Re: Is the ORBit2 stubs/skels breakage unacceptable?

[Michael Meeks <michael ximian com>]

Hi Jeff,

On Sat, 2003-06-28 at 16:45, Jeff Waugh wrote:
> So, I would like to determine whether the stubs/skels breakage in ORBit2
> 0.7.x is acceptable

	As I think the consensus was finally - it should be acceptable. There
are also some quite good reasons to make sure that no tar.gz ships
generated stubs/skels/common - primarily because the new stubs/skels are
smaller, faster and more efficient.

	To make the terminology sufficiently confusing I'll use SAPI/ABI for
stubs/skels API/BI and AAPI/BI for application used API/BI.

	So - we are 100% AA[PB]I backwards compatible. We are also 100% SABI
backwards compatible - however we are not SAPI backwards compatible.
This should not be a problem - since no-one should be shipping generated
stubs/skels [ in fact that can cause some really serious platform
alignment / endianness viciousness with older ORBit2s - so it's a really
stupid idea anyway ].

	So - we could put effort in to make it SAPI backwards compatible - but
this would destroy a large benefit of the new stubs which is to very
substantially reduce the ABI exposure of ORBit2 - which is now extremely
lean and manageable - this being the whole point of the change. It's
also the case that while we are SABI backwards compatible we do that at
some performance cost - so new stubs/skels on a new system will run far
faster than old stubs/skels on a new system.

	Executive summary: every app that ships generated files in the tar.gz
needs fixing since it's broken already on many architectures /
platforms. Also, fixing this will give a brighter, day-glow, faster
future, and we won't have to suffer this pain again.

	HTH,

		Michael.

PS. For those using sufficiently old autotools - you get to do fun
things like this to not distribute things mentioned in BUILT_SOURCES:

DONT_DIST_SOURCE = $(CORBA_SOURCE) $(marshal_sources)

dist-hook:
	for file in $(DONT_DIST_SOURCE) ; do \
	    rm -f $(distdir)/$$file ; \
	done

-- 
 michael@ximian.com  <><, Pseudo Engineer, itinerant idiot