Re: git: 83bf6ab56829 - main - uname: switch machine to HW_MACHINE_ARCH

On Dec 12, 2022, at 13:49, Mark Millard <marklmi@yahoo.com> wrote:

> On Dec 12, 2022, at 11:51, Mark Millard <marklmi@yahoo.com> wrote:
> 
>> Piotr Kubaj <pkubaj_at_anongoth.pl> wrote on
>> Date: Mon, 12 Dec 2022 14:48:57 UTC :
>> 
>>> Reverted. Sorry for the breakage. I think will return with the next
>>> version of this patch and this time I'll make sure to run make universe
>>> on my powerpc64le instead of those pesky universe14 hosts :)
>> 
>> 
>> I expect that any buildworld buildkernel that explicitly has TARGET and
>> TARGET_ARCH set on the command line or in the environment might work,
>> both cross builds and explicitly set to match the system doing the
>> builds (explicit but actually native --in other worlds, a limiting
>> condition of a cross build, a self-hosted cross-style build).

For the above:

I got the TARGET/TARGET_ARCH use wrong: using the likes
of MACHINE=arm explicitly on the make command line or in
the environment is what avoids make doing the wrong thing
via its otherwise automatic internal assignment. It is not
the target that is the problem. See:

https://lists.freebsd.org/archives/freebsd-arm/2022-December/002086.html

>> So I'm not sure that universe builds are a sufficient test context for
>> the specific type of change: You likely need a set of tests that do
>> not assign TARGET or TARGET_ARCH explicitly but are executed in
>> environments that look to be native for each example architecture.
>> You might need to ask folks with systems that you do not have examples
>> of to do some testing of non-explicit native builds.

So the above is invalid.

> I left something implicit for the non-explicit testing
> that may not be obvious: a full sequence for a test
> seems to be something like:
> 
> A) from a pre-change context, build with the change
> B) install the change and boot it
> C) Try a native rebuild and install without any explicit
>    TARGET/TARGET_ARCH assignment.
> 
> Making it through (C) should test the non-explicit case
> for the kind of environment involved.
> 
> Of course, if (B) fails or leads to failure in (C),
> recovering via more from-source build activity in that
> messed up environment could be difficult. Using explicit
> TARGET and TARGET_ARCH assignments might help?

Doing a build test after doing an update to be running
something that contains the change still looks appropriate.
It is MACHINE= that should not be explicitly used in such
a test.

===
Mark Millard
marklmi at yahoo.com