gulp.watch not works if watched glob contains Japanese characters folder name

[aruku7230]

• gulp: 4.00
• node: 8.9.4
• npm: 5.8.0
• OS: macOS 10.12.6

gulpfile.json

const gulp = require('gulp');
const watchGlob = 'folder/**/*';

gulp.task('test', () => {
    gulp.watch(watchGlob, (done) => { done() })
            .on('change', (path) => {
                console.log(`File ${path} changed.`);
            });
});

run test task and change folder/foo file
result output is:

Starting 'test'...
File folder/foo changed.
Starting ''...
Finished '' after 1.45 ms

change watchGlob

const watchGlob = 'フォルダ/**/*';

run test task and change フォルダ/foo file.
result output is:

Starting 'test'...

[rgroothuijsen]

This is most likely a bug in one of the underlying libraries gulp uses for file watching. Interestingly, the フォルダ/** and フォルダ/* glob patterns do react as expected, as does changing a file like フォルダ/foo/bar.

[yocontra]

Can you test it with https://github.com/paulmillr/chokidar directly and see if you can reproduce it? If so the issue goes there, if not we can look into it here.

Thanks for the report.

[aruku7230]

Reproduce it use chokidar. It's a issue with chokidar.

const watchGlob = 'フォルダ/**/*';

gulp.task('test', () => {
    const chokidar = require('chokidar');
    chokidar.watch(watchGlob, (done) => { done() })
            .on('change', (path) => {
                console.log(`File ${path} changed.`);
            });
});

[sttk]

This issue is due to micromatch which is used in chokidar.

[MasatoMakino]

I think it related to NFD filename on macOSX.
NFD's "ダ" is "タ" + " " ".
NFD's "ダ" chrcode deffer from NFC's "ダ" chrcode.

[phated]

We've just added a Bug Bounty of $50 on this issue if anyone wants to solve it.

[pe8ter]

I think I have the issue narrowed down to a RegExp in the nanomatch package. This issue is not restricted to Japanese, but to any directory that doesn't end in 0-9, a-z, A-Z, or _. I opened micromatch/nanomatch#18.

[jonschlinkert]

but to any directory that doesn't end in 0-9, a-z, A-Z, or _.

The regex was written to only match valid path characters, but this clearly was shortsighted and insensitive to non-English languages. I will get that fixed.

[jonschlinkert]

I'd also happily accept a PR to fix it if someone wants to earn the bounty. Just let me know.

[phated]

This is actually going to have to wait until gulp 5.0 to be fixed because it relies on a rewrite of the micromatch parser (which only made sense to be released as a major), and that version will only be used in a version of chokidar that doesn't support all versions of nodejs we support in 4.x