Upon investigating the cell-addition logic for tilemap layers, I found that Manual Mode was creating new cels without properly attaching them. While PR aseprite#5720 resolved the immediate crash, the first interaction in Manual Mode was still incorrectly committing a tile.

To fix this, I implemented an early exit in the commit function that clears the pending data and opts out of the commit process when in Manual Mode. Currently, this leaves "ghost pixels" on the canvas, which I suspect are uncleaned preview tiles. I am working on a fix for the ghosting, but I would appreciate feedback on whether this "early exit" approach is the correct architectural path for the original bug.

Upon investigating the cell-addition logic for tilemap layers, I found
that Manual Mode was creating new cels without properly attaching them.
While PR aseprite#5720 resolved the immediate crash, the first interaction in
Manual Mode was still incorrectly committing a tile.

To fix this, I implemented an early exit in the commit function that
clears the pending data and opts out of the commit process when in
Manual Mode. Currently, this leaves "ghost pixels" on the canvas, which
I suspect are uncleaned preview tiles. I am working on a fix for the
ghosting, but I would appreciate feedback on whether this "early exit"
approach is the correct architectural path for the original bug.

Upon investigating the cell-addition logic for tilemap layers, I found
that Manual Mode was creating new cels without properly attaching them.
While PR aseprite#5720 resolved the immediate crash, the first interaction in
Manual Mode was still incorrectly committing a tile.

To fix this, I implemented an early exit in the commit function that
clears the pending data and opts out of the commit process when in
Manual Mode. Currently, this leaves "ghost pixels" on the canvas, which
I suspect are uncleaned preview tiles. I am working on a fix for the
ghosting, but I would appreciate feedback on whether this "early exit"
approach is a good fix for the original bug.

Upon investigating the cell-addition logic for tilemap layers, I found
that Manual Mode was creating new cels without properly attaching them.
While PR aseprite#5720 resolved the immediate crash, the first interaction in
Manual Mode was still incorrectly committing a tile.

To fix this, I implemented an early exit in the commit function that
clears the pending data and opts out of the commit process when in
Manual Mode. Currently, this leaves "ghost pixels" on the canvas, which
I suspect are uncleaned preview tiles. I am working on a fix for the
ghosting, but I would appreciate feedback on whether this "early exit"
approach is the correct architectural path for the original bug.

Upon investigating the cell-addition logic for tilemap layers, I found
that Manual Mode was creating new cels without properly attaching them.
While PR aseprite#5720 resolved the immediate crash, the first interaction in
Manual Mode was still incorrectly committing a tile.

To resolve this, I implemented an early exit within the commit function to bypass the data-writing process when in Manual Mode. However, this initially prevented the preview layer from being cleared.

To ensure a consistent UI state, I updated drawing_state to explicitly clean the preview buffer when the drawing loop is destroyed. This ensures that "ghost" pixels are removed even when a commit is opted out.

Upon investigating the cell-addition logic for tilemap layers, I found that Manual Mode was creating new cels without properly attaching them. While PR aseprite#5720 resolved the immediate crash, the first interaction in Manual Mode was still incorrectly committing a tile.

To resolve this, I implemented an early exit within the commit function to bypass the data-writing process when in Manual Mode. However, this initially prevented the preview layer from being cleared.

To ensure a consistent UI state, I updated drawing_state to explicitly clean the preview buffer when the drawing loop is destroyed. This ensures that "ghost" pixels are removed even when a commit is opted out.

I have also added a test to make sure no other crash or tile creation is added by the manual tilemode in the future.

Upon investigating the cell-addition logic for tilemap layers, I found that Manual Mode was creating new cels without properly attaching them. While PR aseprite#5720 resolved the immediate crash, the first interaction in Manual Mode was still incorrectly committing a tile.

To resolve this, I implemented an early exit within the commit function to bypass the data-writing process when in Manual Mode. However, this initially prevented the preview layer from being cleared.

To ensure a consistent UI state, I updated drawing_state to explicitly clean the preview buffer when the drawing loop is destroyed. This ensures that "ghost" pixels are removed even when a commit is opted out.

I have also added a test to make sure no other crash or tile creation is added by the manual tilemode in the future.

Upon investigating the cell-addition logic for tilemap layers, I found that Manual Mode was
creating new cels without properly attaching them. While PR aseprite#5720 resolved the immediate
crash, the first interaction in Manual Mode was still incorrectly committing a tile.

To resolve this, I implemented an early exit within the commit function to bypass the
data-writing process when in Manual Mode. However, this initially prevented the preview
layer from being cleared. To ensure a consistent UI state, I updated drawing_state to
explicitly clean the preview buffer when the drawing loop is destroyed.

This ensures that "ghost" pixels are removed even when a commit is opted out. I have also
added a test to make sure no other crash or tile creation is added by the manual tilemode
in the future.