Re: Status and future of 3rd party ABI compatibility layer

[Masao Uebayashi <uebayasi%gmail.com@localhost>]

On Sat, Feb 26, 2011 at 6:35 AM, Joerg Sonnenberger
<joerg%britannica.bec.de@localhost> wrote:
> Hi all,
> there is a lot of code in sys/compat and changes in the kernel API tend
> to require changes in this code too. I would like to know which
> emulations are actually in use, what the status of emulation is (both in
> terms of stability and feature set) and based on that, whether some of
> them shouldn't rest in the Attic.
>
> As far as I can tell we have emulation code for:
>
> Darwin (no GUI, doesn't to have been updated in the last 5 years)
>
> FreeBSD (does it even handle FreeBSD 4?)
>
> IRIX
>
> Linux
>
> OSF1
>
> SunOS/Solaris
>
> SVR3
>
> SVR4
>
> Ultrix
>
> Windows/PECOFF (hasn't this been considered dead quite a while ago?)
>
> I don't think we need to discuss keeping Linux. I also don't count NDIS
> or NetBSD32 in this area. For the rest, I would find the following
> information highly useful to evaluate the maintainance cost and ensure
> future support:
>
> (1) Can the emulation run real world application and which?
>
> (2) If the emulation is considered experimental only:
> What progress has been made in the recent time frame (e.g. this decade)
> on getting it to a useful and stable state? Is this likely to change?

I'd want to hear this too from the project leaders.

>
> (3) How can this emulation be tested? Where can reference binaries be
> obtained? Has someone done the work to integrate them with ATF or other
> scripted test frame works?

We should definitely have a set of test binaries.

One idea (as I've written somewhere) is to have a collection of
syscall stubs and a glue language (e.g. Lua) for each emulation.  This
way all emulated syscalls can be tested by writing scripts, no need to
recompilation.

If used with emulators, no need of real machines too.

>
> (4) Is the emulation in use?
>
> Joerg
>