search

Acly / krita-ai-diffusion

Streamlined interface for generating images with AI in Krita. Inpaint and outpaint with optional text prompt, no tweaking required.
https://www.interstice.cloud
GNU General Public License v3.0
6.98k stars 344 forks source link

Performance regression on auto-complete #1417

Open ronan36880 opened 1 week ago

ronan36880 commented 1 week ago

Commit 1c5e832100ced03cda5148d20c00f8d35ba6b812, which was created created to address #1294 seems to have caused a performance regression on auto-completion (I have confirmed with a git bisect).

I don't fully understand what is going wrong here, but here's the completion file I am using: fluffyrock-unbound-tag-completion.csv

Replication steps:

  1. Enable the fluffyrock-unbound-tag-completion completion file in the settings.
  2. Open a prompt, type by and press space.

It will then take some time to process things:

https://github.com/user-attachments/assets/19dbd639-ad88-457d-aa08-999d01f5795f

Locally reverting this commit seems to have fixed the issue - I wonder if the lstrip function could be much more computationally intensive than expected?

Acly commented 1 week ago

I don't think it has anything to directly with lstrip.

Previously by was stripped to by which is < 3 characters and does not trigger completion After the change, it triggers completion for by, which matches a LOT of tags (I think). Much more than any other 3-letter prefix. And it's just slow because it's a lot.

ronan36880 commented 1 week ago

I don't think it has anything to directly with lstrip.

Previously by was stripped to by which is < 3 characters and does not trigger completion After the change, it triggers completion for by, which matches a LOT of tags (I think). Much more than any other 3-letter prefix. And it's just slow because it's a lot.

Oooh, that's interesting. It does match a lot of tags from a quick look through the csv file. What would be the best way to fix it? It does freeze the UI thread every time after the change.

One obviously workaround would be to edit the csv file, but I'd rather avoid that if possible 😅

Acly commented 1 week ago

I think sorting and cutting off the csv is not a bad idea in general (not by hand obviously) - the tags with low use-count won't work anyway. Not sure if that throws out enough of the by matches though, but it doesn't seem to be a problem for the existing sets which include similar tags.

Don't have a great idea how to fix it in code. Could start auto-complete only at the 4th character, but that makes it less useful. Or special case the by prefix, or a bit more generally anything with a trailing space perhaps...