Client using local scaling can fail to resize (especially with a maximized window)

[marc-legendre]

Hi there!

I stumbled upon an issue where a client using local scaling can fail to resize when the window resizes. This happens in very specific conditions, for example when starting with a maximized window, then resizing it, then maximizing it again.

This happens since 0ff0844 (first released in 1.3.0)

Reproduction

This is easy to reproduce with the VNC app:

• open the VNC app
• maximize the window
• connect
• resize window
  • the viewer resizes correctly
• maximize the window again
  • the viewer does not resize

Here is a screencast demonstrating the issue, using vnc.html from master (047531e), with Chromium (but this is not browser-specific).

Screencast.2024-09-30.11.40.32.webm

Misc. details

I have not completely understood the issue, but I can share some observations / intuitions

• the client resize happens only if !_clientHasExpectedSize()
• but _expectedClientWidth and _expectedClientHeight never change when I resize the window.
• thus if the new window size happens to be the same as the initial size, the client won't resize
• this is why the issue is easily triggered with a maximized window
• this is also why it rarely happens otherwise, and even then it is not even noticeable

[marc-legendre]

I don't understand everything in RFB, but I am wondering if maybe _handleResize() should call _saveExpectedClientSize() like _resize() does ?

diff --git i/core/rfb.js w/core/rfb.js
index 0007bed..2758a1e 100644
--- i/core/rfb.js
+++ w/core/rfb.js
@@ -730,6 +730,7 @@ export default class RFB extends EventTargetMixin {
         window.requestAnimationFrame(() => {
             this._updateClip();
             this._updateScale();
+            this._saveExpectedClientSize();
         });
 
         if (this._resizeSession) {

Indeed, both call _updateClip() and _updateScale(), but only _resize calls _saveExpectedClientSize(), and it feels like _handleResize should.

(note: this patch makes the issue disappear)

[CendioOssman]

Thanks for your bug report! I can reproduce the issue here as well.

I think we didn't fully consider all scenarios in #1617, and we need to reevaluate the design here.

@samhed, can you have a look?