PMMR false positive test?

[garyyu]

On a Travis-Ci test log, I find this:

     Running /home/travis/build/garyyu/grin/target/release/deps/pmmr-9d03adec459ef515
running 10 tests
test cleanup_rewind_files_test ... ok
test pmmr_append ... ok
test compact_twice ... ok
test pmmr_compact_entire_peak ... ok
test pmmr_compact_horizon ... ok
test pmmr_compact_leaf_sibling ... ok
test pmmr_compact_single_leaves ... ok
test pmmr_prune_compact ... ok
test pmmr_reload ... ok
ERROR 2019-03-11T05:45:41Z: grin_store::types: Corrupted storage, could not read an entry from data file: IOErr("failed to fill whole buffer", UnexpectedEof)
ERROR 2019-03-11T05:45:41Z: grin_store::types: Corrupted storage, could not read an entry from data file: IOErr("failed to fill whole buffer", UnexpectedEof)
ERROR 2019-03-11T05:45:41Z: grin_store::types: Corrupted storage, could not read an entry from data file: IOErr("failed to fill whole buffer", UnexpectedEof)
ERROR 2019-03-11T05:45:41Z: grin_store::types: Corrupted storage, could not read an entry from data file: IOErr("failed to fill whole buffer", UnexpectedEof)
ERROR 2019-03-11T05:45:41Z: grin_store::types: Corrupted storage, could not read an entry from data file: IOErr("failed to fill whole buffer", UnexpectedEof)
ERROR 2019-03-11T05:45:41Z: grin_store::types: Corrupted storage, could not read an entry from data file: IOErr("failed to fill whole buffer", UnexpectedEof)
ERROR 2019-03-11T05:45:41Z: grin_store::types: Corrupted storage, could not read an entry from data file: IOErr("failed to fill whole buffer", UnexpectedEof)
ERROR 2019-03-11T05:45:41Z: grin_store::types: Corrupted storage, could not read an entry from data file: IOErr("failed to fill whole buffer", UnexpectedEof)
test pmmr_rewind ... ok
test result: ok. 10 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out

Is this Corrupted storage a false positive test bug? or the designed behavior in this test?

[garyyu]

@antiochp Would you like to take a look? thanks.

[antiochp]

We have something in the code that's attempting to read past the end of a rewound MMR.
i.e. We rewind to pos 3 (for example) and then in the tests we read pos 4.

I'm pretty sure we did not log those ERROR level messages before, but something changed in past few weeks.
We see these during "fast sync" on mainnet/floonet as well.

If we read a non-existent pos from the MMR we return None so these are harmless.

I think @yeastplume was investigating this a bit as well (I suspect we want to simply quieten this logging as its not really an issue).

[garyyu]

This issue can be close by #2685

[0xmichalis]

@garyyu you can close this issue yourself I guess :)

[antiochp]

Closing!