Re: libjava/java/io/natFileDescriptorPosix.cc -r1.25 -r1.26

["Andreas Tobler" <toa at pop dot agri dot ch>]

>-- Original Message --
>> I have this one for darwin, it may also be valid for *BSD?
>> Not reg tested yet.
>
>(Here is the other half, that Andreas and now I'd have in mind.)
>
>Thanks Andreas, how about more generally [so that we need not update
>it for every system as we (a) find them; (b) they begin to support the
>new finer grain flags.  I suspect that both FreeBSD and Darwin system
>headers will support those newer POSIX flags at some point.
>E.g. there is an explicit comment on the matter in FreeBSD 5's header.]

Well, fine with me, I just thought to publish what I found so far.
I couldn't complete testing on this issue. Since I do some other testing
for rth. And, beside I should also follow my day job :)

>In the alternative (i.e. if people think this is too gross) or perhaps
>in addition to this patch (if there is a system that fail to provide
>O_FSYNC as well), the point of use needs to be wrapped and an
>exception thrown to signal the lack of real support.  BTW, I think I
>agree that this is a correct mapping in terms of what O_FSYNC implies
>when present on a system.  I can confirm the version as posted
>completes bootstrap.

I read about SYNC and FSYNC and found that it should match our needs.
FSYNC, afaiu, does a sync on our request versus SYNC which does it on periodic
system interval.

Thanks,
Andreas