platform/upstream/nodejs.git
11 years agoNow working on 0.9.8
isaacs [Fri, 18 Jan 2013 19:49:35 +0000 (11:49 -0800)]
Now working on 0.9.8

11 years agoMerge branch 'v0.9.7-release'
isaacs [Fri, 18 Jan 2013 19:49:12 +0000 (11:49 -0800)]
Merge branch 'v0.9.7-release'

11 years ago2013.01.18, Version 0.9.7 (Unstable) v0.9.7
isaacs [Fri, 18 Jan 2013 16:42:54 +0000 (08:42 -0800)]
2013.01.18, Version 0.9.7 (Unstable)

* V8: Upgrade to 3.15.11.7

* npm: Upgrade to 1.2.2

* punycode: Upgrade to 1.2.0 (Mathias Bynens)

* repl: make built-in modules available by default (Felix Böhm)

* windows: add support for '_Total' perf counters (Scott Blomquist)

* cluster: make --prof work for workers (Ben Noordhuis)

* child_process: do not keep list of sent sockets (Fedor Indutny)

* tls: Follow RFC6125 more strictly (Fedor Indutny)

* buffer: floating point read/write improvements (Trevor Norris)

* TypedArrays: Improve dataview perf without endian param (Dean McNamee)

* module: assert require() called with a non-empty string (Felix Böhm, James Campos)

* stdio: Set readable/writable flags properly (isaacs)

* stream: Properly handle large reads from push-streams (isaacs)

11 years agodoc: Remove mention of child.send 'track' option
isaacs [Fri, 18 Jan 2013 18:17:26 +0000 (10:17 -0800)]
doc: Remove mention of child.send 'track' option

Will be removed very soon.  No point making it public.

11 years agobuffer: Define INFINITY for MSVC compiler
isaacs [Fri, 18 Jan 2013 18:03:54 +0000 (10:03 -0800)]
buffer: Define INFINITY for MSVC compiler

11 years agonpm: Upgrade to 1.2.2
isaacs [Fri, 18 Jan 2013 16:37:32 +0000 (08:37 -0800)]
npm: Upgrade to 1.2.2

11 years agocluster: make --prof work for workers
Ben Noordhuis [Fri, 18 Jan 2013 11:44:10 +0000 (12:44 +0100)]
cluster: make --prof work for workers

Profiling in clustered environments doesn't work out of the box.

By default, V8 writes the profile data of all processes to a single
v8.log.

Running that log file through a tick processor produces bogus numbers
because many events won't match up with the recorded memory mappings
and you end up with graphs where 80+% of ticks is unaccounted for.

Fixing the tick processor to deal with multi-process output is not very
useful because the processes may be running wildly disparate workloads.

That's why we fix up the command line arguments to include
a "--logfile=v8-%p.log" argument (where %p is expanded to the PID)
unless it already contains a --logfile argument.

Fixes #4617.

11 years agov8: add %p option to --logfile, expands to pid
Ben Noordhuis [Fri, 18 Jan 2013 00:23:34 +0000 (01:23 +0100)]
v8: add %p option to --logfile, expands to pid

Useful when you are profiling multiple instances of V8 concurrently.
Submitted upstream: http://code.google.com/p/v8/issues/detail?id=2503

Refs #4617.

11 years agochild_process: do not keep list of sent sockets
Fedor Indutny [Mon, 14 Jan 2013 17:08:20 +0000 (21:08 +0400)]
child_process: do not keep list of sent sockets

Keeping list of all sockets that were sent to child process causes memory
leak and thus unacceptable (see #4587). However `server.close()` should
still work properly.

This commit introduces two options:

* child.send(socket, { track: true }) - will send socket and track its status.
  You should use it when you want to receive `close` event on sent sockets.
* child.send(socket) - will send socket without tracking it status. This
  performs much better, because of smaller number of RTT between master and
  child.

With both of these options `server.close()` will wait for all sent
sockets to get closed.

11 years agoRevert "child_process: do not keep list of sent sockets"
Fedor Indutny [Thu, 17 Jan 2013 23:13:10 +0000 (03:13 +0400)]
Revert "child_process: do not keep list of sent sockets"

This reverts commit db5ee0b3decace9b5d80ee535ce53183aff02909.

11 years agobenchmark: Add resume() in static_http_server
isaacs [Thu, 17 Jan 2013 22:54:59 +0000 (14:54 -0800)]
benchmark: Add resume() in static_http_server

11 years agochild_process: do not keep list of sent sockets
Fedor Indutny [Mon, 14 Jan 2013 17:08:20 +0000 (21:08 +0400)]
child_process: do not keep list of sent sockets

Keeping list of all sockets that were sent to child process causes memory
leak and thus unacceptable (see #4587). However `server.close()` should
still work properly.

This commit introduces two options:

* child.send(socket, { track: true }) - will send socket and track its status.
  You should use it when you want `server.connections` to be a reliable
  number, and receive `close` event on sent sockets.
* child.send(socket) - will send socket without tracking it status. This
  performs much better, because of smaller number of RTT between master and
  child.

With both of these options `server.close()` will wait for all sent
sockets to get closed.

11 years agoRevert "events: Don't crash on events named __proto__"
isaacs [Thu, 17 Jan 2013 21:20:22 +0000 (13:20 -0800)]
Revert "events: Don't crash on events named __proto__"

Unfortunately, it's just too slow to do this in events.js.  Users will
just have to live with not having events named __proto__ or toString.

This reverts commit b48e303af023dc60b6f6590694ba94d9b8108702.

11 years agostdio: Set readable/writable flags properly
isaacs [Thu, 17 Jan 2013 17:52:48 +0000 (09:52 -0800)]
stdio: Set readable/writable flags properly

Set the readable/writable flags properly in net streams that have
a handle passed in (such as TTY streams).

Fix #4606

11 years agoevents: Don't crash on events named __proto__
isaacs [Wed, 16 Jan 2013 19:26:28 +0000 (11:26 -0800)]
events: Don't crash on events named __proto__

This prefixes all event names internally with 'ev'.

11 years agodoc: modify documentation style about 'Stability: 5'
Ryunosuke SATO [Thu, 17 Jan 2013 09:12:03 +0000 (18:12 +0900)]
doc: modify documentation style about 'Stability: 5'

'Stability: 5' is described as 'Locked' not as 'API Locked'
in other documents.

For example:
- `/doc/api/assert.markdown`
- `/doc/api/util.markdown`

This word was injected in 192192a.

11 years agodeps: upgrade libuv to 4ba03dd
Ben Noordhuis [Wed, 16 Jan 2013 22:38:02 +0000 (23:38 +0100)]
deps: upgrade libuv to 4ba03dd

11 years agotest: add module-loading test with empty string
isaacs [Wed, 16 Jan 2013 20:27:53 +0000 (12:27 -0800)]
test: add module-loading test with empty string

11 years agomodule: assert that require() is called with a string
Felix Böhm [Wed, 16 Jan 2013 18:53:16 +0000 (19:53 +0100)]
module: assert that require() is called with a string

as requested in #4577

11 years agolint: Prefer double-quotes over single
isaacs [Wed, 16 Jan 2013 20:07:12 +0000 (12:07 -0800)]
lint: Prefer double-quotes over single

11 years agorepl: fix how to module requiring in code comment
Ryunosuke SATO [Tue, 11 Dec 2012 10:13:22 +0000 (19:13 +0900)]
repl: fix how to module requiring in code comment

This module requiring style is old.
This API has been changed in Node 0.1.16 726865af.

11 years agoTypedArrays: Improve dataview perf without endian param
Dean McNamee [Tue, 15 Jan 2013 18:35:52 +0000 (19:35 +0100)]
TypedArrays: Improve dataview perf without endian param

V8 seems to be particularly slow converting an undefined value to false
in BooleanValue.

Revert this when we upgrade to V8 3.17, or whenever the fix discussed
in http://code.google.com/p/v8/issues/detail?id=2487 lands in V8.

11 years agoassert that require() has a truthy path
James Campos [Wed, 9 Jan 2013 17:01:22 +0000 (09:01 -0800)]
assert that require() has a truthy path

11 years agostream: Properly handle large reads from push-streams
isaacs [Wed, 16 Jan 2013 00:44:29 +0000 (16:44 -0800)]
stream: Properly handle large reads from push-streams

Problem 1: If stream.push() triggers a 'readable' event, and the user
calls `read(n)` with some n > the highWaterMark, then the push() will
return false (indicating that they should not push any more), but no
future 'readable' event is coming (because we're above the
highWaterMark).

Solution: return true from push() when needReadable is set.

Problem 2: A read(n) for n != 0, after the stream had encountered an
EOF, would not trigger the 'end' event if the EOF was pushed in
synchronously by the _read() function.

Solution: Check for ended in stream.read() and schedule an end event if
the length now equals 0.

Fix #4585

11 years agobuffer: improve read/write speed with assert
Trevor Norris [Fri, 11 Jan 2013 00:59:39 +0000 (16:59 -0800)]
buffer: improve read/write speed with assert

Improved assert check order of execution and added additional checks on
parameters to ensure no bad values make it through (e.g. negative offset
values).

11 years agobuffer: floating point read/write improvements
Trevor Norris [Sat, 22 Dec 2012 21:06:50 +0000 (13:06 -0800)]
buffer: floating point read/write improvements

Improvements:
* floating point operations are approx 4x's faster
* Now write quiet NaN's
* all read/write on floating point now done in C, so no more need for
  lib/buffer_ieee754.js
* float values have more accurate min/max value checks
* add additional benchmarks for buffers read/write
* created benchmark/_bench_timer.js which is a simple library that
  can be included into any benchmark and provides an intelligent tracker
  for sync and async tests
* add benchmarks for DataView set methods
* add checks and tests to make sure offset is greater than 0

11 years agohttp: fix duplicate var initialization
Alexandr Emelin [Thu, 10 Jan 2013 07:38:29 +0000 (11:38 +0400)]
http: fix duplicate var initialization

IncomingMessage function contained duplicate initialization
of this._pendings. Line with one of those expressions has been
removed.

11 years agopunycode: update to v1.2.0
Mathias Bynens [Fri, 11 Jan 2013 08:57:24 +0000 (09:57 +0100)]
punycode: update to v1.2.0

This update adds support for RFC 3490 separators for improved
compatibility with web browsers.

11 years agov8: work around String::WriteAscii segfault
Ben Noordhuis [Tue, 15 Jan 2013 22:42:23 +0000 (23:42 +0100)]
v8: work around String::WriteAscii segfault

See http://code.google.com/p/v8/issues/detail?id=2493 for details.
Once the patch lands in upstream V8, this commit can be reverted.

11 years agotls: follow RFC6125 more stricly
Fedor Indutny [Mon, 14 Jan 2013 23:29:46 +0000 (03:29 +0400)]
tls: follow RFC6125 more stricly

* Allow wildcards only in left-most part of hostname identifier.
* Do not match CN if altnames are present

11 years agostreams2: Do not allow hwm < lwm
isaacs [Tue, 15 Jan 2013 00:03:38 +0000 (16:03 -0800)]
streams2: Do not allow hwm < lwm

There was previously an assert() in there, but this part of the code is
so high-volume that the added cost made a measurable dent in http_simple.

Just checking inline is fine, though, and prevents a lot of potential
hazards.

11 years agostream: Do not call endReadable on a non-empty stream
isaacs [Mon, 14 Jan 2013 19:25:39 +0000 (11:25 -0800)]
stream: Do not call endReadable on a non-empty stream

Say that a stream's current read queue has 101 bytes in it, and the
underlying resource has ended (ie, reached EOF).

If you do something like this:

    stream.read(100); // leave a byte behind
    stream.read(0); // read(0) for some reason

then the read(0) will get 0 from the howMuchToRead function.  Since the
stream was ended, this was incorrectly treating the 0 as a "there is no
more in the buffer", and emitting 'end' before that last byte was read.

Why have the read(0) in the first place?  We do this in some cases to
trigger the last few bytes of a net socket (such as a child process's
stdio pipes).  This was causing issues when piping a `git archive` job
to a file: the resulting tarball was incomplete, because it occasionally
was not getting the last chunk.

11 years agobuffer: zero-length buffers shouldn't be slab-backed
Bert Belder [Sat, 5 Jan 2013 21:14:50 +0000 (22:14 +0100)]
buffer: zero-length buffers shouldn't be slab-backed

11 years agotls: allow wildcards in common name
Fedor Indutny [Mon, 14 Jan 2013 17:09:35 +0000 (21:09 +0400)]
tls: allow wildcards in common name

see #4592

11 years agogitignore: ignore perf data files
Ben Noordhuis [Mon, 14 Jan 2013 12:06:29 +0000 (13:06 +0100)]
gitignore: ignore perf data files

11 years agov8: Hardfloat does not imply VFPv3, only VFPv2.
yangguo@chromium.org [Tue, 18 Dec 2012 09:48:51 +0000 (09:48 +0000)]
v8: Hardfloat does not imply VFPv3, only VFPv2.

Raspberry Pi is an example.

BUG=v8:2393

Review URL: https://chromiumcodereview.appspot.com/11570061
Patch from Chi-Thanh Christopher Nguyen <nguyenchithanh@gmail.com>.

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@13232 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

This is a backport of v8/v8@44419ad089b729a1aab4a5f27c0833e2ce69d28b.

11 years agodeps: upgrade libuv to 8e3e60f
Ben Noordhuis [Sun, 13 Jan 2013 00:29:34 +0000 (01:29 +0100)]
deps: upgrade libuv to 8e3e60f

11 years agorepl: fix lint
Nathan Rajlich [Sat, 12 Jan 2013 20:14:39 +0000 (12:14 -0800)]
repl: fix lint

11 years agorepl: make built-in modules available by default
Felix Böhm [Sat, 12 Jan 2013 20:07:06 +0000 (12:07 -0800)]
repl: make built-in modules available by default

Closes #3564.
Closes #4578.

11 years agowindows: add support for '_Total' perf counters
Scott Blomquist [Fri, 11 Jan 2013 21:24:15 +0000 (22:24 +0100)]
windows: add support for '_Total' perf counters

11 years agov8: remove optimization switches
Ben Noordhuis [Thu, 3 Jan 2013 01:33:40 +0000 (02:33 +0100)]
v8: remove optimization switches

Remove compiler switches from $(TOPLEVEL)/deps/v8/build/common.gypi,
we set them globally in $(TOPLEVEL)/common.gypi.

Commit 7b4d95a introduced the switches again, resulting in V8 getting
built without any optimizations.

This commit is essentially a rehash of commit 4b8629d.

11 years agodeps: update v8 to 3.15.11.7
Fedor Indutny [Fri, 11 Jan 2013 20:10:45 +0000 (00:10 +0400)]
deps: update v8 to 3.15.11.7

11 years agoNow working on 0.9.7
isaacs [Fri, 11 Jan 2013 19:25:54 +0000 (11:25 -0800)]
Now working on 0.9.7

11 years agoMerge branch 'v0.9.6-release'
isaacs [Fri, 11 Jan 2013 19:25:37 +0000 (11:25 -0800)]
Merge branch 'v0.9.6-release'

11 years ago2013.01.11, Version 0.9.6 (Unstable) v0.9.6
isaacs [Fri, 11 Jan 2013 18:49:21 +0000 (10:49 -0800)]
2013.01.11, Version 0.9.6 (Unstable)

* V8: update to 3.15.11.5

* node: remove ev-emul.h (Ben Noordhuis)

* path: make basename and extname ignore trailing slashes (Bert Belder)

* typed arrays: fix sunos signed/unsigned char issue (Ben Noordhuis)

* child_process: Fix {stdio:'inherit'} regression (Ben Noordhuis)

* child_process: Fix pipe() from child stdio streams  (Maciej Małecki)

* child_process: make fork() execPath configurable (Bradley Meck)

* stream: Add readable.push(chunk) method (isaacs)

* dtrace: x64 ustack helper (Fedor Indutny)

* repl: fix floating point number parsing (Nirk Niggler)

* repl: allow overriding builtins (Ben Noordhuis)

* net: add localAddress and localPort to Socket (James Hight)

* fs: make pool size coincide with ReadStream bufferSize (Shigeki Ohtsu)

* typed arrays: implement load and store swizzling (Dean McNamee)

* windows: fix perfctr crash on XP and 2003 (Scott Blomquist)

* dgram: fix double implicit bind error (Ben Noordhuis)

11 years agoMerge remote-tracking branch 'ry/v0.8' into master
isaacs [Fri, 11 Jan 2013 18:18:07 +0000 (10:18 -0800)]
Merge remote-tracking branch 'ry/v0.8' into master

Conflicts:
AUTHORS
ChangeLog
Makefile
src/node_version.h

11 years agopath: make basename and extname ignore trailing slashes
Bert Belder [Fri, 11 Jan 2013 17:42:46 +0000 (18:42 +0100)]
path: make basename and extname ignore trailing slashes

Fixes #4536

11 years agotyped arrays: fix sunos signed/unsigned char issue
Ben Noordhuis [Fri, 11 Jan 2013 17:31:47 +0000 (18:31 +0100)]
typed arrays: fix sunos signed/unsigned char issue

The int8_t and uint8_t typedefs on sunos/smartos depend on a number of
compiler directives. Avoid ambiguity and specify signed and unsigned
char explicitly.

Fixes the following build error:

  ../src/stream_wrap.cc: In static member function 'static void*
  node::WriteWrap::operator new(size_t)':
  ../src/stream_wrap.cc:70:49: warning: no return statement in function
  returning non-void [-Wreturn-type]
  In file included from ../src/v8_typed_array.cc:26:0:
  ../src/v8_typed_array_bswap.h: In function 'T
  v8_typed_array::SwapBytes(T) [with T = signed char]':
  ../src/v8_typed_array_bswap.h:150:23:   instantiated from 'T
  v8_typed_array::LoadAndSwapBytes(void*) [with T = signed char]'
  ../src/v8_typed_array.cc:694:7:   instantiated from 'static
  v8::Handle<v8::Value> {anonymous}::DataView::getGeneric(const
  v8::Arguments&) [with T = signed char]'
  ../src/v8_typed_array.cc:738:40:   instantiated from here
  ../src/v8_typed_array_bswap.h:125:16: error: size of array is
  negative

11 years agotest: child process { stdio:'inherit' }
isaacs [Fri, 11 Jan 2013 01:15:53 +0000 (17:15 -0800)]
test: child process { stdio:'inherit' }

11 years agoRevert "typed arrays: fix build on sunos"
Fedor Indutny [Fri, 11 Jan 2013 17:01:14 +0000 (21:01 +0400)]
Revert "typed arrays: fix build on sunos"

This reverts commit 35a137cf8dcec75fd4cc11bbd8eb22cc2c1d7ece.

11 years agotyped arrays: fix build on sunos
Fedor Indutny [Fri, 11 Jan 2013 15:52:49 +0000 (19:52 +0400)]
typed arrays: fix build on sunos

11 years agoRevert "path: fix bugs related to paths with trailing slashes"
Bert Belder [Fri, 11 Jan 2013 15:19:21 +0000 (16:19 +0100)]
Revert "path: fix bugs related to paths with trailing slashes"

It only does the right thing on unix, and the implementation
isn't great either.

This reverts commit bb1c03989f8702e06072e6d9228b52661bf00ace.

11 years agodeps: upgrade libuv to 9aab5d4
Ben Noordhuis [Fri, 11 Jan 2013 12:49:45 +0000 (13:49 +0100)]
deps: upgrade libuv to 9aab5d4

11 years agohttp: Set _dumped=false initially
isaacs [Fri, 11 Jan 2013 02:16:43 +0000 (18:16 -0800)]
http: Set _dumped=false initially

The better to keep the IncomingMessage class isomorphic and avoid
creating additional hidden classes.

11 years agopath: fix bugs related to paths with trailing slashes
Andreas Madsen [Tue, 8 Jan 2013 17:22:04 +0000 (18:22 +0100)]
path: fix bugs related to paths with trailing slashes

11 years agoconfigure: define "arm_fpu" and "arm_neon" for ARM
Nathan Rajlich [Fri, 11 Jan 2013 01:13:59 +0000 (17:13 -0800)]
configure: define "arm_fpu" and "arm_neon" for ARM

v8's common.gypi file expects them to be defined now.

Closes #4534.

11 years agomake binary: allow custom config flags
Nathan Rajlich [Fri, 11 Jan 2013 00:59:09 +0000 (16:59 -0800)]
make binary: allow custom config flags

For example, to cross-compile from my OS X laptop for Raspberry Pi, you would
do something like:

  $ make binary BINARYNAME=node-v`python tools/getnodeversion.py`-linux-arm-pi \
      DESTCPU=arm CONFIG_FLAGS="--dest-os=linux"

11 years agov8: Hardfloat does not imply VFPv3, only VFPv2.
yangguo@chromium.org [Tue, 18 Dec 2012 09:48:51 +0000 (09:48 +0000)]
v8: Hardfloat does not imply VFPv3, only VFPv2.

Raspberry Pi is an example.

BUG=v8:2393

Review URL: https://chromiumcodereview.appspot.com/11570061
Patch from Chi-Thanh Christopher Nguyen <nguyenchithanh@gmail.com>.

git-svn-id: https://v8.googlecode.com/svn/branches/bleeding_edge@13232 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

This is a backport of v8/v8@44419ad089b729a1aab4a5f27c0833e2ce69d28b.

11 years agohttp: Don't switch the socket into old-mode
isaacs [Tue, 8 Jan 2013 04:33:56 +0000 (20:33 -0800)]
http: Don't switch the socket into old-mode

11 years agohttp: Use stream.push() instead of touching _readableState
isaacs [Tue, 8 Jan 2013 04:23:07 +0000 (20:23 -0800)]
http: Use stream.push() instead of touching _readableState

11 years agonet: Use readable.push() instead of private methods
isaacs [Tue, 8 Jan 2013 02:07:37 +0000 (18:07 -0800)]
net: Use readable.push() instead of private methods

11 years agostream: Use push() for Transform._output()
isaacs [Thu, 10 Jan 2013 21:49:19 +0000 (13:49 -0800)]
stream: Use push() for Transform._output()

This also slightly changes the semantics, in that a 'readable'
event may be triggered by the first write() call, even if a
user has not yet called read().

This happens because the Transform _write() handler is calling
read(0) to start the flow of data.  Technically, the new behavior
is more 'correct', since it is more in line with the semantics
of the 'readable' event in other streams.

11 years agostream: Use push() for readable.wrap()
isaacs [Tue, 8 Jan 2013 03:40:08 +0000 (19:40 -0800)]
stream: Use push() for readable.wrap()

11 years agostream: Add readable.push(chunk) method
isaacs [Tue, 8 Jan 2013 02:07:17 +0000 (18:07 -0800)]
stream: Add readable.push(chunk) method

11 years agoblog, changelog: 0.8.17 was released in 2013, not 2012
isaacs [Thu, 10 Jan 2013 15:47:09 +0000 (07:47 -0800)]
blog, changelog: 0.8.17 was released in 2013, not 2012

11 years agoblog: Correct shasums for v0.8.17
isaacs [Thu, 10 Jan 2013 03:05:37 +0000 (19:05 -0800)]
blog: Correct shasums for v0.8.17

11 years agoblog: Add security notice to v0.8.17 post
isaacs [Thu, 10 Jan 2013 01:21:16 +0000 (17:21 -0800)]
blog: Add security notice to v0.8.17 post

11 years agoblog: Post about v0.8.17
isaacs [Thu, 10 Jan 2013 01:08:24 +0000 (17:08 -0800)]
blog: Post about v0.8.17

11 years agoNow working on 0.8.18
isaacs [Thu, 10 Jan 2013 01:07:33 +0000 (17:07 -0800)]
Now working on 0.8.18

11 years agoMerge branch 'v0.8.17-release' into v0.8
isaacs [Thu, 10 Jan 2013 01:07:21 +0000 (17:07 -0800)]
Merge branch 'v0.8.17-release' into v0.8

11 years agotest: run tests in alphabetical order
Ben Noordhuis [Thu, 10 Jan 2013 00:58:30 +0000 (01:58 +0100)]
test: run tests in alphabetical order

11 years ago2012.01.09, Version 0.8.17 (Stable) v0.8.17
isaacs [Thu, 10 Jan 2013 00:04:21 +0000 (16:04 -0800)]
2012.01.09, Version 0.8.17 (Stable)

* npm: Upgrade to v1.2.0
  - peerDependencies (Domenic Denicola)
  - node-gyp v0.8.2 (Nathan Rajlich)
  - Faster installs from github user/project shorthands (Nathan Zadoks)

* typed arrays: fix 32 bit size/index overflow (Ben Noordhuis)

* http: Improve performance of single-packet responses (Ben Noordhuis)

* install: fix openbsd man page location (Ben Noordhuis)

* http: bubble up parser errors to ClientRequest (Brian White)

11 years agotyped arrays: fix 32 bit size/index overflow
Ben Noordhuis [Wed, 9 Jan 2013 23:39:44 +0000 (00:39 +0100)]
typed arrays: fix 32 bit size/index overflow

Fix an out-of-bound read/write bug due to integer wrapping. Reported by
Dean McNamee.

11 years agonpm: Upgrade to v1.2.0
isaacs [Wed, 9 Jan 2013 23:21:30 +0000 (15:21 -0800)]
npm: Upgrade to v1.2.0

11 years agotyped arrays: implement load and store swizzling
Dean McNamee [Sun, 6 Jan 2013 17:59:12 +0000 (18:59 +0100)]
typed arrays: implement load and store swizzling

Implement load and store swizzling operations. This reduces an unneeded
back and forth between types and additionally keeps the value in the
swappable type until it is swapped. This is important for correctness
when dealing with floating point, to avoid the possibility of loading
the bits of a signaling NaN (because it isn't yet swapped) into the FPU.

This additionally produces better code (comments are mine):

gcc version 4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2336.11.00)

setValue<double>:
  movd  %xmm0, %rax         ; fp reg -> gen reg
  bswapq  %rax              ; 64-bit byte swap
  movq  %rax, (%r15,%r12)   ; store

11 years agotyped arrays: swizzle with compiler intrinsics
Dean McNamee [Sat, 5 Jan 2013 21:06:35 +0000 (22:06 +0100)]
typed arrays: swizzle with compiler intrinsics

Implement swizzling with compiler intrinsics and be aware of the native
endianness to correctly swap on big endian machines.

This introduces a template function to swap the bytes of a value,
and macros for the low level swap (taking advantage of gcc and msvc
intrinsics). This produces code like the following (comments are mine):

gcc version 4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2336.11.00)

setValue<double>:
  movd  %xmm0, %rax         ; fp reg -> gen reg
  bswapq  %rax              ; 64-bit byte swap
  movd  %rax, %xmm0         ; gen reg -> fp reg
  movq  %xmm0, (%r15,%r12)  ; store

11 years agotyped arrays: undo local changes
Ben Noordhuis [Tue, 8 Jan 2013 16:08:58 +0000 (17:08 +0100)]
typed arrays: undo local changes

Partially revert 5664dd2, 6573fc3 and 7788a6b to ease landing changes
from upstream.

11 years agoobject_wrap: add missing HandleScope
Fedor Indutny [Fri, 4 Jan 2013 12:37:38 +0000 (16:37 +0400)]
object_wrap: add missing HandleScope

by Sven Panne's suggestion, see [0] for details.

[0]: https://code.google.com/p/v8/issues/detail?id=2468

11 years agostream: fix typo
Shigeki Ohtsu [Wed, 9 Jan 2013 02:17:06 +0000 (11:17 +0900)]
stream: fix typo

Also: Revert "stream: Override addListener as well as on"

This reverts commit 18c985919d93df97561157ab8e7de70f4e22dd44.

11 years agostream: Override addListener as well as on
isaacs [Tue, 8 Jan 2013 22:27:01 +0000 (14:27 -0800)]
stream: Override addListener as well as on

For the compatibility switch.

11 years agostreams: set listener first, emit 'data' later
Fedor Indutny [Tue, 8 Jan 2013 17:56:55 +0000 (21:56 +0400)]
streams: set listener first, emit 'data' later

When switching into compatibility mode by setting `data` event listener,
`_read()` method will be called immediately. If method implementation
invokes callback in the same tick - all emitted `data` events will be
discarded, because `data` listener wasn't set yet.

11 years agodoc: os.cpus() returns times in milliseconds
Tim Bradshaw [Mon, 31 Dec 2012 18:44:51 +0000 (18:44 +0000)]
doc: os.cpus() returns times in milliseconds

11 years agotyped arrays: don't declare as module
Dean McNamee [Sun, 23 Dec 2012 22:43:51 +0000 (23:43 +0100)]
typed arrays: don't declare as module

TypedArrays is not a module, it is attached to the global object.
Don't register it with Node's module system.

11 years agotyped arrays: replace switch with templates
Dean McNamee [Tue, 18 Dec 2012 19:03:41 +0000 (20:03 +0100)]
typed arrays: replace switch with templates

Convert TypedArray's TypeName implementation to template specialization
instead of a switch() statement.

11 years agonet: fix bufferSize include writableStream length
Shigeki Ohtsu [Mon, 7 Jan 2013 01:37:35 +0000 (10:37 +0900)]
net: fix bufferSize include writableStream length

socket.bufferSize missed to include the length of internal buffers in
writableStream.

11 years agodgram: don't assert on send('string')
Ben Noordhuis [Tue, 8 Jan 2013 03:27:34 +0000 (04:27 +0100)]
dgram: don't assert on send('string')

Raise a TypeError when the argument to send() or sendto() is anything
but a Buffer.

Fixes the following assertion:

  $ node -e 'require("dgram").createSocket("udp4").send("BAM")'
  node: ../../src/udp_wrap.cc:220: static v8::Handle<v8::Value>
  node::UDPWrap::DoSend(const v8::Arguments&, int): Assertion
  `Buffer::HasInstance(args[0])' failed.
  Aborted (core dumped)

Fixes #4496.

11 years agobuild: fail w/err msg when missing binutils
Mike Harsch [Mon, 7 Jan 2013 05:29:38 +0000 (22:29 -0700)]
build: fail w/err msg when missing binutils

Building --with-dtrace requires objdump from GNU binutils.
This change inserts a helpful error message if there is a
problem executing objdump.

11 years agodtrace: More style
isaacs [Mon, 7 Jan 2013 20:31:44 +0000 (12:31 -0800)]
dtrace: More style

Continuation lines should be indented with 4 spaces, not a tab.

11 years agodtrace: Make D style more D-ish
isaacs [Mon, 7 Jan 2013 20:20:21 +0000 (12:20 -0800)]
dtrace: Make D style more D-ish

11 years agodtrace: x64 ustack helper
Fedor Indutny [Thu, 27 Dec 2012 03:16:23 +0000 (07:16 +0400)]
dtrace: x64 ustack helper

11 years agodtrace: fix style in ustack helper
Fedor Indutny [Mon, 7 Jan 2013 18:53:50 +0000 (22:53 +0400)]
dtrace: fix style in ustack helper

11 years agodtrace: SeqAsciiString was renamed to SeqOneByteString in v8
isaacs [Mon, 7 Jan 2013 18:30:15 +0000 (10:30 -0800)]
dtrace: SeqAsciiString was renamed to SeqOneByteString in v8

11 years agosrc: pass node_isolate to Undefined()
Ben Noordhuis [Sun, 6 Jan 2013 23:15:59 +0000 (00:15 +0100)]
src: pass node_isolate to Undefined()

11 years agosrc: pass node_isolate to Null()
Ben Noordhuis [Sun, 6 Jan 2013 22:55:56 +0000 (23:55 +0100)]
src: pass node_isolate to Null()

11 years agosrc: pass node_isolate to True() and False()
Ben Noordhuis [Sun, 6 Jan 2013 23:09:31 +0000 (00:09 +0100)]
src: pass node_isolate to True() and False()

11 years agosrc: pass node_isolate to Local<>::New
Ben Noordhuis [Sun, 6 Jan 2013 22:52:35 +0000 (23:52 +0100)]
src: pass node_isolate to Local<>::New

11 years agosrc: pass node_isolate to Integer::New
Ben Noordhuis [Sun, 6 Jan 2013 22:33:09 +0000 (23:33 +0100)]
src: pass node_isolate to Integer::New

11 years agosrc: pass node_isolate to String::Empty
Ben Noordhuis [Sun, 6 Jan 2013 22:09:39 +0000 (23:09 +0100)]
src: pass node_isolate to String::Empty

11 years agosrc: make node_isolate global
Ben Noordhuis [Sun, 6 Jan 2013 22:06:48 +0000 (23:06 +0100)]
src: make node_isolate global

V8 3.15 has new API functions that let you specify the Isolate. V8 and
node.js generally spend 0.5-3.5% of the time in pthread_getspecific(),
looking up the current Isolate. Avoid that overhead by making "our"
isolate global so we can pass it around. The change to the new API is
introduced in follow-up commits.

11 years agochild_process: Pull through untouched stdio streams
isaacs [Mon, 7 Jan 2013 14:29:00 +0000 (06:29 -0800)]
child_process: Pull through untouched stdio streams

Otherwise child procs will never emit a 'close' event if you don't
ever consume their streams, because they will never hit the EOF.