fs.watch doesn't support Chinese characters

[daysv]

info

io.js, v2.3.1
Windows 7, 64bit

[trevnorris]

I just tried this on Linux 3.19 x64 and it ran fine. This may be a windows specific issue.

Can you please post the test script to make sure we're attempting the same thing?

[daysv]

var root = path.join(__dirname,'./doc');
fs.watch(root,function(event,filename){
    console.log(event,filename);
});

Create a new folder and try to rename it to 新建文夹件

[trevnorris]

Can you output filename.length and also new Buffer(filename)?

[daysv]

I can't output them. @trevnorris

[trevnorris]

The command console.log(new Buffer(filename)) doesn't print anything? Make sure it's the only statement in the callback.

[daysv]

sorry.It works and I got them.

[trevnorris]

Can you post the output?

[daysv]

length: 15
buffer: <Buffer c3 a6 c2 96 c2 b0 c3 a5 c2 bb c2 ba c3 a6 c2 96 c2 87 c3 a4 c2 bb c2 b6 c3 a5 c2 a4 c2 b9>

[trevnorris]

Thanks. And yeah. It's borked on Linux too. Didn't test the right thing.

[bnoordhuis]

That's because for better or worse, src/fs_event_wrap.cc encodes the filename as a one-byte (i.e. ISO-8859-1) string. That was changed in 2013 in f674b09, before that the filename was encoded as UTF-8.

On Windows, UTF-8 is the right encoding because libuv decodes the filename to UTF-8 before passing it on to io.js. On Unices, it's complicated; filenames don't have an encoding, they're just byte strings, except on OS X, where they are stored as NFD-normalized UTF-8.

[trevnorris]

@daysv Have a work around for you. Try the following:

fs.watch(PATH, function(event, filename) {
  console.log(Buffer(filename, 'binary').toString());
});

That will reinterpret the one-byte encoded string as UTF8.

[daysv]

@trevnorris Thanks! ^_^

[trevnorris]

@bnoordhuis Thanks for the commit reference and reasoning. Think this, along with the work around, should be documented?

EDIT: Or possibly this work around could be implemented internally to make the difference transparent.

[bnoordhuis]

I think we should fix it but it's part of a broader theme where file path handling is sub-optimal. We probably need some platform-specific code for encoding them to JS strings.

[UncleBill]

I reported same issue at nodejs/node-v0.x-archive#25504.

[seishun]

@bnoordhuis

On Unices, it's complicated; filenames don't have an encoding, they're just byte strings

I don't get it. Doesn't Node.js assume UTF-8 encoding everywhere else?

[UncleBill]

Hi, @seishun , is your patch included in the last release(v5.1.1)? This issue hasn't solved yet.(Tested on windows 10 64 bit)

[seishun]

@UncleBill the fix was postponed to v6.0.0.

[UncleBill]

@seishun Oh, got it. Thanks!

[jasnell]

@trevnorris @bnoordhuis ... thinking about this one further... one possible solution would be to add an encoding option on the fs.watch options. If specified, the filename would be automatically passed as the result of Buffer(filename, 'binary').toString(options.encoding)

[jorangreef]

Thanks @jasnell!