I think what you want here is std::_Exit. It's standard and supported on all platforms and is what std::quick_exit ends up calling anyway (after calling the at_quick_exit handlers).

FWIW -- my two cents here is that I sort of doubt they will actually remove this... since it's been a BSD function for ~30 years. They just don't want people using daemon() or anything that forks because their Apple libs break when you fork. :/

My beautiful code is gone! :'(

Added some comments and a question..

I dream of the day when we don't need Boost for anything. Except maybe the test framework. If that.

Trivially correct. Accepting.

Looks good to me!

Excellent work, sickpig! I won't approve 100% just yet -- I'll wait for deadalnix to have one last look at it.

Oops -- QAPP_ORG_NAME "BitcoinCash" -> "BitcoinABC"

Updated all app bundle-related names to be bitoinabc.org and not bitcoincash.org
Bitcoincash is the protocol. The app org name is bitcoinabc.

Yay!

Modified to use C++11 ranged for loop

Looks good to me. I'll approve as me, and give deadalnix a chance to reply/critique python usage (I am a python nub).

Bah. Arcanist! Why do you vex me?! Fixed with a rebase.

Fixed typo in comment

Aw man thanks a lot!

Ok, I can do that at a later time (like later tonight or tomorrow morning) -- or freetrader -- if you're already set up mind doing it since.. well.. you're already set up?

Yeah agreed that can be a separate PR.. err... diff.

I strongly approve!!

A few comments. Otherwise looks good to me.

This looks good. I like the code refactor -- slightly neater to not call self.start_nodes in each derived class and just move it to the base. All tests pass, as well.

OMFG did you test this?! SUCH A VITAL PIECE OF LOGIC ! It might break everything

I love how you panic, Deadalnix, without bothering to understand what is happening. :)

Yay! Good job! Simple fix, lol..

Looks good to me.

Looks good. Approved.

Ok, this looks good.

.. and as Bitcoin ABC

OK, well I'll accept as myself..

Added some usability suggestions.

In D497#8679, @deadalnix wrote:

Is the problem this is working around happening for core ? What are the next step here. This would be acceptable as a transitional solution, but doesn't seems like a good plan long term.

Added freetrader's recommended changes

All fair points.. I'll implement your recommendations.

Good suggestions by deadalnix.

I can't apply this patch to test it.

Looks Good To Me. However I can't compile it as it complains about missing FastRandomContext::rand64().

Wrestling with arc

Refactored the two test-driver scripts into 1 script for simplicity.

Rebased to include all changes

Ok

I very much approve.

LGTM

I like this diff. A lot.

Apparently you guys like holding locks for trivial checks rather than caching them

Abandoned because it turns out we throw away non-cash ADDR's anyway a few lines later. So it doesn't hurt to accept ADDRs from non-cash nodes in the off chance they are telliing us about a Cash peer.

In D478#8417, @Mengerian wrote:

For networking code, we will always be past UAHF fork. Unless we invent a time machine.

In D479#8414, @freetrader wrote:

The flag should not really be necessary in my view, although I can live with it as an interim measure.

In D479#8388, @zander wrote:

In classic I needed a 3-state.

1. not active
2. active, waiting for > 1MB block
3. active

If you follow that I think you can delete some more overhead elsewhere.

Struggling with arcanist. This should do it.

Eh.. rebase to not pick up unwanted diffs

Use new flag from D479

Assert was in the wrong place.

In D468#8161, @zander wrote:

FYI; the CheckForkWarningConditions() and friends use hard-coded block-counts and show a warning of a nature that I'm not sure is of any value. As such this code has been removed in Classic.

I was on clang-format-4.0.