Adds UNIXServer/Socket support for #4858.

…o merge

with that one path BB.  In a sense you can think of it as an airbubble in that
edge!  Bleed it!

…dentally

found I implemented a FIXME where optimize() was already removing unneeded
exception edges :)

This version should mildly improve CFG creation because it does not add and
then remove these edges.  It also has a nice new canRaiseExceptions() in BB.

* Port 2015 redirect logic for popen.
* Improve cloexec for subprocess IO.
* Make cloexec work for any streams we can get realFileno for.
** This allows it to work for JVM-provided pipes if they can be
   unwrapped.

See #4858

See #4858

In
3fe1cc2,
jruby's version of racc/parser.rb was brought in line with MRI
racc/parser.rb. This brought along a change to the yyparse method which
changed the last argument from false to true. This causes the java racc
parser implementation to unconditionally emit very noisy debug output
whenever yyparse is called, where previously it would not emit such
output. Passing true is safe for the pure ruby parser, as it ignores the
argument entirely and does nothing with it. It is also safe for the
native c parser, as that parser only emits debug info if it is compiled
with DEBUG set. Tenderlove's version of racc disabled debug by default
in
ruby/racc@c5359ab,
but it looks like MRI never picked up that change.
For a time, true was passed and debugging info was not emitted (between
9.1.9.0 and 9.1.12.0). This was because in the merge to bring
racc/parser.rb in line with MRI, the loading of the java cparse library
no longer happened, so the pure ruby version of the library was used
which ignores the debug parameter entirely. This changed in
baeaad1
when the cparse-java.jar was being loaded again for jruby.

Fixes #4880.

Don't unconditionally emit debug level info from RACC

This is for #4384.

For this initial version, methods and constants only have a single
value indicating validation count globally. For constants, this
will aggregate all constant names (which are invalidated
separately). For methods, this will aggregate all class hierarchy
invalidations, which will frequently mean a single invalidation
event will increment this value many times, depending on the size
of the hierarchy below it. This latter point could be improved
but we don't really have a clear place to say "this is the top
invalidation" at the moment.

this will include the jline shaded and relocated in the readline.jar

fixes #4740

There are two changes here, with the first the more likely cause
of #4854:

1. While leaving the OpenFile lock locked, we proceeded to
   attempt to lock the Selector being selected. If the order of
   these locks happens differently elsewhere, it will cause a
   deadlock. Fixed by unlocking the OpenFile lock before locking
   the Selector lock.
2. OpenFile.removeBlockingThread synchronized on both OpenFile's
   lock and the OpenFile instance itself. I could find only a
   handful of other places that lock OpenFile and they do not
   appear to lock anything else, but there's no reason to lock
   OpenFile here anyway.

…nd others.

While slowly figuring out a single-line change which fixed this issue I also
noticed several other mismatches between the two parsers.  Hopefully, at this
point, they are identical.

I also deleted a emacs turd.

RubyParser.y appears to be changed but they were minor whitespace differences
with Ripper.  I tried to make these two files a little bit easier to diff.

…- which

seemed to just break yard but not mri tests) from last fix.  Still one more
regression to go.

…k for

duplicated variables in higher level production f_opt and f_block_opt but now
that check is done at the actual variable itself.

For #4885.

…ifferent task

Signed-off-by: Charles Oliver Nutter <headius@headius.com>

See socketry/nio4r#175.

This does not incorporate the updated MRI 2.4 test.

… twice.