Not completely silent unfortunately.

One test so far: java.lang.System.nanoTime()

We're only about 500% slower than native.

-MD accepts no argument, and luckily the argument we supplied evaluated to
nothing.  Remove it.

the -MP switch allows make to ignore deleted files.

this gives us better dependency tracking.

We need different rules for .s (compiled by 'gas') and .S ('gcc') since
the compiler and assembler take different command line options.

It's as arch-specific as they come.

This implementation stores small objects in pools of similar-sized objects,
while large objects are allocated using a first-fit algorithm.  There is also
a specialized interface for allocating aligned pages.

order to reach includes files for within the hirarchy.

No threading yet, so no real locking

This forces the linker to add a DYNAMIC segment, allowing us to export our
symbols to loaded libraries.

This gives us full control over linker options.

We need the headers so we can provide the symbols to loaded libraries.
This means we cannot use objcopy to munge the binary format to elf32-i386,
as the loader doesn't understand this format.

Instead, add a separate 32-bit only stage (in elf32-i386 format) that loads
the 64-bit image into memory as its payload.  This keeps multiboot happy and
leave our headers alone.