Re: NetBSD/vax compiler bounty?

[Paul Koning <paulkoning%comcast.net@localhost>]

> On Apr 16, 2020, at 2:56 PM, Martin Husemann <martin%duskware.de@localhost> wrote:
> 
> On Thu, Apr 16, 2020 at 12:59:16PM -0400, Paul Koning wrote:
>> Interesting idea.  Is the list of major issues (things that break a
>> kernel or package build or produce wrong code) all that long?  
> 
> We don't really know, and hopefully the list of original causes is
> not that big (once everything is analyzed).
> 
> We only barely know the state of the very hacked version in NetBSD,
> which is able to compile a few simple pkgs from pkgsrc and for
> C++ code totaly breaks exception handling. John is running pkg builds
> and can likely report more details (and this is not about pkgs where
> the source code just does not deal well with VAX floating point).
> 
> Many of the changes between upstream gcc and the NetBSD vax backend are
> not understood well and may be totaly wrong (likely because they have
> been created many versions back and just dragged along through updates).
> 
> This is why I would suggest using the unpatched upstream version as a base,
> and get the test suite running as good as possible.

I agree, that is critical.  For one thing, fixes should be delivered back to the GCC project and that only happens if you deliver a fix to the current code.  For another, I know that some fixes have been made recently in the GCC codebase.  Not to mention some more significant surgery (the new condition code machinery).

	paul