In case of a write failure when using fork() an error would be thrown. The
thrown exception was missing the `errno` property.

- If V8 snapshots are enabled then the hash is only randomized at build time.

---
backport @10366, @10367 and @10402 to 3.6

Add seed to hash of numeric keyed properties.

Minor cleanups of numeric seeded hashing patch.

Split NumberDictionary into a randomly seeded and an unseeded version.
We don't want to randomize the stub cache.

Review URL: http://codereview.chromium.org/9190001/

Fixes #2549.

Make obj.toString and Object.prototype.toString work correctly for typed arrays.

Passing a non-buffer or non-string argument to Socket.prototype.write triggered
an assert:

  Assertion failed: (Buffer::HasInstance(args[0])), function Write,
  file ../src/stream_wrap.cc, line 289.

Fixes #2532.

* Added Docs button and `button` class.
* Refactored download button style into `.button`
* Applied color overrides for download/docs buttons.
* Pointed docs link to latest available docs.

Change udp memory allocation scheme from uv_buf_init to slab allocation. Takes
slab allocation scheme from stream_wrap.

fixes #2521

`punycode` is a third party code which generates a lot of lint errors.
Upstream was contacted in order to fix it in bestiejs/punycode.js#6, but
request was denied.

Therefore, it's reasonable to exclude this file from linting process.

Ref #2456.

* Add assert to prevent parallel writes
* Embed request object instead of using new/delete
* Remove unnecessary WorkReqWrap in favor of uv_work_t
* Use container_of instead of req->data

Along with 2d8af39a and
0ad2717f, this should Fix #2504.

smaller.

Tested in production.

See also http://code.google.com/p/v8/issues/detail?id=1889

The process handle is uninitialized when uv_spawn() fails so don't export the
handle to JS land when that happens. Attempts to close the uninitialized handle
resulted in segmentation faults and memory corruption.

Fixes #2481.

Fixes #2490.

Speeds up the benchmark below by about 680% (0.5s versus 3.4s on my Core 2 Duo).

    var crypto = require('crypto');
    var hash = crypto.createHash('sha1');
    var data = new Buffer(1024);
    for (var i = 0; i < 128 * 1024; ++i) hash.update(data);

Fixes #2494.

Overrides the path to the python binary. Defaults to `python`.

* V8 hash collision fix (Breaks MIPS) (Bert Belder, Erik Corry)

* Upgrade V8 to 3.6.6.15

* Upgrade npm to 1.1.0-beta-10 (isaacs)

* many doc updates (Ben Noordhuis, Jeremy Martin, koichik, Dave Irvine,
  Seong-Rak Choi, Shannen, Adam Malcontenti-Wilson, koichik)

* Fix segfault in node_http_parser.cc

* dgram, timers: fix memory leaks (Ben Noordhuis, Yoshihiro Kukuchi)

* repl: fix repl.start not passing the `ignoreUndefined` arg (Damon Oehlman)

* #1980: Socket.pause null reference when called on a closed Stream (koichik)

* #2263: XMLHttpRequest piped in a writable file stream hang (koichik)

* #2069: http resource leak (koichik)

* buffer.readInt global pollution fix (Phil Sung)

* timers: fix performance regression (Ben Noordhuis)

* #2308, #2246: node swallows openssl error on request (koichik)

* #2114: timers: remove _idleTimeout from item in .unenroll() (James Hartig)

* #2379: debugger: Request backtrace w/o refs (Fedor Indutny)

* simple DTrace ustack helper (Dave Pacheco)

* crypto: rewrite HexDecode without snprintf (Roman Shtylman)

* crypto: don't ignore DH init errors (Ben Noordhuis)

API addition needs to go in master. Also openssl-0.9.8k doesn't have
SSL_CTX_clear_options().

This reverts commit 6f8839d2.

broken in 4a899c92

- If V8 snapshots are enabled then the hash is only randomized at build time.
- Breaks MIPS

---
Backport hash collision workaround to 3.6.
This is made up of 9956, 10351, 10338 and 10330.
This change bakes the string hash key into the snapshot, so
it is determined at build time for shapshot configs.
Review URL: http://codereview.chromium.org/9124004

Refs #2453.

- Save StringPtr if the header hasn't been completely received yet after one
  packet.
- Add one to num_fields and num_values. They were actually one less than the
  number of fields and values.
- Remove always_inline makes debugging difficult, and has negligible
  performance benefits.

SecureContext.setOptions() is backed by SSL_CTX_set_options() which, contrary to
what the name suggests, is additive: it doesn't set options, it adds them to the
already active options.

Hence the need for SecureContext.clearOptions(), which lets you unset active
options.