TODO: keeping fork objects around after run - problems?
TODO: many forks in a row fail to fork all
TODO: $AnyEvent::Fork::Template vs. $TEMPLATE
TODO: $AnyEvent::Fork::Serve::OWNER expose somehow?

1.32 Wed 26 Jan 2022 17:42:30 CET
	- new strategy for reaping children: as local $SIG{CHLD} will
          reset rather than restore signal handlers, we try to reap
          children manually in the fork server. This might cause zombies
          to stay around for a long time in case there is a race and
          no further activity, but perls signal handling is just too
          broken to have other options.

1.31 Thu Nov 10 17:30:45 CET 2016
	- new child API function, AnyEvent::Fork::Serve::run_args.
	- put pid first in child names.

1.3  Thu May 12 18:53:37 CEST 2016
	- the full path to the perl binary is now being passed as argv[0].
	- talk a bit about using AnyEvent::Fork itself after a fork.
	- work around perl 5.19 again breaking backwards compatibility.
        - pod fixes.

1.2  Wed Sep 25 13:05:09 CEST 2013
	- ->fork would not clone, but erase the arguments in the parent
          (analyzed by Christopher Paulicka).
	- mention AnyEvent::Fork::Remote, and how to be compatible to it
          in the description of the run method.
        - AnyEvent::Fork now sets a child watcher for direct children.
        - clarify that one might need to set a child watcher.
        - explicitly call exit on eof in children, to hide(!) win32 perl
          bugs. at least the test suite doesn't complain anymore, but
          the bug is doubtlessly still there.

1.1  Sun Apr 28 15:47:38 CEST 2013
	- (hopefully) remove dependency on common::sense in child code.
        - make dependency on IO::FDPass dynamic in child code.
        - actually open "communications socket" for reading AND writing
          in the child (this does not normally bother syswrite).
	- allow perl path overriding in new_exec.

1.0  Sun Apr 21 14:05:33 CEST 2013
	- implement new_from_fh and to_fh, for easy passing of
          fork objects among processes (EXPERIMENTAL!).
        - sabotage the fork object on run or to_fh.
        - expliticly call exit in child server - not elegant, but
          doesn't disturb any sane perl, and helps broken win32 perls
          to clean up.
        - the write watcher wasn't disabled on write errors.

0.7  Thu Apr 18 22:16:26 CEST 2013
	- only set SIGCHLD to IGNORE while serving requests, reset
          it to the previous value after ->run.
        - minor docpatches.
        - mention AnyEvent::Fork::RPC, but not yte AnyEvent::Fork::Pool.

0.6  Sun Apr  7 00:41:30 CEST 2013
	- this release mostly updates the documentation.
	- add fork+exec replacement example to SYNOPSIS.
        - clarify some eval and run documentation.
        - defuse the documentation a bit - split synopsis into
          an example section, and rewor the problem statement.

0.5   Sat Apr  6 05:30:03 CEST 2013
	- set $SIG{CHLD} to IGNORE in processes forked to avoid
          accumulating zombies.
        - added "typical problems" section.
        - add some unscientific benchmark numbers to indicate relative
        - post-benchmark socket writing optimisations :) (fewer syscalls,
          do more than one write per event loop iteration).

0.2   Fri Apr  5 21:07:59 CEST 2013
	- INCOMPATIBLE CHANGE: AnyEvent::Fork::Template no longer exports
          anything and has renamed the variable.
	- do not leak the internal watcher as return value from run.
        - set $0 to run function name, tune existing $0 names as well.
        - work around a bug in openbsd file descriptor passing on at least
          4.5 and 4.8. might introduce a security problem on that platform.
        - switch to the new IO::FDPass module, which also fixes a bug
          with setting non-blocking mode and gives other improvements.
        - no longer an XS module.

0.01  Thu Apr  4 09:23:33 CEST 2013
	- "technology preview".

0.00  Fri Mar 29 02:15:20 CET 2013
	- post planning, pre writing.