Magnus Hagander <magnus(at)hagander(dot)net> writes:
> Since it didn't really sound like a nice option, here's a third one I
> came up with later. Basically, this one splits things apart so we only
> use one variable, which is sync_method. Instead of using a macro to get
> the open sync bit, it uses a function. This makes the assign hook only
> responsible for flushing and closing the old file.
Okay, but you failed to correctly reproduce the conditions for closing
the old file.
> Thoughts? And if you like it, is it enough to expect the compiler to
> figure out to inline it or should we explicitly inline it?
I don't think we care that much, since it's only invoked when we're
about to do a moderately expensive kernel call.
regards, tom lane