How do these flags get implemented? When I looked at the previous review I assumed that these mapped onto native libproc flags, but now they clearly don't.

DTrace implements these flags itself, entirely in dt_proc.c. It stores them in the proc_handle, I assume for Solaris compatibility, but they don't go any farther into libproc. On FreeBSD, I consider that an abuse, but fixing it would diverge even more from upstream.

I think diverging from upstream here is not problematic, given 1) this code basically never changes upstream and we already have quite a lot of FreeBSD-specific bits, 2) this code is quite a lot harder to understand otherwise. In general I would welcome changes which simplify libdtrace's process handling.

Sounds good to me. I can spend some time on that.

In my opinion, the first step is unifdef illumos. Otherwise, any attempt to simplify just makes things worse. Are we on the same page?

I think that's reasonable. To be clear, I'm not suggesting unifdefing all of libdtrace, just the code that interacts with libproc.

Cool. D41173 does that. I'll rebase this onto that shortly.

Currently we never use this state - do you see any way for this case to happen with the old code?

I think this default case is redundant since you've covered all of the enum cases.

I don't. I'll rip it out.

True, but if dpr_close is corrupted at runtime, this will catch it early and loudly. It's also a documenting assertion that all cases are covered.

AFAIU __unreachable() is a hint to the compiler that "this code path cannot be reached, so optimize accordingly." It will not (by itself) cause the code to fail closed in the case of memory corruption.

To get the effect you want, I'd expect you need to 1) cast the value to an int, 2) use a traditional assert(0).

In the past, there was some directive that would emit a ud2 instruction (on amd64). However, you're right (of course), this presently emits nothing. I'll just remove it.