If the function for a process.nextTick throws an error, then the
splice() never removes that function from the nextTickQueue array. This
makes sure the functions that have been run in _tickCallback get removed
regardless of errors.
Also add a test for this.
This patch standardises the load order for modules. Highest priority is trying to load exactly the file the user specified, followed by native extensions, followed by registered extra extensions, etc.
In full, if we require('foo') having registered '.coffee' as an alternative extension, we try and load the following files in order:
foo
foo.js
foo.node
foo.coffee
foo/index.js
foo/index.node
foo/index.coffee
This patch replaces the path.exists check for module loading with a call to
fs.statSync (or fs.stat for require.async) which ensures that it's not trying
to load a directory.
Any path.join or path.normalize that starts with a / will not go "above" that after normalization. This is important because /../foo is almost *always* some sort of error, and doesn't match the corollary in sh: `cd $p; pwd`
At the worse, this can be a vector for exploits, since a static file server might do path.join(docroot, path.normalize("/"+req)) to get the file. If the normalized request path could be something like "/../../../etc/passwd" then bad things could happen.
Before there was this comment:
Can't strip trailing slashes since module.js incorrectly
thinks dirname('/a/b/') should yield '/a/b' instead of '/a'.
But now, such thinking is corrected.
- Buffer.toString('ascii', 0, 0) incorrectly returns the entire contents
of the buffer. Fix this.
- Provide similar behavior to Buffer.write() and Buffer.copy() when
dealing with 0-length in valid and invalid byte ranges.
This is ever so slightly less efficient than caching based on ID, since the
filename has to be looked up before we can check the cache. However, it's
the most minimal approach possible to get this change in place. Since
require() is a blocking startup-time operation anyway, a bit of slowness is
not a huge problem.
A test involving require.paths modification and absolute loading. Here's the
gist of it.
Files: /p1/foo.js /p2/foo.js
1. Add "/p1" to require.paths.
2. foo1 = require("foo")
3. assert foo1 === require("/p1/foo") (fail)
4. Remove /p1 from require.paths.
5. Add /p2 to require.paths.
6. foo2 = require("foo")
7. assert foo1 !== foo2 (fail)
8. assert foo2 === require("/p2/foo") (fail)
It's an edge case, but it affects how dependencies are mapped by npm.
If your module requires foo-1.2.3, and my module requires foo-2.3.4,
then you should expect to have require("foo") give you foo-1.2.3, and
I should expect require("foo") to give me foo-2.3.4. However, with
module ID based caching, if your code loads *first*, then your "foo"
is THE "foo", so I'll get your version instead of mine.
It hasn't yet been a problem, but only because there are so few
modules, and everyone pretty much uses the latest version all the
time. But as things start to get to the 1.x and 2.x versions, it'll
be an issue, I'm sure. Dependency hell isn't fun, so this is a way to
avoid it before it strikes.
Add NODE_MODULE_CONTEXTS env var
Only one test was modified to check that this works. NEED to go through all
tests and modify them so that
NODE_MODULE_CONTEXTS=1 make test
passes.
Done by not evaluating the code in the first tick.
This breaks one test in test-error-reporting.js but I believe this to be a
V8 error and I have reported it in
http://code.google.com/p/v8/issues/detail?id=764
- Concatenate 'accept', 'accept-charset', 'accept-encoding',
'accept-language', 'connection', 'cookie', and 'x-*' headers.
- For all others, drop duplicates.
- Adds new dgram module, for all data-gram type transports
- Supports both UDP client and servers
- Supports Unix Daemon sockets in DGRAM mode too (think syslog)
- Uses a shared Buffer and slices that as needed to be reasonably
performant.
- One supplied test program so far, test-dgram-pingpong
- Passes test cases on osx 10.6 and ubuntu 9.10u
a) create a layer of indirection in net.Stream to allow swapping in
different read/write implementations and
b) emit an 'fd' event when file descriptors are received over a UNIX pipe,
as finally as a tangential benefit
c) remove a bunch of conditionals from the primary codepaths for
ease-of-reading.
This patch makes buffers the preferred output for fs.read() and
fs.readSync(). The old string interface is still supported by
converting buffers to strings dynamically. This allows to remove the
C++ code for string handling which is also part of this patch.
This patch makes buffers the preferred output for fs.read() and
fs.readSync(). The old string interface is still supported by
converting buffers to strings dynamically. This allows to remove the
C++ code for string handling which is also part of this patch.
There is a difference between errors which happen to a socket - like
receiving EPIPE - an exceptional situation but ultimately okay and the
situation where code throws in a callback - which is not okay.
Fixes test/simple/test-http-exceptions.js
TODO: explain this in docs.
Previously path.dirname('/tmp') incorrectly returned '.'.
Unfortunately module.js incorrectly thinks dirname('/a/b/') should
yield '/a/b', so I can't strip trailing slashes yet. Once module.js
is fixed, then the commented-out code should be activated and a test
written for it.
This patch makes buffers the preferred input for fs.write() and
fs.writeSync(). The old string interface is still supported by
converting strings to buffers dynamically. This allows to remove the
C++ code for string handling which is also part of this patch.
Taking a performance hit on 'hello world' benchmark by enabling this by
default, but I think it's worth it. Hopefully we can improve performance by
resetting the timeout less often - ideally a 'hello world' benchmark would
only touch the one timer once - if it runs in less than 2 seconds. The rest
should be just link list manipulations.
- setTimeout should active the timeout too. (test-net-set-timeout tests
this.)
- 'timeout' event is not automatically followed by an 'error' event. That
is the user is now responsible for destroying the stream if there is an
idle timeout.
In ab068db9b1 this test was broken because (I
think) compile/run errors are set to crash the program instead of being
passed back.
Error reporting is more important than remote loading. Disabling until there
is a fix
- No more single line "node.js:176:9" errors
- No more strange output when error happens on first line due to
module wrapper function.
- A few tests to check these things
