[tests] Add basic coverage reporting for RPC tests

[jamesob]

After the missed-unobfuscation snafu of #6777, I realized that we don't have comprehensive coverage of the RPC commands, nor do we have a method of determining which RPC calls are or aren't covered. I'd like to start writing more RPC tests, but first I'd like to have an easy way of figuring out which tests to write.

In this PR I've provided a basic method of getting a binary coverage report for each RPC command, which should be useful to guide adding tests until we have full coverage of the RPC interface. After we're exercising every RPC command, we can possibly gate builds on that full-coverage invariant using a mechanism like this, if that's preferable.

The report looks as follows:

 $ ./qa/pull-tester/rpc-tests.py --coverage

Initialized coverage directory at /tmp/coverageAHqCbm
Running testscript wallet.py...
Initializing test directory /tmp/testSAwNV3

[ tests run... ]

Tests successful
Uncovered RPC commands:
  - dumpprivkey
  - estimatefee
  - estimatepriority
  - getaccount
  - getaddednodeinfo
  - getaddressesbyaccount
  - getblockchaininfo
  - getblockheader
  - getblocktemplate
  - getconnectioncount
  - getdifficulty
  - getgenerate
  - getinfo
  - getmempoolinfo
  - getmininginfo
  - getnettotals
  - getnetworkhashps
  - getrawchangeaddress
  - getreceivedbyaccount
  - getreceivedbyaddress
  - gettxout
  - gettxoutsetinfo
  - getunconfirmedbalance
  - importprivkey
  - keypoolrefill
  - listaccounts
  - listaddressgroupings
  - listlockunspent
  - listreceivedbyaccount
  - listreceivedbyaddress
  - listsinceblock
  - lockunspent
  - move
  - ping
  - prioritisetransaction
  - setaccount
  - setgenerate
  - signmessage
  - submitblock
  - verifychain
  - verifymessage

Cleaning up coverage data

[maflcko]

Is it required to have the coverage tempdir? Couldn't this be done in memory without disk io?

[jamesob]

@MarcoFalke I can't see how that would work given that we call the individual test scripts in separate subprocesses, but agree that would be preferable if possible.

[jamesob]

Yargh... build failure due to #6778. Going to try rerunning.

[laanwj]

Concept ACK, although there seems to be overlap (at least in purpose) with #6813

[jamesob]

Merge conflict resolved here.

[jamesob]

Rebased (thanks again @MarcoFalke).

[jamesob]

@laanwj @dexX7 how are you two feeling about this PR? I still think it'd be useful, but if you have reservations I'm happy to close it out for now.

[dexX7]

I personally have no preference to be honest. I think it could be handy, if new users dive into Bitcoin Core and start with adding additional tests, and I also like the cleanup and documentation. Then again, it's unclear, whether users really are going to use it. Mini-ACK from my side.

[jamesob]

@dexX7 yeah, apologies for conflating some of the cleanup & doc with this feature. Would this PR be more useful if we enabled it by default for travis builds?

[dexX7]

That's a very interesting idea!

[jamesob]

The Travis build now reports coverage by default:

[dexX7]

That's pretty cool imho: https://travis-ci.org/bitcoin/bitcoin/jobs/89011588#L3943

edit: you beat me, hehe. :)

[laanwj]

Instead of changing authproxy, I'd prefer if this was an object that wrapped an authproxy, and logs and passes on all the calls. I think it could be done that way

[jamesob]

Good suggestion. I've started working on the implementation of this, and I think it's going to simplify things a bit.

[laanwj]

The overview of untested RPC commands is confronting. We should definitely have it.

[maflcko]

Nit: Would it be hard to unfiddle into two commits?

• Cleanup
• New feature

Regardless of that: Concept ACK, I think we could make it fail as soon as all commands are covered.

[jamesob]

@MarcoFalke, going forward I'll certainly be more diligent about separating cleanup changes from new functionality -- unfortunately I think in this case it'll be tough to tease out the distinction between cleanup and coverage here, as there isn't a clean way to do it in terms of git chunks, so I'd have to basically rewrite the PR with that separation in mind. This is test code, and so the commit separability doesn't seem to me to be as important as in other parts of the system (though I understand in general this is what we should aim for). That said, if you think it's important I'd be happy to try.

[jamesob]

@laanwj note here that unfortunately I wasn't able to completely avoid changing authproxy.py: in order to use composition over inheritance in coverage.AuthServiceProxyWrapper (which I think is preferable), we have to make this attribute un-name-mangled, i.e. accessible outside of the class itself.

[jamesob]

Aside from any further feedback, I'm happy with where this is.

[jamesob]

Ping here -- any further feedback?

[maflcko]

Nit: is the required? (Makes diff larger)

[laanwj]

In principle I agree with getting rid of cases of import *, but what I'd propose is to explicitly import the symbols that are used:

from util import initialize_chain, start_nodes, # etc

Using the full name explicitly in every use is a) more typing b) generates a large diff (as @MarcoFalke already says)

[jgarzik]

Indeed; explicitly importing symbols reduces namespace pollution and slightly reduces chances for odd bugs.

[jamesob]

I can file a cleanup PR afterwards doing this more broadly; would you guys prefer I (i) revert this diff here, (ii) import the symbols explicitly instead of the util namespace, or (iii) leave as-is for now?

[laanwj]

I'd prefer (ii), but if you want to revert this and do it in another PR it's fine with me too. I don't like (iii) because it implies changing this now and changing this (partially) back again later.

[jamesob]

👍 I'll do (ii) sometime tonight.

[jamesob]

FWIW I'd argue that the existing method of importing the util namespace rather than particular symbols results in less pollution, but this is a small thing.

[maflcko]

utACK 37f4481

[laanwj]

ACK after squashing into one commit

[maflcko]

reACK b5cbd39

[jamesob]

Squashed.