Sometimes, cargo test fails on Mac OSX about tox

[NamsooCho]

running 542 tests
test toxcore::crypto_core::tests::crypto_init_test ... ok
alice_pk = PublicKey([127, 71, 215, 228, 18, 40, 100, 184, 186, 158, 178, 170, 102, 5, 177, 21, 105, 82, 207, 123, 8, 6, 230, 142, 123, 192, 140, 27, 108, 197, 150, 123]),
 alice_sk = [238, 115, 125, 75, 116, 121, 183, 189, 118, 66, 194, 5, 247, 157, 36, 177, 136, 141, 224, 72, 69, 103, 117, 36, 84, 108, 253, 58, 162, 135, 104, 162],
 bob_pk = PublicKey([137, 7, 228, 12, 185, 93, 189, 33, 161, 239, 221, 155, 122, 211, 161, 135, 7, 189, 188, 94, 101, 145, 66, 221, 174, 235, 183, 127, 82, 234, 248, 116]),
 bob_sk = [181, 105, 85, 80, 187, 92, 227, 190, 237, 114, 222, 133, 222, 192, 84, 207, 231, 91, 161, 187, 178, 130, 137, 241, 101, 230, 166, 6, 62, 208, 61, 251],
 nonce = Nonce([219, 2, 66, 226, 29, 162, 129, 210, 141, 14, 132, 203, 191, 114, 32, 190, 185, 97, 184, 82, 143, 24, 237, 45])
test toxcore::crypto_core::tests::increment_nonce_number_test_zero_plus_0xff00 ... ok
test toxcore::crypto_core::tests::increment_nonce_number_test_0xff0000_plus_0x011000 ... ok
thread 'toxcore::crypto_core::tests::encrypt_precompute_test' panicked at 'called `Result::unwrap()` on an `Err` value: ()', libcore/result.rs:1009:5

Opps! It succeeded. Refer below comment

[kurnevsky]

Is it reproducible? If so can you catch generated keys and nonce when this test fails?

[kpp]

Nice catch!

[NamsooCho]

running 542 tests
encrypt_precompute_test
 alice_pk = PublicKey([130, 229, 192, 148, 62, 201, 246, 193, 93, 224, 30, 219, 87, 87, 2, 12, 62, 30, 48, 232, 117, 82, 8, 130, 163, 193, 49, 245, 218, 6, 96, 46]),
 alice_sk = [171, 208, 8, 222, 230, 181, 77, 126, 123, 252, 115, 199, 62, 25, 118, 30, 226, 141, 140, 96, 9, 44, 222, 35, 68, 140, 215, 130, 170, 114, 59, 150],
 bob_pk = PublicKey([42, 17, 171, 186, 151, 85, 130, 197, 12, 103, 66, 216, 205, 54, 68, 65, 135, 25, 87, 132, 39, 179, 211, 13, 181, 207, 212, 85, 14, 83, 122, 102]),
 bob_sk = [97, 30, 222, 95, 222, 225, 16, 217, 220, 125, 253, 76, 81, 148, 83, 82, 23, 123, 180, 120, 176, 48, 1, 6, 96, 5, 61, 147, 124, 11, 78, 209],
 nonce = Nonce([243, 128, 244, 180, 215, 104, 117, 98, 88, 250, 155, 217, 110, 165, 210, 188, 185, 18, 138, 238, 59, 80, 207, 211])

This is captured correctly.

[NamsooCho]

encrypt_precompute_test failure alternates with encrypt_data_symmetric_test failure

[kpp]

    #[test]
    // test uses "bare" functions provided by `sodiumoxide`, with an exception
    // of the tested function
    fn encrypt_precompute_test() {
        let alice_pk = PublicKey([130, 229, 192, 148, 62, 201, 246, 193, 93, 224, 30, 219, 87, 87, 2, 12, 62, 30, 48, 232, 117, 82, 8, 130, 163, 193, 49, 245, 218, 6, 96, 46]);
        let alice_sk = SecretKey([171, 208, 8, 222, 230, 181, 77, 126, 123, 252, 115, 199, 62, 25, 118, 30, 226, 141, 140, 96, 9, 44, 222, 35, 68, 140, 215, 130, 170, 114, 59, 150]);

        let bob_pk = PublicKey([42, 17, 171, 186, 151, 85, 130, 197, 12, 103, 66, 216, 205, 54, 68, 65, 135, 25, 87, 132, 39, 179, 211, 13, 181, 207, 212, 85, 14, 83, 122, 102]);
        let bob_sk = SecretKey([97, 30, 222, 95, 222, 225, 16, 217, 220, 125, 253, 76, 81, 148, 83, 82, 23, 123, 180, 120, 176, 48, 1, 6, 96, 5, 61, 147, 124, 11, 78, 209]);

        let alice_plaintext = b"Hi, Bob.";
        let alice_precomputed_key = encrypt_precompute(&bob_pk, &alice_sk);

        let nonce = Nonce([243, 128, 244, 180, 215, 104, 117, 98, 88, 250, 155, 217, 110, 165, 210, 188, 185, 18, 138, 238, 59, 80, 207, 211]);

        let ciphertext = seal_precomputed(alice_plaintext, &nonce, &alice_precomputed_key);

        let bob_precomputed_key = encrypt_precompute(&alice_pk, &bob_sk);
        let bob_plaintext = open_precomputed(&ciphertext, &nonce, &bob_precomputed_key).unwrap();

        assert_eq!(alice_plaintext, &bob_plaintext[..]);
    }

$ cargo test encrypt_precompute_test
   Compiling tox v0.0.7 (/home/humbug/tox)
    Finished dev [unoptimized + debuginfo] target(s) in 8.56s
     Running target/debug/deps/tox-d08d73a27a9a5786

running 1 test
test toxcore::crypto_core::tests::encrypt_precompute_test ... ok

[kurnevsky]

Perhaps it's mac-specific? @NamsooCho can you try this test that @kpp posted? Does it fail?

[kpp]

Do you use system libsodium(with SODIUM_LIB_DIR)?

[NamsooCho]

Do you use system libsodium(with SODIUM_LIB_DIR)?

Yes

[NamsooCho]

Perhaps it's mac-specific? @NamsooCho can you try this test that @kpp posted? Does it fail?

Yes. I will do it tomorrow.

[NamsooCho]

The test that @kpp posted always success on Mac.
But irregularly cargo test still fails.

failures:

---- toxcore::crypto_core::tests::encrypt_data_symmetric_test stdout ----
thread 'toxcore::crypto_core::tests::encrypt_data_symmetric_test' panicked at 'called `Result::unwrap()` on an `Err` value: ()', libcore/result.rs:1009:5
note: Run with `RUST_BACKTRACE=1` for a backtrace.


failures:
    toxcore::crypto_core::tests::encrypt_data_symmetric_test

test result: FAILED. 541 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out

[kpp]

@NamsooCho try to put sodium::init() in the beginning of the test

[NamsooCho]

Putting crypto_init() solves this problem.

[NamsooCho]

When making it like this

pub fn crypto_init() -> bool {
//    ::sodiumoxide::init().is_ok()
    true
}

never fails on cargo test

[NamsooCho]

Maybe It is because, one thread do job of sodium, other thread call crypto_init().

[kpp]

never fails on cargo test

Did you remove libsodium from brew and did you build tox with static && newest libsodium?

[NamsooCho]

Did you remove libsodium from brew and did you build tox with static && newest libsodium?

Yes

[kpp]

Maybe that's the reason of the fix?

[NamsooCho]

No. even after removing libsodium from brew, it fails irregularly.

[kpp]

But you said:

When making it like this .... never fails on cargo test

[NamsooCho]

Yes. after making crypto_init() to blank, it always success.

[kpp]

Wow... does it fail with cargo test -- --test-threads=1 with non-blank crypto_init?

[NamsooCho]

Test with cargo test -- --test-threads=1 with non-blank crypto_init never fail.

[kpp]

What is the frequency of failing with cargo test -- --test-threads=4 with non-blank crypto_init?

[NamsooCho]

Run 9 times, 1 fails.

[kpp]

How many times did you run cargo test -- --test-threads=1?

[NamsooCho]

about 10 times.

[kpp]

Run 30 times more =)

[NamsooCho]

40 times all success

[kpp]

From https://libsodium.gitbook.io/doc/internals :

Thread safety

Initializing the random number generator is the only operation that requires an internal lock.

sodium_init() should be called before any other functions. It picks the best implementations for the current platform, initializes the random number generator and generates the canary for guarded heap allocations.

On POSIX systems, everything in libsodium is guaranteed to always be thread-safe.

[kpp]

Maybe that's why tests fail?

[NamsooCho]

Yes. I guess.

[kpp]

Good catch. I will try to discover how to fix it. One solution is to put crypto_init in every test fn =(

[kpp]

Depends on rust-lang/rfcs#1664