Sigh, I'm an idiot. I forgot that USE_REPL_SNPRINTF doesn't just
replace snprintf, it replaces the entire *printf family; see
port.h lines 137ff. So actually we're OK as far as these %z and
argument-reordering concerns go. Maybe the comments in configure
could use a bit of work though.
There's maybe also an argument for reverting b929614f5, because
actually that code did do something useful, ie allow us to work on
platforms without %ll. But I'm inclined to leave that alone;
it's an extra configure test to detect a case that probably no longer
occurs in the wild. Moreover, since %ll and %z are both C99-isms,
and the former had considerable currency even before C99 (evidence:
gaur/pademelon) it's pretty hard to credit that a platform's *printf
would fail the %ll test yet pass the %z test. So I think we're
likely OK without it, even on dinosaur platforms.
regards, tom lane