feat: add support for exclusion patterns with '!' prefix

[jordikroon]

Resolve #2095

Add exclusion pattern support to allow users to exclude specific paths from triggering reloads. Patterns prefixed with '!' are now parsed as exclusions and stored separately. The globalWatcher checks if events match any exclusion patterns before processing them, preventing excluded paths from triggering reloads while still allowing them to be matched for filtering purposes.

• Add isExclude field to track exclusion patterns
• Parse '!' prefix during pattern initialization
• Rename allowReload to matchesEvent for clarity on matching logic
• Extract exclusion check into separate allowReload wrapper
• Add isExcludedEvent method to globalWatcher for event filtering
• Filter excluded events before debouncing in event listener

[AlliBalliBaba]

I wonder if it would make more sense to stick to the bash/zsh syntax and do exclusions via !(file1|file2). Something like:

watch "/app/!(public|vendor|.git)/**/*.php"

[jordikroon]

In my honest opinion the (ext)glob syntax is the most superior but the drawback is could also be very complicated.
By only allowing the ! prefix, we make it more similar to how git/eslint ignore files work according to their specification(https://git-scm.com/docs/gitignore). Something that is more self explanatory and intuitive.

Curious to what @dunglas thinks, as he is the initial proposer of this feature.

[AlliBalliBaba]

IMO it's hard to make a pattern matching implementation that's very fast. If speed isn't critical, you can probably just do something similar to matchCurlyBracePattern().

boilerplate (haven't tried):

// check for the following syntax: /path/!(vendor,storage)/
func matchWithExclusions(pattern string, fileName string) bool {
	inclusionAndExclusions := expandExclusions(pattern)
	inclusion := inclusionAndExclusions[0]
	exclusions := inclusionAndExclusions[1:]

	for _, exclusion := range exclusions {
		if matchPattern(exclusion, fileName) {
			return false
		}
	}

	return matchPattern(inclusion, fileName)
}

// !(dir1|dir2)/path -> []string{"*/path", "dir1/path", "dir2/path"}
func expandExclusions(s string) []string {
	before, rest, found := strings.Cut(s, "!(")
	if !found {
		return []string{s}
	}

	inside, after, found := strings.Cut(rest, ")")
	if !found {
		return []string{s} // no closing ')'
	}

	out := expandExclusions(before+"*"+after)
    for _, subPattern := range strings.Split(inside, "|") {
        out = append(out, expandExclusions(before+subPattern+after)...)
    }

	return out
}

[henderkes]

Very silly idea... but couldn't we just offer a regex mode instead?

[jordikroon]

I don't think a regex mode would be achievable without a BC break. Unless we add an extra config option exclude or something.

[henderkes]

We could arbitrarily decide on a pattern like #regex# to use the regex mode and fall back to the regular glob pattern one otherwise. Or a regex_watch directive instead.

[AlliBalliBaba]

Yeah we could also just allow something like this:

watch "regex:^/app/(?!storage/|vendor/)[^/]+(?:/.*)?\.php$"

Just would be nice to have a common enough syntax, so long-term the pattern matching could also be moved to the watcher dependency itself.

[jordikroon]

I like the idea. I will wait a couple of days to give others the opportunity to give their comments then I will implement if nobody objects.

[withinboredom]

Why not use a library? https://github.com/gobwas/glob?

[henderkes]

I'm not opposed to it either, but I think the biggest issue with the current implementation of the globs plus the proposed exclusion pattern is standardization. With regexes everyone know their syntax (bar a few dialect specific details).

I don't know how fast or slow regexes are in Go, though. After profiling our C++ regexes once I nearly had a heart attack.

[dunglas]

I would also prefer switching to a popular library like https://github.com/gobwas/glob.
We should also use the most "standard" glob syntax as possible (the one supported by the library sounds good enough).

Supporting regex could be nice, but it's a bit late, I would prefer having only one way to do the thing.

[dunglas]

See also gobwas/glob#67

[henderkes]

The library doesn't seem that widespread to me and saw its last commit a decade ago.

Supporting regex could be nice, but it's a bit late, I would prefer having only one way to do the thing.

It will only ever get later - I'm typically very happy with using existing libraries, but if they aren't actively maintained and don't support what we need, it's better to switch to a standard approach now, rather than later.

[dunglas]

There is https://github.com/bmatcuk/doublestar, that looks popular and maintained.

[henderkes]

Looks good, but doesn't support exclusions. :/

[henderkes]

e-dant/watcher#103
Perhaps we can just wait for this or help upstream an (existing?) implementation. With it interfacing with C there's got to be a bash globbing library, which would be optimal for people to be familiar with.