Ceph

For what it's worth, it's not a blocker for me. The "plain" format is documented to not being supported and it only happens if you have access to the admin socket. In other words, I was looking for troubles and stumbled on what appears to be a bug.

Line numbers are relative to b9a127e

src/client/Client.cc:113 is not safe
    src/osdc/Objecter.cc:2509 is not safe

src/osd/OSD.cc:912 is not safe because dump_ops_in_flight does not test if f is null

src/mon/AuthMonitor.cc contains > 10 if to deal with the fact that f is null or not
    src/mon/MDSMonitor.cc contains < 10 if to deal with the fact that f is null or not
    src/mon/Monitor.cc contains > 10 if to deal with the fact that f is null or not
    src/mon/MonmapMonitor.cc contains < 10 if to deal with the fact that f is null or not
    src/mon/OSDMonitor.cc:2195 and src/mon/OSDMonitor.cc:2218 have an unsafe call where there is no risk to reach it http://tracker.ceph.com/issues/7388
    src/mon/Monitor.cc contains > 10 if to deal with the fact that f is null or not
    src/mon/PGMonitor.cc contains < 10 if to deal with the fact that f is null or not
    src/osd/OSD.cc contains < 10 if to deal with the fact that f is null or not
    src/osd/ReplicatedPG.cc contains < 10 if to deal with the fact that f is null or not
    src/test/bench/small_io_bench_fs.cc src/test/common/get_command_descriptions.cc src/test/common/test_sloppy_crc_map.cc and src/test/crush/indep.cc use hard-coded arguments

There is a total ol 44 call to new_formatter found in master but it appears that there are various strategies to deal with the return of a null formatter, a number of which are custom. For instance osd dump will print() if the formatter is null while osd crush rule list will fall back to json_pretty.

Adding an argument to new_formatter would require to go over all calls to new_formatter, add NULL where it appears to handle the case where it returns null and add the default where it does not. Another solution is to modify the three calls that are unsafe.