Merge remote-tracking branch 'joyent/v0.12' into v1.x

I was originally going to do this after the v0.11.15 release, but as
that release is three weeks overdue now, I decided not to wait any
longer; we don't want the delta to get too big.

Conflicts:
lib/net.js
test/simple/simple.status

PR-URL: https://github.com/iojs/io.js/pull/236
Reviewed-By: Bert Belder <bertbelder@gmail.com>
Reviewed-By: Fedor Indutny <fedor@indutny.com>

smalloc: fix bad assert for zero length data

If the data length passed to smalloc.alloc() the array_length will be
zero, causing an overflow check to fail. This prevents that from
happening.

Signed-off-by: Trevor Norris <trev.norris@gmail.com>

net: make connect() input validation synchronous

Socket.prototype.connect() sometimes throws on bad inputs
after an asynchronous operation. This commit makes the input
validation synchronous. This commit also removes some hard
coded IP addresses.

PR-URL: https://github.com/joyent/node/pull/8180
Fixes: https://github.com/joyent/node/issues/8140
Reviewed-By: Trevor Norris <trev.norris@gmail.com>
Reviewed-By: Fedor Indutny <fedor@indutny.com>
Reviewed-By: Timothy J Fontaine <tjfontaine@gmail.com>

docs: update to authors file

PR-URL: https://github.com/joyent/node/pull/8964
Reviewed-by: Trevor Norris <trev.norris@gmail.com>

doc: added TC meeting minutes 2014-12-30

closes #211

PR-URL: https://github.com/iojs/io.js/pull/229
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>

build: i18n: add icu config options

Make "--with-intl=none" the default and add "intl-none" option to
vcbuild.bat.

If icu data is missing print a warning unless either --download=all or
--download=icu is set. If set then automatically download, verify (MD5)
and unpack the ICU data if not already available.

There's a "list" of URLs being used, but right now only the first is
picked up. The logic works something like this:

* If there is no directory deps/icu,
  * If no zip file (currently icu4c-54_1-src.zip),
    * Download zip file (icu-project.org -> sf.net)
  * Verify the MD5 sum of the zipfile
    * If bad, print error and exit
  * Unpack the zipfile into deps/icu
* If deps/icu now exists, use it, else fail with help text

Add the configuration option "--with-icu-source=..."

Usage:
  * --with-icu-source=/path/to/my/other/icu
  * --with-icu-source=/path/to/icu54.zip
  * --with-icu-source=/path/to/icu54.tgz
  * --with-icu-source=http://example.com/icu54.tar.bz2

Add the configuration option "--with-icu-locals=...".  Allows choosing
which locales are used in the "small-icu" case.

Example:
    configure --with-intl=small-icu --with-icu-locales=tlh,grc,nl

(Also note that as of this writing, neither Klingon nor Ancient Greek
are in upstream CLDR data. Serving suggestion only.)

Don't use hard coded ../../out paths on windows. This was suggested by
@misterdjules as it causes test failures.  With this fix, "out" is no
longer created on windows and the following can run properly:

    python tools/test.py simple

Reduce space by about 1MB with ICU 54 (over without this patch). Also
trims a few other source files, but only conditional on the exact ICU
version used. This is to future-proof - a file that is unneeded now may
be needed in future ICUs.

Also:
  * Update distclean to remove icu related files
  * Refactor some code into tools/configure.d/nodedownload.py
  * Update docs
  * Add test

PR-URL: https://github.com/joyent/node/pull/8719
Fixes: https://github.com/joyent/node/issues/7676#issuecomment-64704230
[trev.norris@gmail.com small change to test's whitespace and logic]
Signed-off-by: Trevor Norris <trev.norris@gmail.com>

doc: added TC meeting minutes 2014-12-17

Closes #163

PR-URL: https://github.com/iojs/io.js/pull/178
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Bert Belder <bertbelder@gmail.com>

doc: util: document --trace-deprecation

Document the --trace-deprecation flag and the `process` properties that
affect util.deprecate().

Fixes: https://github.com/iojs/io.js/issues/190
PR-URL: https://github.com/iojs/io.js/pull/191
Reviewed-By: Jonathan Ong <me@jongleberry.com>

lib: introduce process module

This makes require('process') always return a reference to the global
process object.

PR-URL: https://github.com/iojs/io.js/pull/206
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

url: improve parsing speed

The url.parse() function now checks whether an escapable character is in
the URL before trying to escape it.

PR-URL: https://github.com/joyent/node/pull/8638
[trev.norris@gmail.com: Switch to use continue instead of if]
Signed-off-by: Trevor Norris <trev.norris@gmail.com>

doc: add tls server.close() callback docs

Also, tests to confirm its existence.

PR-URL: https://github.com/iojs/io.js/pull/217
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

dgram: close() should accept a callback

Like net, http, and https server.close, and socket.end(), etc.

PR-URL: https://github.com/iojs/io.js/pull/217
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

dgram: make 'close' event async

Emit the close event asynchronously, after the close, as it is with the
net/http close events.

PR-URL: https://github.com/iojs/io.js/pull/217
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

doc: be specific about the unit of `maxBuffer`

The maxBuffer option was not self-documenting, so document the unit and
its effect.

PR-URL: https://github.com/iojs/io.js/pull/209
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

dgram: changes Socket.close() to return itself

This commit adds a return statement to the dgram.Socket.close()
function that returns itself after it finishes. This follows along
the functionality of the more popular and, dare I say, father-library
`lib/net.js`.

PR-URL: https://github.com/iojs/io.js/pull/214
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

dgram: change Socket.bind() to return itself

This commit changes `lib/dgram.js` Sockets to, when
they are bound to a port / IP, return themselves. This
is done in order to allow chaining of methods and be
in accordance with the `lib/net.js` library.

PR-URL: https://github.com/iojs/io.js/pull/214
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

configure: remove compiler detection

The GCC version is no longer relevant since only 4.8 and newer are
supported. It's probably safe to assume clang on mac since V8 does
so too.

PR-URL: https://github.com/iojs/io.js/pull/205
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

src: remove cpplint error using NOLINT directive

PR-URL: https://github.com/iojs/io.js/pull/202
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

doc: document time() and timeEnd() relationship

`console.time()` and `console.timeEnd()` are very closely related. It's
useful to reference them both from each other.

Previously, console.time() did not mention that it needed to be paired
with a call to console.timeEnd() to be useful, and timeEnd() also failed
to mention that console.time() needed to be called first.

References in both directions have been added.

PR-URL: https://github.com/iojs/io.js/pull/198
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

doc: fix v8 doc headings so 'make doc' passes

Previously the code that builds the Table of Contents threw
an exception because of jump from an H1 heading directly to an H3
heading.

By changing the H3 heading to an H2, 'make doc' works again. This is
also consistent with other docs like console.markdown which use
H2 for method call documentation.

PR-URL: https://github.com/iojs/io.js/pull/203
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

assert: fix deepEqual regression

Change of Object.keys in ES6 breaks assert.deepEqual about primitive
values.

V8: https://code.google.com/p/v8/issues/detail?id=3443

Previously deepEqual depends on Object.key that throws an error for
a primitive value, but now Object.key does not throw.

PR-URL: https://github.com/iojs/io.js/pull/193
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

configure: remove workarounds for GCC < 4.8

They can't compile io.js anyway.

PR-URL: https://github.com/iojs/io.js/pull/187
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

lib: micro-optimize url.resolve()

Replace the call to Array#splice() with a faster open-coded version
that creates less garbage.

Add a new benchmark to prove it.  With the change applied, it scores
about 5% higher and that is nothing to sneeze at.

PR-URL: https://github.com/iojs/io.js/pull/184
Reviewed-By: Chris Dickinson <christopher.s.dickinson@gmail.com>

benchmark: rename url.parse() benchmark

Rename the url.parse() benchmark from url.js to url-parse.js.
A follow-up commit is going to add another one for url.resolve().

PR-URL: https://github.com/iojs/io.js/pull/184
Reviewed-By: Chris Dickinson <christopher.s.dickinson@gmail.com>

lib: micro-optimize EventEmitter#removeListener()

Replace the call to Array#splice() with a faster open-coded version
that creates less garbage.

Add a new benchmark to prove it.  With the change applied, it scores
a whopping 40% higher.

PR-URL: https://github.com/iojs/io.js/pull/185
Reviewed-By: Chris Dickinson <christopher.s.dickinson@gmail.com>

benchmark: fix printing of large numbers

Don't use Number#toPrecision(), it switches to scientific notation for
numbers with more digits than the precision; use Number#toFixed().

PR-URL: https://github.com/iojs/io.js/pull/185
Reviewed-By: Chris Dickinson <christopher.s.dickinson@gmail.com>

benchmark: pre-optimize url.parse() before start

Force V8 to optimize url.parse() before starting the actual benchmark.
Tries to minimize variance between successive runs caused by the
optimizer kicking in at different points.

It does not seem to have much impact, CPU times are roughly the same
before and afterwards; url.parse() quickly plateaus at a local optimum
where most time is spent in V8 builtins, notably Runtime_StringSplit()
and Object::GetElementWithReceiver() calls originating from
deps/v8/src/uri.js, with no recurring optimize/deoptimize cycles that
I could spot.

Still, I don't see any downsides to pre-optimizing the function being
benchmarked so in it goes.

PR-URL: https://github.com/iojs/io.js/pull/132
Reviewed-By: Chris Dickinson <christopher.s.dickinson@gmail.com>
Reviewed-By: Fedor Indutny <fedor@indutny.com>

doc: add details for http res/req end callback

Add documentation for the callback parameter of http.ClientRequest's and
http.ServerResponse's end methods.

Signed-off-by: Julien Gilli <julien.gilli@joyent.com>

debugger: improve clearBreakpoint error and docs

Currently clearBreakpoint error is confusing, it says "Script not found"
when there is no breakpoint, also documentation doesn't include
signature for clearBreakpoint.

PR-URL: https://github.com/iojs/io.js/pull/175
Reviewed-By: Miroslav Bajtoš <miroslav@strongloop.com>

fs: deprecate exists() and existsSync()

These methods don't follow standard conventions, and shouldn't
be used anyway.

Fixes: https://github.com/iojs/io.js/issues/103
PR-URL: https://github.com/iojs/io.js/pull/166
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

doc: document callback of http res/req end method

Add documentation for the callback parameter of http.ClientRequest's and
http.ServerResponse's end methods.

PR-URL: https://github.com/iojs/io.js/pull/181
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Brendan Ashworth <squirrelslikeacorns@gmail.com>

lib,src: remove post-gc event infrastructure

Remove the 'gc' event from the v8 module and remove the supporting
infrastructure from src/.  It gets the axe because:

1. There are currently no users.  It was originally conceived as
   an upstreamed subset of StrongLoop's strong-agent GC metrics,
   but the strong-agent code base has evolved considerably since
   that time and has no use anymore for what is in core.

2. The implementation is not quite sound.  It calls into JS land
   from inside the GC epilog and that is unsafe.  We could fix
   that by delaying the callback until a safe time but because
   there are no users anyway, removing it is all around easier.

PR-URL: https://github.com/iojs/io.js/pull/174
Reviewed-By: Trevor Norris <trev.norris@gmail.com>

stream: switch _writableState.buffer to queue

In cases where many small writes are made to a stream
lacking _writev, the array data structure backing the
WriteReq buffer would greatly increase GC pressure.

Specifically, in the fs.WriteStream case, the
clearBuffer routine would only clear a single WriteReq
from the buffer before exiting, but would cause the
entire backing array to be GC'd. Switching to [].shift
lessened pressure, but still the bulk of the time was
spent in memcpy.

This replaces that structure with a linked list-backed
queue so that adding and removing from the queue is O(1).
In the _writev case, collecting the buffer requires an
O(N) loop over the buffer, but that was already being
performed to collect callbacks, so slowdown should be
neglible.

PR-URL: https://github.com/joyent/node/pull/8826
Reviewed-by: Timothy J Fontaine <tjfontaine@gmail.com>
Reviewed-by: Trevor Norris <trev.norris@gmail.com>

lib: fix guard expression in timer.unref()

Fixes the following assertion on slow systems, like our ARM buildbot:

    $ out/Debug/node test/simple/test-timers-unref.js
    node: ../src/async-wrap-inl.h:101: v8::Handle<v8::Value>
    node::AsyncWrap::MakeCallback(uint32_t, int,
    v8::Handle<v8::Value>*): Assertion `cb_v->IsFunction()' failed.
    Aborted

The reason it only manifests on slow systems is that the test starts
a 1 ms interval timer, then defers timer.unref.bind({}) to the next
tick.  On fast systems, the test completes in under a millisecond,
before the callback is called.

This commit makes timer.unref() check that the receiver actually has
a timeout callback property.

Fixes #13.

PR-URL: https://github.com/iojs/io.js/pull/165
Reviewed-By: Rod Vagg <rod@vagg.org>

src: fix windows build error

Fix a Windows-only build error that was introduced in
commit 1183ba4 ("zlib: support concatenated gzip files").

Rename the NO_ERROR and FAILED enumerations, they conflict
with macros of the same name in <winerror.h>.

PR-URL: https://github.com/joyent/node/pull/8893
Reviewed-By: Fedor Indutny <fedor@indutny.com>
Reviewed-By: Rod Vagg <rod@vagg.org>
Reviewed-by: Timothy J Fontaine <tjfontaine@gmail.com>

debugger: fix unhandled error in setBreakpoint

Fix Interface.setBreakpoint() to correctly handle an attempt to set a
breakpoint in the current script when there is no current script.
This usually happens when the debugged process is not paused.

Fixes: https://github.com/joyent/node/issues/6453
PR-URL: https://github.com/joyent/node/pull/6460
Reviewed-By: Chris Dickinson <christopher.s.dickinson@gmail.com>

test: run tests in parallel, common improvements

* Allow running tests in mixed parallel/sequential modes
* Add -J flag for running tests on all available CPUs
* Support TEST_THREAD_ID in test/common.js and use it for tmpDir and PORT
* make: use -J flag

Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
PR-URL: https://github.com/iojs/io.js/pull/172
Fix: iojs/io.js#139

test: split test in parallel/sequential

Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
PR-URL: https://github.com/iojs/io.js/pull/172
Fix: iojs/io.js#139

zlib: support concatenated gzip files

Reviewed-By: Fedor Indutny <fedor@indutny.com>
PR-URL: https://github.com/joyent/node/pull/6442

doc: clarify create{Read,Write}Stream fd option

Clarify the fd option: it is preferred to the path parameter, omits
the "open" event if given, and is available on WriteStreams as well.

PR-URL: https://github.com/joyent/node/issues/7707
Fixes: https://github.com/joyent/node/issues/7707
Fixes: https://github.com/joyent/node/issues/7708
Fixes: https://github.com/joyent/node/issues/4367
Reviewed-By: Chris Dickinson <christopher.s.dickinson@gmail.com>

fs: add access() and accessSync()

fs.exists() and fs.existsSync() do not follow the typical
error first callback convention. access() and accessSync()
are added as alternatives in this commit.

Fixes: https://github.com/joyent/node/pull/8714
PR-URL: https://github.com/iojs/io.js/pull/114
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Domenic Denicola <domenic@domenicdenicola.com>

doc: change write() after end() streams example

Currently there's an example using http.ServerResponse stream, which
has a known bug and will not throw an error while writing after end().
Changed to a writable stream from fs which behaves as expected.

Fixes https://github.com/joyent/node/issues/8814.

PR-URL: https://github.com/iojs/io.js/pull/155
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Brendan Ashworth <squirrelslikeacorns@gmail.com>

lib: reintroduce v8 module

I introduced this module over a year ago in a pull request as the v8
module but it was quickly subsumed by the tracing module.

The tracing module was recently removed again and that is why this
commit introduces the v8 module again, including the new features it
picked up commits d23ac0e and f8076c4.

PR-URL: https://github.com/iojs/io.js/pull/131
Reviewed-By: Chris Dickinson <christopher.s.dickinson@gmail.com>
Reviewed-By: Christian Tellnes <christian@tellnes.no>
Reviewed-By: Thorsten Lorenz <thlorenz@gmx.de>
Reviewed-By: Trevor Norris <trev.norris@gmail.com>

src: move BE/LE buffer conversion to StringSlice()

Move the big endian to little endian conversion logic for UCS2 input
from src/string_bytes.cc to src/node_buffer.cc; StringSlice() is the
only function that actually needs it and with this commit, a second
copy is avoided on big endian architectures.

src: redo unaligned access workaround

Introduce two-byte overloads of node::Encode() and StringBytes::Encode()
that ensure that the input is suitably aligned.

Revisits commit 535fec8 from yesterday.

src: fix addon loader regression

Fix a regression that was introduced in commit a38b9178 by removing the
bad check.  Also rearrange the addon loading logic to ensure that the
list of pending addons remains in a consistent state when the shared
object fails to load; in particular, when an addon self-registers first,
then hits a dynamic linker error in a later constructor.

Fixes the following asserting when loading a .node shared object:

    node: ../src/node.cc:1944: void node::node_module_register(void*):
    Assertion `(modpending) != (nullptr)' failed.

Fixes strongloop/strongops#233.

PR-URL: https://github.com/iojs/io.js/pull/154
Reviewed-By: Ryan Graham <ryan@strongloop.com>

build: remove support for VS 2010 and 2012

They can't compile io.js anyway.

PR-URL: https://github.com/iojs/io.js/pull/156
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Bert Belder <bertbelder@gmail.com>

domain: add soft deprecation notice

Per the TC meeting on 2014-12-10, domains will be "soft deprecated"
until a suitable replacement API is available; at which time they
will be fully deprecated. Full deprecation will include references
to replacement API and the application of util.deprecate to the domain
api.

PR-URL: https://github.com/iojs/io.js/pull/141
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

test: add test for spawnSync() env option

PR-URL: https://github.com/joyent/node/pull/8845
Reviewed-by: Trevor Norris <trev.norris@gmail.com>

lib,src: fix spawnSync ignoring its 'env' option

PR-URL: https://github.com/joyent/node/pull/8546
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>

docs: clarify that 'io.js' is the correct name

Refs #118.

PR-URL: https://github.com/iojs/io.js/pull/151
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Rod Vagg <rod@vagg.org>

doc: added node-forward TC meeting 2014-10-15

PR-URL: https://github.com/iojs/io.js/pull/148
Reviewed-By: Mikeal Rogers <mikeal.rogers@gmail.com>

domain: forward args to .run(fn)

Adds the feature to define arguments for the function called in
domain.run(), this is supposed to be useful when a function is called from
another context and some values from the current context are needed as
arguments, it's similar to the callback from setTimeout or setInterval.

PR-URL: https://github.com/iojs/io.js/pull/15
Reviewed-By: Chris Dickinson <christopher.s.dickinson@gmail.com>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

doc: added TC meeting minutes for 2014-12-10

closes #112

PR-URL: https://github.com/iojs/io.js/pull/144/
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Trevor Norris <trev.norris@gmail.com>
Reviewed-By: Mikeal Rogers <mikeal.rogers@gmail.com>

module: fix regression in require ../

Fixes regression in require system that prevented loading relative
packages via main property in package.json where the file is not
index.{ext}. The regression was introduced in commit 36777d2.

PR-URL: https://github.com/iojs/io.js/pull/145
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

build: fix dtrace-enabled build on os x

Commit 691d55f introduces a regression on OS X when dtrace is enabled
(the default.)  This commit rectifies that by removing the erroneous
platform check.

Fixes the following build error:

    CXX(target) /Users/bnoordhuis/src/iojs/out/Release/obj.target/node/src/node_dtrace.o
    ../src/node_dtrace.cc:27:10: fatal error: 'node_provider.h' file not found

PR-URL: https://github.com/iojs/io.js/pull/150
Reviewed-By: Miroslav Bajtoš <miro.bajtos@gmail.com>

doc: add TC notes from october 29th

PR-URL: https://github.com/node-forward/node/pull/41
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
@rvagg: cherry-picked 2c7847b and e7dec60 from master to v0.12

doc: meeting notes from oct 9th TC meeting

Fixes https://github.com/node-forward/node/issues/4

PR-URL: https://github.com/node-forward/node/pull/9
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
@rvagg: cherry-picked 3cb5f3d from master to v0.12

docs: reword project messaging

Rewords project messaging in README to make the overall project
messaging a bit clearer. More discussion to be found in iojs/io.js#24.

PR-URL: https://github.com/iojs/io.js/pull/36
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
Reviewed-By: Jeremiah Senkpiel <fishrock123@rocketmail.com>
Reviewed-By: Rod Vagg <rod@vagg.org>

build: configure return exit status from gyp

Previously, 'configure' would not return an exit status
if gyp blows up. This can be tested via:

    date >> node.gyp ; ./configure && echo A-OK

You will get "A-OK" even though gyp had failed.

PR-URL: https://github.com/joyent/node/pull/8856
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

crypto: separate altname extensions with ", "

In newly introduced `SafeX509ExtPrint` I forgot to insert separators
between extensions, which lead to the "DNS:...DNS:..." thing for npm.

Fix: iojs/io.js#105
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>
PR-URL: https://github.com/iojs/io.js/pull/113.patch

test: mark more tests as flaky

Marking these two tests as flaky, since they have been failing
intermittenly in recent builds:
test-debug-signal-cluster
test-cluster-basic

crypto: disable SSLv3 if shared OpenSSL lacks it

Some distributions disable SSLv3 due to POODLE.  In such a case, disable
the specific SSLv3 methods and throw an exception, much like the code
already does for SSLv2.  The SSLv23* code is retained because this is
OpenSSL's terminology for "no version in particular".

Reviewed-By: Fedor Indutny <fedor@indutny.com>
PR-URL: https://github.com/iojs/io.js/pull/101

deps: update libuv to 1.0.2

PR-URL: https://github.com/joyent/node/pull/8847
Reviewed-by: Trevor Norris <trev.norris@gmail.com>

benchmark: add url benchmarks

Based on the ad-hoc benchmark from joyent/node#8638 plus an additional
benchmark for user:pass auth URLs.

PR-URL: https://github.com/iojs/io.js/pull/102
Reviewed-by: Chris Dickinson <christopher.s.dickinson@gmail.com>

lib: optimize require() path walking

Remove a speed bump from commit 36777d2 by reusing the result of the
previous stat() system call.  It's a code path that gets called many
thousands of times at startup in most applications so shaving off an
extra system call can have an appreciable impact on startup times.

PR-URL: https://github.com/iojs/io.js/pull/130
Reviewed-by: Chris Dickinson <christopher.s.dickinson@gmail.com>

build: do not generate support for libuv's probes

Dtrace probes were removed from libuv recently, but their usage by node
was not completely removed, causing build breaks on SmartOS.

Even though the build is working on other platforms, these probes are
not fired by libuv anymore, so there's no point in using them on these
platforms too.

Reviewed-by: Trevor Norris <trev.norris@gmail.com>

src: fix unaligned access in ucs2 string encoder

Seen with g++ 4.9.2 on x86_64 Linux: a SIGSEGV is generated when the
input to v8::String::NewFromTwoByte() is not suitably aligned.

g++ 4.9.2 emits SSE instructions for copy loops.  That requires aligned
input but that was something StringBytes::Encode() did not enforce until
now.  Make a properly aligned copy before handing off the input to V8.

We could, as an optimization, check that the pointer is aligned on a
two-byte boundary but that is technically still UB; pointers-to-char
are allowed to alias other pointers but the reverse is not true:
a pointer-to-uint16_t that aliases a pointer-to-char is in violation
of the pointer aliasing rules.

See https://code.google.com/p/v8/issues/detail?id=3694

Fixes segfaulting test simple/test-stream2-writable.

PR-URL: https://github.com/iojs/io.js/pull/127
Reviewed-by: Trevor Norris <trev.norris@gmail.com>

test: fix test-fs-symlink-dir-junction-relative

  * The test no longer relies on being invoked from a particular
    working directory to function properly.
  * fs.symlink() and fs.symlinkSync() are both tested.
  * The test now cleans up after itself.

This commit fixes https://github.com/iojs/io.js/issues/126

PR-URL: https://github.com/iojs/io.js/pull/129
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

test: add test for spawnSync() env option

PR-URL: https://github.com/joyent/node/pull/8845
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

lib,src: fix spawnSync ignoring its 'env' option

PR-URL: https://github.com/joyent/node/pull/8546
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>

build: fix build with systemtap providers enabled

The "dtrace" script version include in systemtap-sdt-devel-2.6-3
(part of Fedora 21) no longer ignores unknown command line
arguments, but will instead error out and refuse to run.
This patch adds a separate condition to node's gyp input so
that on Linux it will run dtrace without the -xnolibs
argument that trips it up on systemtap-std-devel-2.6-3.

PR-URL: https://github.com/joyent/node/pull/8846
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

Merge 'node/v0.12'

This merge is effectively a no-op because io.js has already
cherry-picked all the patches from node it needs.

The merge commit serves to establish a new merge base for future merges.

src: remove the tracing module entirely

PR-URL: https://github.com/iojs/io.js/pull/124
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

doc: mention callback for http res/req write & end

Add documentation for the callback parameter of http.ClientRequest's and
http.ServerResponse's write and end methods.

async-wrap: add event hooks

Call a user-defined callback at specific points in the lifetime of an
asynchronous event. Which are on instantiation, just before/after the
callback has been run.

**If any of these callbacks throws an exception, there is no forgiveness
or recovery. A message will be displayed and a core file dumped.**

Currently these only tie into AsyncWrap, meaning no call to a hook
callback will be made for timers or process.nextTick() events. Though
those will be added in a future commit.

Here are a few notes on how to make the hooks work:

- The "this" of all event hook callbacks is the request object.

- The zero field (kCallInitHook) of the flags object passed to
  setupHooks() must be set != 0 before the init callback will be called.

- kCallInitHook only affects the calling of the init callback. If the
  request object has been run through the create callback it will always
  run the before/after callbacks. Regardless of kCallInitHook.

- In the init callback the property "_asyncQueue" must be attached to
  the request object. e.g.

  function initHook() {
    this._asyncQueue = {};
  }

- DO NOT inspect the properties of the object in the init callback.
  Since the object is in the middle of being instantiated there are some
  cases when a getter is not complete, and doing so will cause Node to
  crash.

PR-URL: https://github.com/joyent/node/pull/8110
Signed-off-by: Trevor Norris <trev.norris@gmail.com>
Reviewed-by: Fedor Indutny <fedor@indutny.com>
Reviewed-by: Alexis Campailla <alexis@janeasystems.com>
Reviewed-by: Julien Gilli <julien.gilli@joyent.com>

async-wrap: explicitly pass parent

When instantiating a new AsyncWrap allow the parent AsyncWrap to be
passed. This is useful for cases like TCP incoming connections, so the
connection can be tied to the server receiving the connection.

Because the current architecture instantiates the *Wrap inside a
v8::FunctionCallback, the parent pointer is currently wrapped inside a
new v8::External every time and passed as an argument. This adds ~80ns
to instantiation time.

A future optimization would be to add the v8::External as the data field
when creating the v8::FunctionTemplate, change the pointer just before
making the call then NULL'ing it out afterwards. This adds enough code
complexity that it will not be attempted until the current approach
demonstrates it is a bottle neck.

PR-URL: https://github.com/joyent/node/pull/8110
Signed-off-by: Trevor Norris <trev.norris@gmail.com>
Reviewed-by: Fedor Indutny <fedor@indutny.com>
Reviewed-by: Alexis Campailla <alexis@janeasystems.com>
Reviewed-by: Julien Gilli <julien.gilli@joyent.com>

async-wrap: expose async-wrap as binding

Expose basic hooks for AsyncWrap via the async_wrap binding. Right now
only the PROVIDER types are exposed. This is a preliminary step before
more functionality is added.

PR-URL: https://github.com/joyent/node/pull/8110
Signed-off-by: Trevor Norris <trev.norris@gmail.com>
Reviewed-by: Fedor Indutny <fedor@indutny.com>
Reviewed-by: Alexis Campailla <alexis@janeasystems.com>
Reviewed-by: Julien Gilli <julien.gilli@joyent.com>

src: remove unnecessary template parameter

The template class information is received via the type of the first
argument. So there is no need to use Wrap<T>(handle).

PR-URL: https://github.com/joyent/node/pull/8110
Signed-off-by: Trevor Norris <trev.norris@gmail.com>
Reviewed-by: Fedor Indutny <fedor@indutny.com>
Reviewed-by: Alexis Campailla <alexis@janeasystems.com>
Reviewed-by: Julien Gilli <julien.gilli@joyent.com>

src: all wraps now use actual FunctionTemplate

Instead of simply creating a new v8::Object to contain the connection
information, instantiate a new instance of a FunctionTemplate. This will
allow future improvements for debugging and performance probes.

Additionally, the "provider" argument in the ReqWrap constructor is no
longer optional.

PR-URL: https://github.com/joyent/node/pull/8110
Signed-off-by: Trevor Norris <trev.norris@gmail.com>
Reviewed-by: Fedor Indutny <fedor@indutny.com>
Reviewed-by: Alexis Campailla <alexis@janeasystems.com>
Reviewed-by: Julien Gilli <julien.gilli@joyent.com>

node: fix throws before timer module is loaded

An edge case could occur when the setImmediate() in _fatalException()
would fire before the timers module had been loaded globally, causing
Node to crash.

PR-URL: https://github.com/joyent/node/pull/8110
Signed-off-by: Trevor Norris <trev.norris@gmail.com>
Reviewed-by: Fedor Indutny <fedor@indutny.com>
Reviewed-by: Alexis Campailla <alexis@janeasystems.com>
Reviewed-by: Julien Gilli <julien.gilli@joyent.com>

node, async-wrap: remove MakeDomainCallback

C++ won't deoptimize like JS if specific conditional branches are
sporadically met in the future. Combined with the amount of code
duplication removal and simplified maintenance complexity, it makes more
sense to merge MakeCallback and MakeDomainCallback.

Additionally, type casting in V8 before verifying what that type is will
cause V8 to abort in debug mode if that type isn't what was expected.
Fix this by first checking the v8::Value before casting.

PR-URL: https://github.com/joyent/node/pull/8110
Signed-off-by: Trevor Norris <trev.norris@gmail.com>
Reviewed-by: Fedor Indutny <fedor@indutny.com>
Reviewed-by: Alexis Campailla <alexis@janeasystems.com>
Reviewed-by: Julien Gilli <julien.gilli@joyent.com>

async-wrap: move MakeCallback to .cc

MakeCallback is too large a function to be inlined. Likewise, only
having header files will not allow for any part of AsyncWrap to be
exposed cleanly via NODE_MODULE_CONTEXT_AWARE_BUILTIN().

PR-URL: https://github.com/joyent/node/pull/8110
Signed-off-by: Trevor Norris <trev.norris@gmail.com>
Reviewed-by: Fedor Indutny <fedor@indutny.com>
Reviewed-by: Alexis Campailla <alexis@janeasystems.com>
Reviewed-by: Julien Gilli <julien.gilli@joyent.com>

src: remove Async Listener

Async Listener was the name of the user-facing JS API, and is being
completely removed. Instead low level hooks directly into the mechanism
that AL used will be introduced in a future commit.

PR-URL: https://github.com/joyent/node/pull/8110
Signed-off-by: Trevor Norris <trev.norris@gmail.com>
Reviewed-by: Fedor Indutny <fedor@indutny.com>
Reviewed-by: Alexis Campailla <alexis@janeasystems.com>
Reviewed-by: Julien Gilli <julien.gilli@joyent.com>

test: mark test-net-GH-5504 as flaky on linux

That test can trigger a bug on some older Linux kernels.

PR-URL: https://github.com/iojs/io.js/pull/124
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

test: runner support for flaky tests

Adding --flaky-tests option, to allow regarding flaky tests failures
as non-fatal.

Currently only observed by the TapProgressIndicator, which will
add a # TODO directive to tests classified as flaky. According to the
TAP specification, the test harness is supposed to treat failures
that have a # TODO directive as non-fatal.

doc: set logical umask in process.umask example

0644 seems to be the desired mode for new files (as it is a very weird
umask), and to achieve that the correct umask would be 0022.

PR-URL: https://github.com/joyent/node/pull/8039
Reviewed-by: Trevor Norris <trev.norris@gmail.com>

url: change hostname regex to negate invalid chars

Regarding joyent/node#8520

This changes hostname validation from a whitelist regex approach
to a blacklist regex approach as described in https://url.spec.whatwg.org/#host-parsing.

url.parse misinterpreted `https://good.com+.evil.org/`
as `https://good.com/+.evil.org/`.  If we use url.parse to check the
validity of the hostname, the test passes, but in the browser the
user is redirected to the evil.org website.

smalloc: don't allow to dispose typed arrays

PR-URL: https://github.com/joyent/node/pull/8743
Reviewed-by: Trevor Norris <trev.norris@gmail.com>

path: refactor normalizeArray()

The normalizeArray() function now avoids using the slow Array#splice()
method to improve performance and now also filters out empty path parts.

Code that pre-filtered empty parts has been removed.

PR-URL: https://github.com/joyent/node/pull/8724
Reviewed-by: Trevor Norris <trev.norris@gmail.com>

node.cc: use nullptr instead of NULL

PR-URL: https://github.com/iojs/io.js/pull/124
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

node.cc: fix bad assert

It was my mistake to change an assert check. This changes it back to how
the assert was originally done.

Fixes: c131c1f "modules: adding load linked modules feature"
Signed-off-by: Trevor Norris <trev.norris@gmail.com>

uv: float patch to revert tty breakage

Float https://github.com/libuv/libuv/commit/484a3a9 to fix incorrect
indentation in REPL.

crypto: store thread id as pointer-sized value

uv_thread_t is a HANDLE (void pointer) on Windows, which means that
on 64-bit windows it cannot be stored with CRYPTO_THREADID_set_numeric
without potential data loss.

PR-URL: https://github.com/iojs/io.js/pull/124
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

deps: update libuv to 1.0.1

PR-URL: https://github.com/joyent/node/pull/8785
Reviewed-by: Trevor Norris <trev.norris@gmail.com>

openssl: don't define SIXTY_FOUR_BIT_LONG on Windows

On Windows a long integer is always 32-bits, even when the target
architecture uses 64-bit pointers.

PR-URL: https://github.com/iojs/io.js/pull/124
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

docs: fix project name

Renamed node.js to io.js and updated links to external resources.

PR-URL: https://github.com/iojs/io.js/pull/42
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>

Revert "openssl: don't define SIXTY_FOUR_BIT_LONG on Windows"

This reverts commit 878cc3e532ec5feb7145ba0523db2c6383df748b.

Reverted for breaking the x86_64 Linux build:

    In file included from ../deps/openssl/openssl/include/openssl/bn.h:1:0,
                     from ../deps/openssl/openssl/crypto/bn/asm/../bn_lcl.h:115,
                     from ../deps/openssl/openssl/crypto/bn/asm/x86_64-gcc.c:1:
    ../deps/openssl/openssl/include/openssl/../../crypto/bn/bn.h:813:20: note: previous declaration of 'bn_add_words' was here
     BN_ULONG bn_add_words(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,int num);
                        ^
    ../deps/openssl/openssl/crypto/bn/asm/x86_64-gcc.c:210:15: error: conflicting types for 'bn_sub_words'
     BN_ULONG bn_sub_words (BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,int n)

openssl: don't define SIXTY_FOUR_BIT_LONG on Windows

On Windows (and potentially other LP64 platforms), a long integer is
always 32-bits, even when the target architecture uses 64-bit pointers.

Signed-off-by: Bert Belder <bertbelder@gmail.com>

test-require-resolve: use case insensitive compare

The test fixtures directory is derived from the path to the currently
running script, which is itself specified on the command line. That
means that the case of the fixtures dir may not match what the test
expects (when executed on a case-insensitive file system).

PR-URL: https://github.com/iojs/io.js/pull/116
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>