_asn1_type' panicked at 'called `Result::unwrap()` on an `Err` value:
ErrorStack([
    Error { 
        code: 109052098,
        library: "asn1 encoding routines",
        function: "asn1_cb",
        reason: "unknown tag",
        file: "crypto\\asn1\\asn1_gen.c",
        line: 266,
        data: "tag=PRINTABLESTRING:Hello Test"
    },
    Error {
        code: 109052098,
        library: "asn1 encoding routines",
        function: "ASN1_generate_v3",
        reason: "unknown tag",
        file: "crypto\\asn1\\asn1_gen.c",
        line: 94
    }
])', openssl\src\asn1.rs:1068:18
...
failures:
    asn1::tests::asn1_string_from_asn1_type

[sfackler]

It seems like it is definitely a real failure to me. OpenSSL has previously had race conditions in its internal state initialization that could cause errors like this during concurrent calls.

[bkstein]

A race condition, yes. That's possible. I was not aware of cargo test executing multiple threads by default. Should we restrict testing to a single thread?

$ cargo test -- --test-threads=1

[sfackler]

The general expectation is that these APIs should be thread safe. It seems inadvisable to test exclusively in a single threaded context.