RE: [Axiom-developer] RE: cross-compiling Axiom

[Gabriel Dos Reis]

On Tue, 5 Sep 2006, C Y wrote:

| Just so I can see if I'm successfully following this or not, can I try
| and list a few points and see if I am understanding them correctly?
|
| We want to use gcc and the GNU tools to create a binary targeted for HP
| Unix Debian Linux on a Sparc, or Windows on an x86 Linux host, that can
| be copied from Linux to the target system with ftp or a USB stick and
| run with no further compiling.  The target system is in no way required
| for anything except running the final binary. Is this what Linux
| distributors want to be able to do?

Yes.

| And is the issue that GCL cannot
| currently build using the non-host-system binutils and include files
| that are required for this to work?  Or will Axiom also have trouble
| even if we get GCL itself to do this?

If we can get GCL do it, we most certainly can have Axiom move there.

| Things We Can Do:
|
| 1.  Camm and the Debian team can build GCL and Axiom on a wide variety
| of target platforms, but it is not clear if they are doing this using
| the "normal" way Linux distributions want to or have some special
| provisions like having at least one machine of every target type
| available.  Based on Gaby's comments I take it the Debian solution is
| not general enough to meet everyone's needs, however they are doing it.

I'm not familiar enough with the way the Debian people do it.  My
discussion with Benjamin was quite extensive and from what I
understood, we need to do a bit more and then we can take on.
But, I suspect that the linux distros are familiar with the way debian
people do it. Apart from that and other reasons I alluded to I don't
know more about why they do it the way they prefer to do it.

| 2.  We can build on Linux and sometimes Windows using existing tools on
| those platforms, but we cannot build a Linux binary on Windows or a
| Windows binary on Linux.  (Correct?)

yes.

| Moving to other platforms, our options are:
|
| 1.  Produce a working GCL for a platform, then build Axiom on that
| platform using the GCL made for it
|
| 2.  Produce a working GCL and a working Axiom binary for a platform,
| with no compiling work done on the target system.
|
| Is that the distinction here Gaby?

yes.  More precisely, for #2 we would need a simulator for the target
to accomodate GCL (for saving image).

| Doing 1 is NOT cross-compiling,
| while #2 is?  And #2 is the one we currently cannot do?

#2 is the one we cannot currently do.

Elaborating on your #1, if we can produce #1 plus a "cheap" simulator
to build Axiom, then we would be in a very good shape.  Simulators
already exist in the GNU tools chain (see the link I gave earlier
about the way we test GCC on targets we don't have access to).  We
need to convince Axiom to work with GNU toolchains -- that is what
I've been spending time on so far.


-- Gaby