Just happened again on a non-experimental builder:
https://internal.askmonty.org/buildbot/builders/kvm-bintar-quantal-x86/builds/7171/steps/test/logs/stdio

rpl.rpl_checksum_cache 'innodb_plugin,row' w1 [ fail ]  Found warnings/errors in server log file!

        Test ended at 2017-05-17 19:19:31

line

InnoDB: Error: trying to do an operation on a dropped tablespace 6

^ Found warnings in /usr/local/mariadb-10.0.31-linux-i686/mysql-test/var/1/log/mysqld.2.err

ok

 - saving '/usr/local/mariadb-10.0.31-linux-i686/mysql-test/var/1/log/rpl.rpl_checksum_cache-innodb_plugin,row/' to '/usr/local/mariadb-10.0.31-linux-i686/mysql-test/var/log/rpl.rpl_checksum_cache-innodb_plugin,row/'

I suspect that this must have something to do with the previous tests that were run on the same instance. The logs are no longer available for either failure, and I was unable to repeat the problem locally by running all 6 combinations of the test 10 times each.

I think that we can simply remove this message. The only issuers of it are the debug function lock_rec_block_validate() and the change buffer merge. While in the former function the error should ideally never occur, we happen to know that DISCARD TABLESPACE and TRUNCATE TABLE and possibly DROP TABLE are breaking locks inside InnoDB.

When it comes to the change buffer merge, the message simply is useless noise. We know perfectly well that a tablespace can be dropped while a change buffer merge is pending.. And the code is prepared to handle that, which is demonstrated by the fact that whenever the message was issued, InnoDB did not crash.

In MariaDB 10.2, the function is fil_space_acquire() or fil_space_acquire_silent() instead of fil_inc_pending_ops(). Also there, we should suppress the message during change buffer merge, but not necessarily in other places.

Starting with MariaDB 10.2.2, the message is a warning: Trying to access missing tablespace.
It could be issued in more places than the change buffer merge. I will only suppress the message in the change buffer merge.