[plt-dev] Release Announcement for v4.2.3, third call

From: Matthew Flatt (mflatt at cs.utah.edu)
Date: Fri Dec 4 16:47:56 EST 2009

At Mon, 30 Nov 2009 21:48:45 -0500, Sam TH wrote:
> On Mon, Nov 30, 2009 at 9:34 PM, Matthew Flatt <mflatt at cs.utah.edu> wrote:
> > At Mon, 30 Nov 2009 21:29:07 -0500, Eli Barzilay wrote:
> >> [Matthew]
> >> * scheme/unsafe/ops? (wasn't mentioned last time)
> >
> > I don't think any of these need to be in the release announcement.
> 
> It seems like the unsafe operations are now a significant part of the
> performance model, but they've never been announced more widely than
> the mailing list.  Are we waiting for something?

I think we're waiting until we've figured out how to use it well enough
or packaged it better for a wide audience.

The problem with local bindings that Will Farr pointed out is a good
example of where support for unsafe operations needs improvement.
Another good example is an experiment that I committed yesterday (which
I know that you noticed): there's now an `flvector' type that avoids
the two extra indirections of an `f64vector' in unsafe mode. The
`flvector' operations are also inlined better in safe mode. But
`flvector' works less well for FFI interactions, the safe operations
are typically less efficient than safe operations on plain vectors, and
even the unsafe operations offer a benefit only when combined with
other operations to avoid unboxing.



Posted on the dev mailing list.