Alvaro Herrera <alvherre@2ndquadrant.com> writes:
> We recently noticed that vacuum buffer counters wraparound in extreme
> cases, with ridiculous results.
Ugh.
> I propose to backpatch this.
+1 for widening these counters, but since they're global variables, -0.2
or so for back-patching. I don't know of any reason that an extension
would be touching these, but I feel like the problem isn't severe enough
to justify taking an ABI-break risk.
Also, %zd is the wrong format code for int64. Recommended practice
these days is to use "%lld" with an explicit cast of the printf argument
to long long (just to be sure). That doesn't work safely before v12,
and if you did insist on back-patching further, you'd need to jump
through hoops to avoid having platform-specific format codes in a
translatable string. (The side-effects for translation seem like
an independent argument against back-patching.)
regards, tom lane