Consider defaulting to `--debounce=0s` on Linux

[georgefst]

Describe the feature you’d like to be implemented

On trying out ghciwatch for the first time, I was surprised to find that it felt noticeably less snappy than ghcid. It turns out this is because reloads are delayed by half a second for debouncing purposes. This is unnecessary on Linux (and possibly BSDs etc. ?) since inotify reliably sends a single CLOSE_WRITE event when a file is written to disk. From my own past experiments with exec-on-save behaviour, I have found that other common OSs are disappointingly unsophisticated, which is what makes debouncing necessary.

List alternatives to the feature and their pros and cons

• Look in to what exactly ghcid does.
• Reduce the default debounce interval across the board. I seem to remember I tested pretty thoroughly on Windows and MacOS before deciding 0.1s was safe here.

Additional context

From the fact that --debounce=0s works perfectly fine on my Linux machine, in that ghciwatch doesn't attempt to reload twice, I'm assuming that the Rust file watching library being used is already delegating to CLOSE_WRITE on Linux. I haven't actually checked this.

[9999years]

Hmm. I'm not sure about 0 seconds, but 0.1 seconds is probably reasonable. What editor are you using? I found that the filesystem events detected varied a lot based on the editor in use, see my notes here:

ghciwatch/src/event_filter.rs

Lines 26 to 65 in d5b00e3

	Events when I write an existing source file in vim:
	• FileEventKind(Modify(Name(Any))), "/Users/wiggles/mwb4/src/App.hs", file_type: Some(File)
	• FileEventKind(Create(File)), "/Users/wiggles/mwb4/src/App.hs", file_type: Some(File)
	• FileEventKind(Modify(Data(Content))), "/Users/wiggles/mwb4/src/App.hs", file_type: Some(File)
	• FileEventKind(Modify(Metadata(Ownership))), "/Users/wiggles/mwb4/src/App.hs", file_type: Some(File)
	• FileEventKind(Modify(Name(Any))), "/Users/wiggles/mwb4/src/App.hs~", file_type: None
	• FileEventKind(Remove(File)), "/Users/wiggles/mwb4/src/App.hs~", file_type: None
	• FileEventKind(Create(File)), "/Users/wiggles/mwb4/src/4913", file_type: None
	• FileEventKind(Remove(File)), "/Users/wiggles/mwb4/src/4913", file_type: None
	• FileEventKind(Modify(Metadata(Ownership))), "/Users/wiggles/mwb4/src/4913", file_type: None
	^ This is technically a rename due to How It Works. We can tell it's not a "real rename" because
	there's no removed `.hs` files.
	^ I think the only real way to tell if it's a rename or not is to check if the paths are in the
	loaded module set or not?
	Events when I write an existing source file in VSCode:
	• FileEventKind(Modify(Metadata(Any))) "/Users/wiggles/mwb4/src/App.hs" file_type: Some(File)
	• FileEventKind(Modify(Data(Content))) "/Users/wiggles/mwb4/src/App.hs" file_type: Some(File)
	^ A `Modify(Data(Content))` and `Modify(Metadata(_))` and nothing else for that path.
	Events when I create a new file:
	• FileEventKind(Create(Folder)), "/Users/wiggles/mwb4/src/Foo", file_type: Some(Dir)
	• FileEventKind(Create(File)), "/Users/wiggles/mwb4/src/Foo/MyModule.hs", file_type: Some(File)
	• FileEventKind(Modify(Data(Content))), "/Users/wiggles/mwb4/src/Foo/MyModule.hs", file_type: Some(File)
	^ There's a `Create(File)` and `Modify(Data(_))`and nothing else for that path.
	Events when I `rm -rf` a directory:
	• FileEventKind(Remove(File)), "/Users/wiggles/mwb4/src/Foo/MyModule.hs", file_type: None
	• FileEventKind(Remove(Folder)), "/Users/wiggles/mwb4/src/Foo", file_type: None
	^ Note that these both have `file_type: None`.
	^ There's a `Remove(File)` event and nothing else for that path.
	Events when I `mv` a module to a new location:
	• FileEventKind(Create(File)), "/Users/wiggles/mwb4/src/Foo/MyModule.hs", file_type: None
	• FileEventKind(Modify(Data(Content))), "/Users/wiggles/mwb4/src/Foo/MyModule.hs", file_type: None
	• FileEventKind(Modify(Name(Any))), "/Users/wiggles/mwb4/src/Foo/MyModule.hs", file_type: None
	• FileEventKind(Modify(Name(Any))), "/Users/wiggles/mwb4/src/Foo/MyCoolModule.hs", file_type: Some(File)
	^ There's a `Modify(Name(_))` event for the new and old paths. The new path has a `file_type: Some(File)`.
	^ No good way to tell _what_ file it was renamed from.
	^ Why is the old (removed) file marked as `Create(File)`???

[georgefst]

I am now actually noticing occasional interruptions in ghciwatch when using 0s, indicating multiple events arriving in quick succession.

I'm still fairly sure this could be fixed with some filtering, since the behaviour of CLOSE_WRITE shouldn't be editor-dependent. Based on your notes there, I assume that corresponds to FileEventKind(Modify(Data(Content))).

Anyway, 0.1s is probably a nice clean compromise. I'm unlikely to notice such a delay.