I think it might make sense to split the switch to the jemalloc memalign() into a separate commit.

Looks correct

I wonder if a chrootat(fd, path) that allows a NULL path would be more general?

Require explicit __FOO_DEFAULT

Require explicit DEFAULT value

Define default value

Improve manpage text

Add note about future plans

Tweak comments

I've been debating if __FOO_DEFAULT and __FOO_OPTIONS should have the __ prefix. I think it's arguable that they are part of the public interface, but I'm not sure.

I did realize as I was working on makeman support that we should also implement what ports calls MULTI options to tame the LLVM backend support, but I'd like to land this stack first as I need to get back to the work that motivated it.

Improve description file

Remove .It from description

Add tools/build/options/LIBC_MALLOC

• Add BROKEN_SINGLE_OPTIONS
• Add copy to kern.opts.mk
• Unset __SINGLE_OPTIONS

Fix typo

@jrtc27 rightly points out I should add makeman (and thus showconfig) support. I'll look into that in followup commits.

Fix typo

Ports also has RADIO, GROUP, and MULTI for different set types, but this is the one I want now.

Rebase on D41481

In D41457#944650, @jrtc27 wrote:

Though please note the .PATH GC'ing in the commit message

• Fix SYM_MAPS
• Remove unused .PATH

Use LIBC_SRCTOP relative paths

Use LIBC_SRCTOP relative path

I don't have a strong preference. I just think it's kind of lame to have a heirarchy but not really. That being said, .PARSEDIR is a bit weird in that you have to use it in a .foo or force an expansion immediately so maybe that argues for sticking with LIBC_SRCTOP.

This change and D41421 conflict.

It seems extremely late in the release to be doing a malloc update.