-
Notifications
You must be signed in to change notification settings - Fork 36
(Windows) Unknown and unexpected results #93
Comments
Ah, thanks for taking this for a spin! Those results are roughly in line with what I'm seeing here, too.
I do expect some amount of noise in the modification results. Some platforms emit a modification event immediately after a file is created which watcher-stress isn't currently expecting. I've filed #95 to make sure we're being consistent here. It doesn't look like this is happening for you, but I've also seen rapid modifications of the same file be coalesced into a single modification event (MacOS does this pretty aggressively). So some amount of "missed" modifications is not unusual either. 75% seems awfully high though... I'll 🔎 and see if there's something else going on 😄
This is a known platform limitation, although I didn't have an issue for it, yet. I've filed one at #96. Basically,
"Split" renames happen when a rename can't be correlated properly and is reported as separate creation and deletion events. "Half unknown" refers to one of those two events having an unknown kind. So this means that these events are being delivered like:
To contrast, "split exact" events would both be
Alas, we'll never be able to quite get all the way there 😅 There just isn't high enough fidelity in the underlying event stream delivered by the OS. At high enough event rates, we're going to drop some events, too. I've been living on atom/atom#16124's branch for a few days now we already seem to be good enough for a comfortable editing experience, though. We're solidly ahead of |
Ahh, I think Windows is a lot more aggressive about reporting modifications for directories. I'll have to see if I can filter those out somehow. |
Forgot to mention that a |
Prerequisites
Description
When testing using
watcher-stress
, I'm getting quite a bit of unexpected modification results (~75%), half of the renamed results are "split half unknown" (48%), and all deletion results are unknown.Steps to Reproduce
watcher-stress --logging-dir C:\Users\abneyl\watcher-logs\polling\ --root C:\temp --exercise parallel --watcher-count 1
Expected behavior:
Accuracy to be 100% on all types?
Actual behavior:
I got the following output:
Reproduces how often:
Seems to be 100% of the time, at the least 4 different runs all seem to be in the same ballpark.
Versions
@atom/watcher@0.0.7
Additional Information
Attached are zips of the logs generated, as well as the test directory.
watcher-93-root-0.zip
watcher-93-polling.zip
If this should instead be filed over on
watcher-stress
just let me know!The text was updated successfully, but these errors were encountered: