mirror of
https://github.com/rustdesk/rustdesk.git
synced 2026-10-14 07:50:14 +00:00
rustdeskandClaude Opus 5.5
f0acecb8af
The shape shown before stays while the one in use decodes, so a switch does not flash the default cursor. It also stayed once the shape in use was known to be lost (rejected by the core, or not decoding): the controller kept drawing, say, an I-beam for a pointer it never got, until the peer changed shape. Both the native and the painted cursor now leave the default one, as master did for the native cursor. A shape that does not decode is marked in updateCursorData, whether it was sent or fetched again, so both paths agree. The client also selects a shape it rejects as it arrives. It showed the shape before until the peer selected the rejected one again, and the default cursor from then on; it now looks the same both times. Sciter ignores an id it has no shape for, as it did. The default and forbidden cursors are made with the model. They were made the first time a build asked for one, and their decode lands later with nothing to draw again, so that first build showed the shape before in their place. A fetch failing after the session was cleared marks nothing in the new one, and a fetch that throws at once tells the listeners after the build that asked for it, not during it. The forbidden cursor, shown while input is off, became the cursor shown last, so a raster made after input came back showed it until the shape's pixels were fetched, or for good if they never came. Only a shape stands in for a shape now; CursorModel.shown keeps the rule for both builders. _checkView asks for a single registration again: every native cursor is made once per session, and one made twice leaks an HCURSOR on Windows. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01EZ49AbZJYfm8NTp5yDPMab