fs.readSync(0, new Buffer(0), 0, 0, 0) throws Error: Offset is out of bounds

[eahutchins]

Reading 0 bytes into a 0 sized buffer should be a no-op to make it more like read(2):

If count is zero, read() returns zero and has no other results.

[eahutchins]

fs.readSync(0, new Buffer(1), 0, 0, 0) fails with Error: ESPIPE, invalid seek on stdin, but a real empty file works and returns 0.

[glasser]

Just hit this issue as well.

[glasser]

Some notes: fs.read has the same issue. fs.write and fs.writeSync explicitly check for length 0 and do an early return.

http://pubs.opengroup.org/onlinepubs/9699919799/functions/read.html is the POSIX read spec.

[isaacs]

Working as intended. Please see discussion in #5888.

[isaacs]

Ah, ok, so the issue in the OP is still an issue, though.

[glasser]

Yeah, so:

• If we want to support being able to get errors out of read(0), then we can implement that (but you currently can't do that). When I first looked at the code in node_file.cc, I thought that you could just remove the (off >= buffer_length) check (and rely purely on the (off + len > buffer_length) check). But that might run into integer overflow issues, and admittedly the error message is better. Perhaps just change the check to (off > buffer_length)? The idea here would be, it's OK for the block we're trying to read to start anywhere in the buffer, including at the end.
• If we don't want to support that, then the length > 0 check in read is fine.

[trevnorris]

Why wouldn't

offset <= buffer_length
length <= buffer_length
offset + length <= buffer_length

be sufficient?
On Jul 23, 2013 5:29 PM, "David Glasser" [email protected] wrote:

Yeah, so:

If we want to support being able to get errors out of read(0), then we
can implement that (but you currently can't do that). When I first looked
at the code in node_file.cc, I thought that you could just remove the (off

= buffer_length) check (and rely purely on the (off + len >
buffer_length) check). But that might run into integer overflow
issues, and admittedly the error message is better. Perhaps just change the
check to (off > buffer_length)? The idea here would be, it's OK for
the block we're trying to read to start anywhere in the buffer,
including at the end.

If we don't want to support that, then the length > 0 check in read is
fine.

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/5685#issuecomment-21456240
.

[glasser]

That sounds right to me, but I just try not to make assertions about safety in the face of overflow without thinking very carefully :)

[jasnell]

@trevnorris ... any further thoughts on this one?

[trevnorris]

@jasnell If it hasn't been fixed yet then should be easy to do. Just need to change the range checks, and return early if the operation is a noop.