test: Reset global args between test suites #17366

Ideally there wouldn't be any globals in Bitcoin Core. However, as we still have globals, they need to be reset between runs of test cases. One way to do this is to run each suite in a different process. make check does that. However, ./src/test/test_bitcoin when run manually or on appveyor is a single process, where all globals are preserved between test cases.

This leads to hard to debug issues such as #15845#pullrequestreview-310852164.

Fix that by resetting the global arg for each test suite. Note that this wont reset the arg between test cases, as the constructor/destructor is not called for them.

Addendum: This is not a general fix, only for -segwitheight. I don't know if clearing all args can be done with today's argsmanager. Nor do I know if it makes sense. Maybe we want datadir set to a temp path to not risk accidentally corrupting the default data dir?

Thanks to @mzumsande for debugging this

ACK fa07b8beb598642655b1207afd275b801ff8cec2

Fix that by resetting the global arg for each test suite. Note that this wont reset the arg between test cases, as the constructor/destructor is not called for them.

Could clarify in OP or title that this is not a general fix? In other words it applies only to -segwitheight specifically, right? :)

@practicalswift Done

ACK fa07b8beb598642655b1207afd275b801ff8cec2

To catch similar cases going forward: perhaps we could run src/test/test_bitcoin as a single process in one Travis job?

That way we could also catch cases where src/test/test_bitcoin prints error messages which was the case previously.

Let me know if such a Travis job would be accepted: if so I can implement.

That way we could also catch cases where src/test/test_bitcoin prints error messages which was the case previously.

What "error messages" and "previously" are you referring to?

What "error messages" and "previously" are you referring to?

Sorry, context here: #15352, #15944, #16277 :)

ACK fa07b8beb598642655b1207afd275b801ff8cec2, I also tested that this fixes the issue in #15845.

Let me know if such a Travis job would be accepted: if so I can implement.

No opinion from me