Tom Lane wrote:
> Bruce Momjian <pgman@candle.pha.pa.us> writes:
> > OK, here is a version of newNode that is a macro.
>
> If you use memset() instead of MemSet(), I'm afraid you're going to blow
> off most of the performance gain this was supposed to achieve.
Yep.
> > Does anyone have additional suggestions? The only thing I can suggest
> > is to make a clear-memory version of palloc because palloc always calls
> > MemoryContextAlloc() so I can put it in there. How does that sound?
>
> I do not think palloc should auto-zero memory. Hard to explain why,
> but it just feels like a bad decision. One point is that the MemSet
> has to be inlined or it cannot compile-out the tests on _len. palloc
> can't treat the length as a compile-time constant.
Right, palloc shouldn't. I was thinking of having another version of
palloc that _does_ clear out memory, and calling that from a newNode()
macro. We already know palloc is going to call MemoryContextAlloc, so
we could have a pallocC() that calls a new MemoryContextAllocC() that
would call the underlying memory allocation function, then do the loop
like MemSet to clear it.
It would allow newNode to be a macro and prevent the bloat of having
MemSet appear everywhere newNode appears; it would all be in the new
function MemoryContextAllocC().
> > The regression tests do pass with this patch, so functionally it works
> > fine.
>
> Speed is the issue here, not functionality...
I was just proving it works, that's all.
--
Bruce Momjian | http://candle.pha.pa.us
pgman@candle.pha.pa.us | (610) 359-1001
+ If your life is a hard drive, | 13 Roberts Road
+ Christ can be your backup. | Newtown Square, Pennsylvania 19073