why msync() is not quoted by MMAP_NEED_MSYNC?

[Wim Lewis]

On Friday, January 2, 2004, at 01:05  PM, Katsuhiro Kondou wrote:
> While I installed latest cvs version on my iBook, I found
> innd dies when msync() is called at innd/icd.c.  msync()
> returns EINVAL, but from msync(2), I don't think innd
> misuses msync().  I'm suspecting potential bug in msync()
> on OSX, and have already reported to Apple.

This would not surprise me at all --- there are a bunch of bugs in 
Darwin where the newer BSD VM APIs call into the Mach VM system. A 
while ago I discovered that madvise() was completely broken; from 
looking at the Darwin sources I can only suppose that the programmer 
was very, very confused. It might be worth checking into msync() as 
well...