Travis-CI memory reference errors in CLI Tests

[MichelSantos]

Travis-CI has detected some memory reference problem in the CLI wallet integration tests. These errors cannot always be triggered.

• Some Travis results indicate that the memory reference problem is connected with the construction of the fixture.

• Other Travis results also indicate that the memory reference problem is connected with delays after app1->shutdown in the preceding unit test.

• The memory reference errors can be imitated outside of Travis by the inclusion of the following code (which is patently ridiculous by intentionally breaking the functionality) at the beginning of the cli_fixture constructor.

std::cout << "CLI_Fixture A" << std::endl;

app1->shutdown();
fc::usleep(fc::milliseconds(1000));

std::cout << "CLI_Fixture B" << std::endl;

These combined test results suggest a timing problem where the conclusion of one unit test is interfering with the startup of the next unit test as in cli_multisig_transaction-saving_keys_wallet_test, and cli_create_htlc-cli_sign_message.

[abitmore]

Duplicate of #1303 ? Also see #1626, #1856 and #1784. If it's the same issue or similar, it's has been there for quite some time, which is annoying, but imo it's not blocking anything. I'd say leave it there now. I had a plan to fix it after merged bitshares/bitshares-fc#134 (but that one is a bit stale).

[MichelSantos]

Yes the nature of the same problem is the same as exposed by #1303 which is a problem of one instance of an object from a prior test interfering, through an as yet unconfirmed mechanism (although I have my guesses), with another instance in the subsequent test. The fix in #1626 attempts to fix it by allowing the environment sufficient time to shut down the prior instance of the cli_fixture members.

However the delaying action from #1626 is absent when the prior test creates an graphene::app::application directly rather than a cli_fixture. This causes a problem for two pairs of test sequences in the suite: cli_multisig_transaction-saving_keys_wallet_test, and cli_create_htlc-cli_sign_message. In both cases, the prior test directly creates an instance of graphene::app::application and the subsequent test indirectly creates an instance of graphene::app::application by instantiating a cli_fixture.

I have no objection to leaving out this PR for now because I have high confidence at this point that the problem is in the test suite and not in the underlying objects in Core. Please advise if you recommend to close this PR #2068 or to PR into another branch.

[abitmore]

I didn't know you created another app object outside of cli_fixture. Approved #2068.

[abitmore]

The underlying reason is described in OP of #1784.

[pmconrad]

Hopefully fixed by #2068. Failure of merge build was due to timeout in sonar scanner.