<div dir="ltr">Hi John,<div><br></div><div>That sounds very promising!<br>Let me know when you have a PR available so I can test it out.</div><div><br></div><div>Quite sure I measured 10x, but maybe we measured something slightly different - or it's hardware-dependent.</div><div>(The issue with scrolling happened only in my closed-source project, which has a behavior similar to ListVIew.)</div><div><br></div><div>Florian Kirmaier</div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Mon, 1 Jan 2024 at 16:37, John Hendrikx <<a href="mailto:john.hendrikx@gmail.com">john.hendrikx@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
  
    
  
  <div>
    <p>Hi,</p>
    <p>I think I've found the issue. Note that on my machine I can't
      really reproduce the problem as clearly (Windows 10) as the
      performance degradation is not quite as bad as mentioned in the
      issue (maybe a factor of 2, and scrolling was unaffected), but
      it's probably what you are seeing as little else changed in the PR
      that seems to be the root cause.  Still, VisualVM did hint at the
      probable culprit (SimpleSelector#applies).</p>
    <p>The problem is that BitSet (used for storing and comparing CSS
      style classes) has special logic to do fast containsAll checks if
      it detects it is compared against another BitSet. However, due to
      a read only wrapper on one of the sets this detection fails, and
      it will fall back to standard containsAll logic which is a bit
      slower.</p>
    <p>The fix should be relatively trivial, and I'll create a PR soon.<br>
    </p>
    <p>I did note a few other things in my investigation:</p>
    <p>- JFXCentral has around 1000 style classes, which means a BitSet
      as currently implemented, could use 125 bytes of memory for each
      StyleClassSet if a high bit was set<br>
      - Most selectors involve only 1 style class, sometimes 2 and
      rarely 3 or more<br>
      - Most nodes have 1 style class, sometimes 2 and rarely 3 or more<br>
      <br>
      Now, if I were to implement this kind of logic knowing that the
      number of possible style classes can be quite high, and knowing
      that most selectors and nodes will rarely have 3 or more style
      classes assigned to them, I'd probably not use a non-sparse bit
      set as it is currently implemented.</p>
    <p>I've played around a bit with a sparse bitset, and although I did
      not see amazing (overall) performance gains, the change (after
      applying the regression fix) did shave off another 10-20 ms for
      the "loadTimeFX" measurement in the JFXCentral application.</p>
    <p>--John<br>
    </p>
    <p><br>
    </p>
    <div>On 31/12/2023 15:24, Christopher
      Schnick wrote:<br>
    </div>
    <blockquote type="cite">
      
      <p>Hello, <br>
        <br>
        I just tested this with our JavaFX application and can confirm
        that there are massive differences. It takes around 1-2 seconds
        to completely apply all application stylesheets in JavaFX 20 but
        takes around 6-7 seconds in JavaFX 21.<br>
      </p>
      <div>On 12/31/2023 3:00 PM, Florian
        Kirmaier wrote:<br>
      </div>
      <blockquote type="cite">
        
        <div dir="ltr">Hi Everyone,<br>
          <br>
          Sorry for the delay - but I couldn’t find the time to extract
          the TestApplication for this bug.<br>
          Luckily, I found another application, which is also open
          source, which is affected by the application.
          <div><br>
            I'm speaking about <a href="https://www.jfx-central.com/" target="_blank">https://www.jfx-central.com/</a> -
            both the desktop and web versions are affected.<br>
            I’ve seen a performance deterioration of 10x when switching
            pages when using JavaFX21 compared to JavaFX20.<br>
            <br>
            I’ve created a ticket with further instructions on how to
            test it:<br>
            <a href="https://bugs.openjdk.org/browse/JDK-8322795" target="_blank">https://bugs.openjdk.org/browse/JDK-8322795</a><br>
            <br>
            Greetings<font color="#888888"><br>
              <br>
              Florian Kirmaier</font></div>
        </div>
        <br>
        <div class="gmail_quote">
          <div dir="ltr" class="gmail_attr">On Fri, 27 Oct 2023 at
            21:31, Andy Goryachev <<a href="mailto:andy.goryachev@oracle.com" target="_blank">andy.goryachev@oracle.com</a>>
            wrote:<br>
          </div>
          <blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">
            <div>
              <div lang="EN-US">
                <div>
                  <p class="MsoNormal"><span>Please create a ticket,
                      Florian.  Would it be possible to profile the
                      application when scrolling?</span></p>
                  <p class="MsoNormal"><span> </span></p>
                  <p class="MsoNormal"><span>Thank you</span></p>
                  <p class="MsoNormal"><span>-andy</span></p>
                  <p class="MsoNormal"><span> </span></p>
                  <p class="MsoNormal"><span> </span></p>
                  <p class="MsoNormal"><span> </span></p>
                  <div id="m_3386554746980567036m_9047044901516997342mail-editor-reference-message-container">
                    <div>
                      <div style="border-right:none;border-bottom:none;border-left:none;border-top:1pt solid rgb(181,196,223);padding:3pt 0in 0in">
                        <p class="MsoNormal" style="margin-bottom:12pt"><b><span style="font-size:12pt;color:black">From: </span></b><span style="font-size:12pt;color:black">openjfx-dev
                            <<a href="mailto:openjfx-dev-retn@openjdk.org" target="_blank">openjfx-dev-retn@openjdk.org</a>>
                            on behalf of Florian Kirmaier <<a href="mailto:florian.kirmaier@gmail.com" target="_blank">florian.kirmaier@gmail.com</a>><br>
                            <b>Date: </b>Friday, October 27, 2023 at
                            04:20<br>
                            <b>To: </b><a href="mailto:openjfx-dev@openjdk.java.net" target="_blank">openjfx-dev@openjdk.java.net</a>
                            <<a href="mailto:openjfx-dev@openjdk.java.net" target="_blank">openjfx-dev@openjdk.java.net</a>><br>
                            <b>Subject: </b>Performance Regression in
                            21 - CSS</span></p>
                      </div>
                      <div>
                        <p class="MsoNormal"><span style="font-size:11pt">Hi everyone,<br>
                            <br>
                            I've noticed that some parts of one of my
                            applications is significantly slower with
                            21. It's fast with 20.<br>
                            The application heavily uses (and reuses)
                            TextFlow with a Cell pattern.<br>
                            When I scroll, it is smooth with 20, but has
                            big freezes with 21.<br>
                            <br>
                            I've tried all the commits that happened in
                            between, and pin-pointed it down to the
                            following:<br>
                            ticket: <a href="https://bugs.openjdk.org/browse/JDK-8304959" target="_blank">https://bugs.openjdk.org/browse/JDK-8304959</a><br>
                            PR: <a href="https://github.com/openjdk/jfx/pull/1070" target="_blank">https://github.com/openjdk/jfx/pull/1070</a><br>
                            commit: <a href="https://github.com/openjdk/jfx21u/commit/3fa02ee96a6dadbc20cacbf399a2d65df708eee1" target="_blank">https://github.com/openjdk/jfx21u/commit/3fa02ee96a6dadbc20cacbf399a2d65df708eee1</a><br>
                            <br>
                            <br>
                            According to the description and the
                            discussion in the PR - this wasn't supposed
                            to change any performance.<br>
                            Is this regression known?<br>
                            Otherwise, I should create a ticket for it.<br>
                            <br>
                            Greetings Florian</span></p>
                      </div>
                    </div>
                  </div>
                </div>
              </div>
            </div>
          </blockquote>
        </div>
      </blockquote>
    </blockquote>
  </div>

</blockquote></div>