Version `9` strips leading `./`

[romainmenke]

In version 8 :

./gulp/assets/js/*.js

becomes

./gulp/assets/js/app.js

In version 9 :

./gulp/assets/js/*.js

becomes

gulp/assets/js/app.js

---

Is this intentional?
I didn't find anything about it in the Changelog.

To me this seems like a bug because find does it like version 8

;find ./gulp/assets/js/*.js
./gulp/assets/js/app.js

[isaacs]

From the changelog:

• Paths are returned in the canonical formatting for the platform in question.

• Patterns ending in / will still be restricted to matching directories, but will not have a / appended in the results. In general, results will be in their default relative or absolute forms, without any extraneous / and . characters, unlike shell matches. (The mark option may still be used to always mark directory matches with a trailing / or .)

The new algorithm no longer has the ability to track the use of extraneous dot and slash characters, because each path matched is an object in a tree, and patterns are optimized aggressively. This allows for much faster searching (like, 5-500x faster in many cases), but there's no way to know with certainty which form of the pattern matched any given path.

[romainmenke]

Paths are returned in the canonical formatting for the platform in question.

Can you give examples of this?
Is this the direct cause of what I am seeing?

---

Patterns ending in / will still be restricted to matching directories, but will not have a / appended in the results. In general, results will be in their default relative or absolute forms, without any extraneous / and . characters, unlike shell matches. (The mark option may still be used to always mark directory matches with a trailing / or .)

This section seems focussed on path endings, not path prefixes.

[romainmenke]

This allows for much faster searching (like, 5-500x faster in many cases), but there's no way to know with certainty which form of the pattern matched any given path.

I never thought this package was slow, I did however like that it just worked :/

Webpack for example just doesn't work when the leading ./ is omitted.

[msev]

hi @romainmenke

In my webpack configuration i needed to change that :

glob.sync('./dev/scripts/*.js')

by this :

globSync(path.resolve(process.cwd(), './dev/scripts/*.js'))

[romainmenke]

@msev Thank you :)
Good to know that others are facing the same issue in the same context!

I however know how to fix it but I first wanted to know if this was a bug or not.

[geekact]

globSync(path.resolve(process.cwd(), './dev/scripts/*.js'))

@msev maybe this can be globSync(path.resolve('./dev/scripts/*.js')) or globSync(path.join(process.cwd(), './dev/scripts/*.js'))

[isaacs]

If you want to get the absolute paths, you can do globSync('dev/scripts/*.js', { absolute: true }). Using path.resolve() in on the pattern in this way will break on Windows, because the pattern will contain \ characters.

[msev]

thanks @isaacs

[isaacs]

Ah, sorry, @romainmenke, I missed your other message:

Paths are returned in the canonical formatting for the platform in question.

Can you give examples of this? Is this the direct cause of what I am seeing?

So, in a nutshell, this means that the returned paths are the "normal" relative paths (or absolute paths if { absolute: true } is used) for the platform it's running on. So if you do .////foo/./../bar/*.js, then the returned paths on posix platforms will be like bar/baz.js rather than .////foo/./../bar/baz.js. On Windows, they'll be bar\\baz.js, using the platform-specific path separator.

The reason for this is that it's not tracking the full string that it walked down to find the match, it's just doing a walk over the file system and hanging onto the parts of the pattern that are still relevant at any given point. In other words, it's doing a file search, not a string expansion. Doing it in this way is radically more efficient, and makes a lot of weird edge cases easier to handle, plus it makes deduplication tremendously easier, and allows for much more efficient handling of .. and ** within patterns, but it means I dropping the information about extra slashes and dots in the pattern.

Patterns ending in / will still be restricted to matching directories, but will not have a / appended in the results. In general, results will be in their default relative or absolute forms, without any extraneous / and . characters, unlike shell matches. (The mark option may still be used to always mark directory matches with a trailing / or .)

This section seems focussed on path endings, not path prefixes.

This part in the middle:

In general, results will be in their default relative or absolute forms, without any extraneous / and . characters, unlike shell matches.

[romainmenke]

The reason for this is that it's not tracking the full string that it walked down to find the match, it's just doing a walk over the file system and hanging onto the parts of the pattern that are still relevant at any given point. In other words, it's doing a file search, not a string expansion. Doing it in this way is radically more efficient, and makes a lot of weird edge cases easier to handle, plus it makes deduplication tremendously easier, and allows for much more efficient handling of .. and ** within patterns, but it means I dropping the information about extra slashes and dots in the pattern.

@isaacs That's fair.

The issue is also relatively simple to work around.

That others are facing the exact same issue seems to indicate that the ecosystem depends on the old behavior.

Webpack for example could handle foo/bar.js and ./foo/bar.js in the same way, but it doesn't, likely because it didn't need to.

Arguably it might have been better to publish such an extensive rewrite as a new package.
At the same time other packages should not ossify around incorrect assumptions (e.g. relative paths start with ./)

---

Maybe good to call out this case in the changelog with an example?
I would not have bothered to create an issue if I had known this was intended behavior.

[isaacs]

Yeah, as a maintainer, the thing about a community ossifying around your old code is that it's really hard to know when going into a massive overhaul like this exactly what can be discarded and what needs to be preserved, so long-coming much-needed semver major upgrades usually involve a bit of this back and forth. No matter how you try to get the word out there, sometimes the only real way to measure the importance of a feature is by ripping it out, lol

I can see a couple of ways forward (not mutually exclusive).

1. Add a dotRelative option (spelling tbd) which would tell glob to prepend ./ onto any relative path it returns. (I'd push this to be an option in PathScurry, as well.)
2. Yes, I agree this should be called out more clearly in the changelog. It's sort of implicitly there, but clearly not as obvious as it needs to be, I agree.
3. I could track whether a pattern starts with ./, and imply dotRelative like I imply absolute for absolute patterns. But, doing this would entail kind of a lot of extra work and be fairly kludgey. The relative/absolute bitflag would need to be a trinary, minimatch would have to change its pattern optimization, etc. So I'm not very inclined to go that way.
4. Don't add support for it, be intransigent that '' is a perfectly valid relative path, ./foo/bar is the same as foo/bar, etc. That's not really my style, unless there's some compelling reason not to add a given option, but it's worth mentioning. Webpack really shouldn't be treating them differently, but it is what it is.

[isaacs]

Oh, another question, if the result starts with ../ then does it still need the ./ ahead of it? In other words, is ./../foo/bar required, or is ../foo/bar ok with webpack or other tools that require a dot ahead of relative paths?

I'm assuming that the differentiation here stems from the fact that require('foo/bar') is meaningfully different from require('./foo/bar'), in which case ../ would be just as good as ./ for establishing relative-ness.

[isaacs]

@romainmenke Care to try out #500 and see if it solves your issue?

npm i github:isaacs/node-glob#pull/500

And then:

glob(myPattern, { dotRelative: true })
// or whatever function you're using to get glob results