libcompat: Use WORLDTMP sysroot
ClosedPublic
Actions

Adding powerpc as this effect powerpc64 and I can't test it (the comments about -m32 are only about amd64, powerpc64 builds without additional changes)

Rebase

Harbormaster completed remote builds in B46016: Diff 106997.Jun 16 2022, 10:07 PM

brooks edited parent revisions, added: D35510: ncurses: add SRCHDRS (ncurses_dll.h) to tinfo SRCS; removed: D34696: amd64: improve -m32 support in machine/*.h.Jun 16 2022, 10:08 PM

All the prerequisites for this change have now landed. We'd like to get this in to reduce diffs to CheriBSD before we add support for multiple libcompat ARCHs in the same buildworld.

jhb accepted this revision.Jun 30 2022, 5:41 PM

This revision is now accepted and ready to land.Jun 30 2022, 5:41 PM

I like this simplification.
arm would have trouble here, but we no longer make compat libs for arm.
Also arm64 doesn't support -m32, and it's a fair ways away from that, but this will add that requirement if we want to build 32-bit arm libraries on 64-bit arm via libcompat.
I think these are both fine, and note them in passing to show they were considered when this change was made.

Closed by commit rG724123b9f4d3: libcompat: Use WORLDTMP sysroot (authored by brooks). · Explain WhyJul 1 2022, 7:33 AM

This revision was automatically updated to reflect the committed changes.

brooks added a commit: rG724123b9f4d3: libcompat: Use WORLDTMP sysroot.

Brooks followed up in email to this commit with the following that seems useful to preserve here.

A better reference would have been a good idea, but listing them all
would be a bit excessive. In this case, more time passed between
committing those changes and this one than I'd planned. You can see the
set with:

git log --grep -m32 a09ea2bbc3053d3b2afdb2f7444e5e57a96bb891~..HEAD

In D34697#808973, @imp wrote:

. . .
Also arm64 doesn't support -m32, . . .

I'll note that arm64 has historically made a little non-lib32 use of -m32. As of a commit back in 2018-Jun-27 share/mk/src.opts.mk has:

# aarch64 needs arm for -m32 support.
.elif ${__T} == "aarch64" && ${__llt:Marm*} != ""
__DEFAULT_DEPENDENT_OPTIONS+=   LLVM_TARGET_ARM/LLVM_TARGET_AARCH64

The commit was described via:

QUOTE
LLVM_TARGET_ALL: Default LLVM_TARGET_ARM on for aarch64.
This is needed for -m32 support which is used in the kernel cloudabi32 module. Tweak the style to make it easier to understand.
END QUOTE

In D34697#808973, @imp wrote:

. . .
arm would have trouble here, but we no longer make compat libs for arm.
. . .

Is that for aarch64 ending up with armv7 (or AArch32) lib32 support? If yes, any notes about what the troubles would be?
(I do not have the background to read into the wording anything specific about the difficulties.)