Feedback requested: Support for -XX:AllocateHeapAt on ZFS?

Hi,

The code backing the `-XX:AllocateHeapAt=path` option to java calls the
function posix_fallocate to reserve the space for the file in the file
system. This will fail and return `EINVAL` if the underlying file system
does not support the operation.

On FreeBSD this is particularly relevant because the commonly used ZFS
file system does not support this operation.

For rationale see:
https://freebsd-current.freebsd.narkive.com/5pbDPeIT/heads-up-posix-fallocate-support-removed-from-zfs-lld-affected
https://lists.freebsd.org/pipermail/freebsd-current/2018-February/068447.html


OpenBSD is using the ftruncate function instead, which does not give the
same guarantees as posix_fallocate. Likewise MacOSX will attempt to set
the size via fcntl, but falls back to ftruncate if that does not work.

To make things even more interesting, the glibc implementation of
posix_fallocate used on most Linux based systems deviates from the posix
standard by 'emulating' the feature if not supported by the underlying
file system.
https://www.man7.org/linux/man-pages/man3/posix_fallocate.3.html#CAVEATS

My question is: Do we want to keep the strict posix compliance, which in
practice means this feature is not supported on ZFS, or do we want to be
less strict like the other platforms mentioned above?

With the typical file system sizes of today, I don't think there should
be any big dangers of using a less strict approach, but I'd like to hear
others thought on the subject.

Take care!
Harald