## Version information Thunar 4.21.6, built from the release tarball. Ubuntu 26.04, GTK 3.24.52. The code is unchanged on master as of 2026-09-16. ## Current behaviour A file dragged from the main view onto a folder in the tree side pane is refused. The drag cursor shows "no drop" over every folder in the tree and nothing happens on release. There is no error and no warning on stderr. ## Steps to reproduce 1. View > Side Pane > Tree. 2. Leave hidden files hidden (View > Show Hidden Files off, the default). 3. In the tree, expand your home directory. 4. Drag any file from the main view onto any folder under home in the tree. The drop is refused. Expected: the file is copied or moved into that folder. This is reliable under a home directory because a home directory normally holds a few dozen hidden directories (39 on the machine this was found on), which is what makes the view's row indices and the model's row indices disagree. A minimal version of the same thing: ```sh mkdir -p /tmp/thunar-dnd/.hidden /tmp/thunar-dnd/target ``` Point Thunar at `/tmp/thunar-dnd`, expand it in the tree with hidden files hidden, and drag a file onto `target`. The tree shows one child, the model has two, so the index the drop handler resolves is not the row under the pointer. With hidden files **shown**, nothing is filtered, the two index spaces coincide, and drops work. That on/off behaviour is the quickest confirmation of the cause. ## Cause 4.21.5's "Use GtkTreeModelFilter for tree view side pane" (#1460) put a `GtkTreeModelFilter` between `ThunarTreeModel` and the `GtkTreeView`. Four call sites in `thunar-tree-view.c` still take a `GtkTreePath` from the **view** and resolve it against `view->model`, the **child** model, with no `gtk_tree_model_filter_convert_path_to_child_path`. `thunar_tree_view_get_dest_actions`: ```c /* determine the path for x/y */ if (gtk_tree_view_get_path_at_pos (GTK_TREE_VIEW (view), x, y, &path, NULL, NULL, NULL)) { /* determine the iter for the given path */ if (gtk_tree_model_get_iter (GTK_TREE_MODEL (view->model), &iter, path)) ``` `path` is in view coordinates, `view->model` is the child model. The iter is either invalid or points at a different row, so `file` comes back NULL, `thunar_file_accepts_drop` is never consulted, `actions` stays 0, and the drop is refused. Nothing warns, because `gtk_tree_model_get_iter` simply returns FALSE. ## Two more sites, by inspection I have not reduced either of these to reliable steps, so I am reporting them as code defects rather than as confirmed symptoms. Both are the same mistake. **`thunar_tree_view_get_preferred_toplevel_path`.** Its early-return branch returns the **cursor** path, which is in view coordinates, while every other branch returns `gtk_tree_model_get_path (GTK_TREE_MODEL (view->model), &iter)`, a child path. Its only caller, `thunar_tree_view_set_cursor`, feeds the result straight back into `view->model`, so the two branches cannot both be right. I do see the tree side pane intermittently stop highlighting the current directory after file operations, which would fit, but I cannot trigger it on demand. **`thunar_tree_view_key_press_event`**, in the `GDK_KEY_Left` and `GDK_KEY_Right` device lookups. These act on top-level rows, where in practice little is filtered, so I would not expect this one to misbehave often. It is still wrong. ## Fix Attached patch ([thunar-4.21.6-tree-pane-fixes.patch](/uploads/f3c709d319ea2196dc374526497bb282/thunar-4.21.6-tree-pane-fixes.patch)) adds a helper: ```c static gboolean thunar_tree_view_get_child_iter (ThunarTreeView *view, GtkTreePath *path, GtkTreeIter *iter) ``` which converts through `gtk_tree_model_filter_convert_path_to_child_path` before resolving, and uses it at the three lookup sites. In `get_preferred_toplevel_path` the cursor path is converted to a child path up front instead, so that every return value is a child path, as the caller assumes. With it applied, dropping onto tree folders works again with hidden files hidden. The patch also carries fixes for #1857, #1832 and a stale tab-title binding, which are all in the same area and all fall out of the same 4.21.5 change. Happy to split it into separate commits if that is easier to review.