Should be included with the commit that changes the behaviour

The doc change should be included in the same commit that changes the behaviour

In the commit message, this fixes the build

I find the commit message a bit confusing because it doesn't explicitly mention that -U's behaviour changes, but I believe the three individual changes are correct:

OK for spdx/license info

OK. Please include a reference to the source of the float conversion routines in the commit message.

LGTM. We can address leap seconds etc. after commit.

Looks fine to me

Yeah, my motivation here is just the warning from make makeman. I'll note in the commit message that this is NFC.

There are three possible timestamps of interest:

1. As stored in the filesystem
2. Specified in a mtree file
3. Overridden timestamp for reproducible builds (cmdline arg or SOURCE_DATE_EPOCH)

We need to make sure that all four combinations work -- mtree mode / directory scan, -T / no -T

Update comments per @brooks, use XNM instead of XAS

Use IEEE754 instead of the FPU

I thought requesting the API only where needed would be preferred.

What do you think of just adding it in secure/ssh.mk?

We might need to implement some of these in the future but seems reasonable for now to continue stubbing them out

I notice that in Linux this is 11 or 13 by default on aarch64 if configured for 16k/64k pages (respectively), but I'm sure this is reasonable for the foreseeable future.

Adding a new port is a good time to check portlint/portclippy/portfmt for useful suggestions. That said, I appreciate the value of keeping formatting close to the other OpenJDK ports. I mention this for completeness.

Cirrus-CI test run: https://cirrus-ci.com/build/5750132458651648

Changes to this section over time include:

Also, makefs has some reasonable existing tests; it should be straightforward to add a test for -T with Rock Ridge TF.

As an aside the way makefs uses stampst.st_ino as a flag feels kind of hacky to me, but is consistent with what's already done.

Upstream issue open in https://github.com/llvm/llvm-project/issues/132216

I think this is fine. backticks and checkyesno moving are minor improvements (arguably not bugs).

It looks like <= PAGE_SIZE would indeed be correct

https://ci.freebsd.org is currently down; I'll take a look soon

OK, even though these seem kind of silly (i.e., not much more meaning than #define THOUSAND 1000).

Let's add a comment describing the functionality here, and make it explicit that the stub implementations are what Linux provides when CONFIG_REF_TRACKER is not defined. (Make it clear that this is distinct from the case in Linuxkpi where we sometimes provide stubs that are actually insufficient / not correct, but function because the driver code calling them does not actually need the results.) Other than that LGTM.

Incidentally looking at Linux code to check this answers one of the questions I had about SOF. Linux has:

Worth putting the comment from the commit message in the src?

OK, or we could just have the header guards.

diff against correct version

Remove ObsoleteFiles.inc entry

add OptionalObsoleteFiles.inc

use elfcopy name

We could always install it as elftc-objcopy and add LINKS+= ${BINDIR}/${PROG} ${BINDIR}/objcopy in the MK_LLVM_OBJCOPY=no case and set EFI_OBJCOPY?= elftc-objcopy always; reviewers let me know if you think that's preferable/simpler.

Is there a reason not to apply (some of) the patches in the repo directly?

Who is supported to cherry-pick to releng?