[oi-dev] OpenIndiana and illumos, part 2

[Garrett D'Amore]

On Fri, 2010-11-19 at 23:00 +0000, Alasdair Lumsden wrote:
> On 18 Nov 2010, at 20:21, Albert Lee wrote:
> 
> > OK, Garrett's attempt forced people to respond really defensively...
> > so let's try again.
> > 
> > For the record, here are goals (and anti-goals) of illumos that are
> > relevant to OpenIndiana:
> > 
> > - The ability to run software built for Solaris 10 using interfaces
> > that weren't already subject to change between releases (obsolete
> > interfaces need not apply)
> > - The ability to run software built for Solaris 11 on a best-effort
> > basis, since it is both an unknown and a moving target
> > - No consideration for the converse ability to run software built for
> > illumos on Solaris 11 (that is not to say it won't happen to work, but
> > it won't be a concern for development)
> > - Freedom to introduce new features in a way that is incompatible with
> > Solaris 11
> > - Freedom to remove features with known and limited affect (obviously
> > not subverting compatibility, but compromises are inevitable)
> 
> All completely fine - the only point I would like clarification on is "No consideration for the converse ability to run software built for illumos on Solaris 11".
> 
> I'm completely fine with it not being a development concern, but I'm struggling to understand why, eg GNU Nano, compiled on Illumos wouldn't work on Solaris 11. As far as I know nobody has proposed modifying the compiler suites (with Sun Studio we can't, for example) and most 3rd party software hasn't been written to use specific Illumos features (yet). So why would a binary compiled on Illumos look any different to one compiled on Solaris 11?
> 
> I'm definitely *not* asking Illumos to change their stance on this, this is more so that myself (and no doubt others) can fully comprehend the ramifications/future considerations of the statement.
> 
> I think it would be a shame if FOSS software compiled on Illumos/OI didn't run on Solaris 11, but obviously it wouldn't be the end of the world. It would just be useful.
> 
> 

We won't go out of our way to bust this, but we won't limit ourselves
either.  So if we e.g. add a useful function, perhaps from a new POSIX
standard or perhaps from Linux, and the software is compiled to use it
(because autoconf detects it for example), then the resulting binary
might not work on Solaris 11.

	- Garrett