As the name implies, dwdebug can "parse" DWARF debug info from ELF
object files, but it is supposed to be much more than that.  The
design includes what I would call a pretty standard set of debugger
objects -- targets, address spaces, debugfiles, symtabs, symbols, etc.

Then, of course, you can attach debugfiles (populated by the DWARF
backend) to address spaces, which are themselves associated with targets.

I only prototyped one target for testing, and that's a Linux userspace
ptrace-based target.  Right now it just can read/write memory.

As usual for an early project, the APIs are incomplete, probably subject
to massive changes, etc.

There are two demo programs at the moment.  targetdumpval will load up
a linux ptrace target (attaches to a running process) and can dump a
memory value on every debugging event.

dumpdebuginfo loads DWARF debug data from an ELF object file into
symtabs and symbols, then dumps it to the screen.  If you run it on the
Linux kernel (vmlinux), it will generate output of ~300MB and take
quite some time.  If you redirect to /dev/null, it'll take about 5 seconds.

Nothing has been optimized; I'm sure there's a lot of memory wastage in the
parsing/symtab population processes; the DWARF expression and location list
"evaluators" are missing (I just hacked enough expression evaluation in to
get us started)...