Re: backwards compatibility/cross-compiler compatibility

[Benjamin Scherrey <scherrey@switchco.com>]

I'm shocked and extremely happy that compiler vendors are talking
about a standard ABI for *anything*! I was very disappointed when the
ISO standards group decided not to take up the issue. I know that a
standard ABI for all platforms is not viable (or even useful) but at
least those operating on the same platform should have an agreement or
the option of a "compatibility mode" for people who need to write such
code. The most obvious issues are name mangeling of symbols, vtables,
and calling conventions. I understand that backwards compatibility is
not a reasonable option but once a standard ABI convention is
established and given a little time to mature, why would
"release-to-release" compatibility be so problematic? Also, is there a
spot on the web where we can watch this effort or are the vendors
being closed doors on this?

	thanx & later,

		Ben Scherrey

Jason Merrill wrote:
> 
> FWIW, a number of compiler vendors (including Cygnus) have started meeting
> recently to discuss the possibility of a common ABI for ia64 toolchains.
> At the first meeting, Mike Ball of Sun asserted that "release-to-release
> compatibility is a chimera".  Apparently they and SGI both have tried to
> set up schemes for backwards compatibility of their libraries, but have
> since abandoned them as unworkable.
> 
> The problem is that for STL, or any other template library, the ABI is
> basically the whole library.  This could change with support for separate
> compilation, but that is the current state of affairs.  As a result, I
> don't think that trying to maintain compatibility between versions of STL
> is of a high priority.
> 
> The ABI group is focusing on compatibility between compilers for code
> written for the freestanding library before we try to tackle the full
> library.
> 
> Jason