Cache offset and length in document nodes

[Amir-P]

Fixes #244

Performance comparison with master ran locally on an M1 Mac:

Analyzing tests for scrolling
Average Frame Build Time: Target: 87.48 Reference: 87.62
90th Percentile Frame Build Time: Target: 96.96 Reference: 96.93
90th Percentile Frame Build Time: Target: 98.92 Reference: 99.47
Average GPU Frame Time: Target: 0.00 Reference: 0.00
Performance tests found no regression


Analyzing tests for editing
Average Frame Build Time: Target: 129.21 Reference: 762.61
90th Percentile Frame Build Time: Target: 135.17 Reference: 769.02
90th Percentile Frame Build Time: Target: 137.28 Reference: 770.56
Average GPU Frame Time: Target: 0.00 Reference: 0.00
Performance tests found no regression

[codecov]

Codecov Report

Attention: Patch coverage is 94.66667% with 4 lines in your changes missing coverage. Please review.

Project coverage is 88.04%. Comparing base (e725636) to head (c4a937b).
Report is 4 commits behind head on master.

Files with missing lines	Patch %	Lines
...ackages/parchment/lib/src/document/attributes.dart	33.33%	2 Missing ⚠️
packages/parchment/lib/src/document/embeds.dart	33.33%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #408      +/-   ##
==========================================
+ Coverage   87.94%   88.04%   +0.09%     
==========================================
  Files          62       62              
  Lines       10364    10419      +55     
==========================================
+ Hits         9115     9173      +58     
+ Misses       1249     1246       -3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Amir-P]

This change is supposed to improve the performance of editor specially when editing and having cursor. I was hoping you guys can take a look, test things out, and give feedback on it. Many thanks in advance. @maelchiotti @simonbengtsson

[simonbengtsson]

Sure! Tried now both on master and on the perf branch both with and without 100k characters. In summary I didn't really observe any differences. The 100k characters did not seem to matter and the experience was the same both in master and in the perf branch. The editing experience is almost lag free when writing normally so tried spamming the keyboard as fast as I could instead and then some extreme lag is observable. Could very much be related to my persistence code however and not related to fleather editor.

Attaching a video of that lag when keyboard spamming for reference. Here I have 100k characters in the editor and am using the perf branch but was about the same no matter what. Note that I stopped spamming the keyboard when I saw a visible character at the | line so all characters added after that is added by lag ie when I'm not touching the keyboard.

pref_branch_100k_characters.mov

[Amir-P]

Thank you @simonbengtsson for testing it. It's weird that you're not facing any issues with the master maybe your computer is too powerful! According to our benchmark tests we've seen ~3x improvement in average frame build time on macOS (from ~450ms to ~130ms for a frame).
BTW, what a nice looking app you have there!

[amantoux]

Sure! Tried now both on master and on the perf branch both with and without 100k characters. In summary I didn't really observe any differences. The 100k characters did not seem to matter and the experience was the same both in master and in the perf branch. The editing experience is almost lag free when writing normally so tried spamming the keyboard as fast as I could instead and then some extreme lag is observable. Could very much be related to my persistence code however and not related to fleather editor.

The number of paragraphs (LineNode or BlockNode) makes a huge difference.
If you have 1000+ paragraphs things get pretty sticky

[amantoux]

LGTM!
Awesome performance comming in :)

[amantoux]

I'm fine with this but it would have been nice to have it in another PR :)

[amantoux]

:)