WIP: Optimize shadow hinting #3269
Conversation
Avoids iterating through all DOM nodes to walk through shadow DOM elements. Instead, we make use of the nyxt class 'nyxt-shadow-dom'. Note that the use of 'do' loop macro in 'rqs-nyxt-id' makes sure we get that parenscript generates the desired code in javascript.
Avoid using 'flet' in 'rqsa' and 'rqs-nyxt-id'. And move the functions into a separate parenscript macros.
@heiwiper let's keep that task outside the scope of this PR. Feel free to open an issue about it, and I'll share how I'd do it. To validate that there are no performance penalties, take a URL with many links and compare the performance before commit b43ec71 and at the tip of the working branch. URL https://en.wikipedia.org/wiki/Europe is a good candidate, since it hits almost 5k hints. I'll mark this PR as a draft. |
|
@heiwiper I've briefly meditated on the subject and we may be looking at the issue in a fundamental wrong way. The performance penalty may be coming purely from the inadequacy of the JS algorithm (and PS is thus irrelevant). I believe you have followed the implementation from saka-key (which, btw, is a draft and isn't used anywhere). I think that a better source of inspiration would be the one from LinkHints. Let me know if you'd like to work on it or if I should take over it. Thanks. For future reference, a test URL for this feature - https://developer.servicenow.com/dev.do. |
|
Hello André, I haven't had much time to work on this PR lately. I'm switching over to a new job and I'm using some of my free time to adapt to the new tech stack at work. I believe I won't be able to get back to contributing until around the end of the next month. So even though I would like to work on this issue, I would understand if someone else takes over due to the high priority. Sorry for not communicating my unavailability earlier. |
|
@heiwiper no worries! I hope you're doing well. |
Description
This draft PR should address the performance issue of
hint-modewhich significantly slows down hint overlays rendering when performing thefollow-hintcommand in a page with thousands of hintable elements.Fixes #3254
Discussion
The change that has been introduced so far is to separate the
fletdeclared function in the macrosnyxt/ps:rqsaandnyxt/ps:rqs-nyxt-idwhich is suspected to cause some delay, considering that a lambda function has to be declared during every call to these two macros.I have also optimized the two recursive functions which are now macros. Instead of walking through all DOM elements to check for shadow root elements, a query is now performed using the existing custom class
nyxt-shadow-root. Note that there's still need to confirm that this change does not introduce any regression bugs, so I would like to ask for more guidance about how to write a test forhint-mode.Apparently, the performance hasn't improved much compared to using the macros
qsaandqs-nyxt-id. The next part that I will try to tackle is the call tonyxt/ps:rqs-nyxt-idin the functionnyxt/mode/hint:hint-elements, I will try see whether it would be possible to refactor that part of the code.Checklist:
Everything in this checklist is required for each PR. Please do not approve a PR that does not have all of these items.
cd /path/to/nyxt/checkout git submodule add https://gitlab.common-lisp.net/nyxt/py-configparser _build/py-configparser:documentations written in the aforementioned style. (It's OK to skip the docstring for really trivial parts.)changelog.lispwith my changes if it's anything user-facing (new features, important bug fix, compatibility breakage).migration.lispentry for all compatibility-breaking changes.(asdf:test-system :nyxt)and(asdf:test-system :nyxt/gi-gtk)) and they pass.