Created attachment 1112210 [details]
rpm db4 messages should go to stderr

Description of problem:

Some rpm pre/post scripts rely on a shell out to rpm to determine pre-existing installed package information/versioning. (e.g. `rpm -q --provides mysql-server 2> /dev/null`)

Unfortunately some non-fatal warnings can be output if, for example, a read lock on the rpmdb is held open at this time, leading to the warning picked up as the output and causing breakage:

Freeing read locks for locker 0x49e: 482061/47810768245328 
mysql-server-5.1.73-5.el6_6.x86_64

(both lines on stdout)

Putting aside the question of what is holding the read lock open for a second, this behaviour is obviously erroneous -- warnings/errors should be output to stderr, not stdout. 

How reproducible:

`rpm -q --provides mysql-server 2> /dev/null` 

While read lock held open on rpmdb. Cannot provide a reproducer for this as unfortunately is enivornment dependent.

Actual results:

Freeing read locks for locker 0x49e: 482061/47810768245328 
mysql-server-5.1.73-5.el6_6.x86_64

Expected results:

Freeing read locks for locker 0x49e: 482061/47810768245328 (on stderr)
mysql-server-5.1.73-5.el6_6.x86_64 (on stdout)

Additional info:

The behaviour is caused by db4, which if no msgfile is specified, defaults to stdout. There is a simple fix, however, which without modifying default db4 behaviour that merely involves rpm passing stderr as the msgfile at initialisation (patch supplied).