Run tests in memory sanitizer #17460

We run the tests in the thread (and other sanitizers), see git grep sanitizer ./ci, but not in the memory sanitizer. This seems like it could be improved. See e.g. #17449

On a second thought, it might be better to just run with valgrind: https://github.com/bitcoin-core/secp256k1/pull/687#issuecomment-551794708

@MarcoFalke I've tried to add Valgrind to our Travis build before: Valgrind is very good but also very very very slow. Running the unit tests and the functional tests under Valgrind takes hours(!) :(

Perhaps DynamoRIO's Dr. Memory could be a faster option. I haven't investigated that option yet.

Another alternative could be to run a subset of the tests under Valgrind and then use MSAN as a complement. With MSAN running all the tests including those excluded from the Valgrind run due to being too slow.

Maybe running a single functional test (and/or unit test) in valgrind would be a good start. #17449 would have been caught when Bitcoin Core just boots up on mainnet.

Maybe running a single functional test (and/or unit test) in valgrind would be a good start.

Agreed: to start with something under valgrind would be better nothing :)

#17449 would have been caught when Bitcoin Core just boots up on mainnet.

Do you mean "boots up on mainnet" as in when IBD is finished on mainnet?

FWIW, I've only gotten valgrind to warn about the #17449 case when WarningBitsConditionChecker::Condition(…) is being run. That happens in UpdateTip given that we're done with IBD.

Perhaps one would expect valgrind to warn earlier in the execution, such as when {CreateChain,Select}Params(…) is run. But that is not the case :)

#17568 (comment)