Spago run --watch runs(/compiles) twice with vim

[googleson78]

I'm not sure if this is a bug with spago or something stemming from my setup (I don't have the same issue with stack --file-watch though), so sorry in advance if this is a mistake on my part!

Repro steps:

0. spago init
1. spago run --watch
2. After the initial run, save src/Main.purs (even without modifying it). I use vim to do this step.

Expected behaviour:
Files are compiled and run once.

Observed behaviour:
Files are run (and possibly compiled?) twice.

Sample output after a save:

File changed, rebuilding: "/home/googleson78/git/spago-watch-double-compile-repro/src/Main.purs"
Compiling Main
Build succeeded.
🍝
Success! Waiting for next file change.
Type help for available commands. Press enter to force a rebuild.
Build succeeded.
🍝
Success! Waiting for next file change.
Type help for available commands. Press enter to force a rebuild.

spago version is 0.8.5.0

[klntsky]

My neovim does this too. But not nano.
I believe this is related: https://vi.stackexchange.com/questions/11629/set-backupcopy-yes-doesnt-still-writes-the-file-twice

[googleson78]

Thanks @klntsky!

Setting backupcopy=yes resolves this problem, however I am still left wondering what the difference between spago and stacks file change detection is.

[f-f]

I literally copypasted the implementation from Stack's one, so I'm not sure. I'll take a look

[f-f]

@googleson78 which OS are you on? (I'm asking because I'm trying to tweak the debounce logic but it doesn't seem to have any effect on macOS)

[googleson78]

Linux (Debian testing)

[f-f]

@googleson78 could you try the branch from #362?

If that doesn't work you could try tweaking the debounce time here

[klntsky]

May it be so that debounce settings do not apply if a file was deleted and re-created, just like Vim does?

[klntsky]

I tried building from #362 and I still get all File changed, triggering a build messages duplicated when using nvim.

[f-f]

May it be so that debounce settings do not apply if a file was deleted and re-created, just like Vim does?

Oh right, now it makes sense, thanks. Then I cannot figure out why it's different from Stack's behaviour

[klntsky]

Because the code is different (it diverged)

[f-f]

Yeah, though the changes from the original are limited:

• I removed the color stuff
• Generalized from IO to m
• Wired in our glob stuff
• Removed this clearing of the dirtiness flag (so you'd get another rebuild if you changed files while the thing was building)

And this last thing might be the reason for this happening, but I still don't understand why

[f-f]

Update: I tried to apply the last fix I mentioned above, from here, and it didn't help :(
So I'll give up on this, unless someone here has some idea on how to make progress

[klntsky]

I think I can give it a try.

[klntsky]

I see no way other than re-implementing debouncing.

[klntsky]

#371

[f-f]

The reason why this didn't happen with Stack but it happened with Spago just dawned on me: Stack watches single files, but we match file modification events against globs instead

[klntsky]

So what exactly causes doubled event triggering?

[f-f]

@klntsky I did not verify this but I suspect that if you watch the single file you might get less events

[helinko]

Nice to see this fixed! Could we get it released soon?

[f-f]

@helinko yup, I'm going to cut a release in the next days!