[RFC] type safe trees

[Daniel Jacobowitz]

On Fri, Jun 25, 2004 at 10:06:15AM -0400, Richard Kenner wrote:
>     You're seriously telling me that you'd even consider using such a
>     machine as a develpoment platform?
> 
> For doing development work on the *compiler*, of course not, but it's
> often all you *have* for doing development of the toolchain.
> 
>     Surely everything is going to be cross-compiled until you've got
>     production hardware, a solid toolchain and a good chunk of the system
>     ported.
> 
> And that's my point.  You're going to be a in a difficult environment
> for quite a while.  Getting GCC bootstrapped is one of the best tests of
> the compiler that exists and is an early step.

Maybe it was a convenient early step in the past, but I don't think
that this is a valid reason to preserve the dependence on only C (and
Ada, and C++ and Java if you want target libraries!).  There are plenty
of other early tests available in C; bring up a shell, GNU coreutils,
and so forth.

A cross compilation environment for a hosted system is not a difficult
environment any more.  We've taken most of the voodoo out of it.

>     GCC has very good support for cross compiling. IMHO this is the reason
>     it can be rapidly ported to new targets.
> 
> Right.  And that's what I don't want to damage.

But the support for cross compiling doesn't help your argument at all;
I don't understand what your point is here.

>     Take a look at all the targets supported by gcc. I'd guess half of
>     those simply aren't capable of hosting gcc. 
> 
> Sure.  But I'm talking about those that *are*.  I'm talking about a new
> machines, meant to be marketting as a general development.  Indeed nothing
> I'm saying applies to the embedded environment: that environment isn't
> influenced at all by what language the compiler uses.

I cross-compile to self-hosted systems all the time already.  So do
lots of other people on this list.  It's actually exceedingly
convenient.

-- 
Daniel Jacobowitz