Re: -Wformat-zero-length

On tor, 2011-07-07 at 18:09 -0400, Tom Lane wrote:
> I wrote:
> > Peter Eisentraut <peter_e@gmx.net> writes:
> >> I was adding gcc printf attributes to more functions in obscure places,
> >> and now I'm seeing this in pg_upgrade:
> 
> >> relfilenode.c:72:2: warning: zero-length gnu_printf format string [-Wformat-zero-length]
> 
> > Shouldn't it be prep_status("\n")?  If not, why not?
> 
> On closer inspection, it appears to me that prep_status should never be
> called with a string containing a newline, period, and the test it
> contains for that case is just brain damage.  The only reason to call it
> at all is to produce a line like
> 
>     message ......................
> 
> where something more is expected to be added to the line later.  Calls
> that are meant to produce a complete line could go directly to pg_log.
> 
> This in turn implies that transfer_all_new_dbs's use of the function is
> broken and needs to be rethought.

I think this got a bit besides the point.  There is probably some
bogosity in the logging implementation, but I think the prep_status("")
call is correct and purposeful at that point, namely to clear the line.
The question is, do we consider empty format strings a bug worth warning
about, or should we shut off the warning?