Move dtrace tools to cross-tools

[emaste]

Summary

libelf and libdwarf need their headers installed, and that implies the legacy stage at the moment.

The dtrace tools are needed for the build only, so they can move to the cross tools stage.

Posting for review of the approach for now; some tweaking may be necessary.

[imp]

Modulo a couple of comments, and maybe an if simplification, this looks good to me.

Makefile.inc1

1238 ↗

(On Diff #4421)

How we've handled this in the past is to have 'reach over' makefiles in the elf toolchain libs. However, I'm not seeing an easy way to do that which doesn't replicate a bunch of directories / makefiles, so this should be good.

1455 ↗

(On Diff #4421)

So there's two issues here. One is that we can't use the host's ctfmerge tools if they are too old. the other issue is that we can't use it at all if the target arch is different.

I don't think the MACHINE != TARGET test is needed at all. We can successfully build pc98 kernels with i386 ctfmerge tools. That's the only case where MACHINE_ARCH == TARGET_ARCH where MACHINE != TARGET.

This would save a wrapped if which is kinda ugly the way this is wrapped too :)

Might want a quick comment to this effect too:

1. Dtrace tools are required when the host's tools are too old, or when
2. we're on a different architecture from the target.

[emaste]

Update comment and remove unnecessary .if, per @imp

[ngie]

Makefile.inc1

1238 ↗

(On Diff #4423)

• Can these be built in parallel?
• Can this list be parameterized out, like I did with buildtools?

1462–1463 ↗

(On Diff #4423)

Where are lib/libdwarf and lib/libelf built via build-tools (or is it cross-tools, I'm so confused...)?

[emaste]

Makefile.inc1

1238 ↗

(On Diff #4423)

Propose a follow-on patch perhaps?

1462–1463 ↗

(On Diff #4423)

With this change they are built during legacy, so these dependencies are incorrect. I think there's no need for an explicit dependency any longer.

[imp]

Makefile.inc1

1238 ↗

(On Diff #4423)

don't worry about building them in parallel. Please. Don't.
There's no need, and this list should be tiny. If it needs to be parallelized, we're doing something horribly wrong.
the legacy stage needs to be absolutely bulletproof. Don't do anything to make it less so.

[emaste]

Rebase after committing the libelf move

[imp]

This looks good to go.
Still don't think the TARGET != MACHINE test is needed, but I couldn't convince myself that there were no cases where it was needed.
Maybe it still is for the case where you have an up-to-date build box, but you have old sources / binaries that you are upgrading so the
ctf tools in that scenario aren't up to date, and this test forces it always. But I just wonder how it is different than other things that are
unconditionally built in the cross tools section... It is most like the btxld, which has similar tests.

I guess this is a long way of saying that this will work, but there might be a simpler way to accomplish the same thing, but there might
not so go ahead :)

[emaste]

In D2144#68384, @imp wrote:

This looks good to go.
Still don't think the TARGET != MACHINE test is needed, but I couldn't convince myself that there were no cases where it was needed.

Yeah, I'm not certain either. It was added in rS233416 by @gonzo.

[emaste]

Rebase after markj@ change
libelf and libdwarf change already committed

[emaste]

Makefile.inc1

1397 ↗

(On Diff #8767)

This line will also be deleted.

[markj]

Looks reasonable to me.