Scenario migration — behavioral findings

Behavioral asymmetries and probable bugs surfaced while migrating tests/e2e/* to scenario form. Each entry follows the pattern:

Source test • Action sequence • e2e claim vs scenario observation • assessment

Findings here are evidence, not conclusions: the migration runs the scenario through the same Editor instance the e2e drives, just without the keymap/render/screen-scrape layer, so if observations differ it's the editor's behavior that differs, not the framework's.

---

1. MoveLineEnd parks cursor at last text cell, not one past

Source: layout / cursor scenarios. Sequence: Action::MoveLineEnd on "hello". Expectation: cursor at byte 5 (one past the last char) and at text col 5 in screen coordinates. Observation: EditorTestApi::primary_caret() reports byte 5 (matches), but cursor_screen_position() reports (col=4, row=0) — one column short. Assessment: Possibly a confusion between "buffer cursor at EOL byte position" and "screen column of cursor". When we render to the terminal, a cursor at byte 5 of "hello" should sit at column 5 (immediately after 'o'), not column 4 (on top of 'o'). Pinned in tests/semantic/migrated_layout_cursor.rs::layout_cursor_after_move_line_end_lands_one_past_last_char.

2. ToUpperCase with no selection uppercases the entire buffer

Source: tests/e2e/case_conversion.rs-adjacent migrations. Sequence: Action::ToUpperCase on "hello" with no selection. Expectation (one of two reasonable): (a) no-op, since there's nothing selected; or (b) uppercase word/line under cursor. Observation: Uppercases the entire buffer ("hello" → "HELLO") and parks the cursor at the end of the upcased range. Assessment: Behavior is consistent (the editor's "upcase" command falls back to whole-buffer when no selection exists), but worth recording so a future change to "no-op without selection" is surfaced. Pinned in migrated_bulk::bulk_uppercase_with_no_selection_uppercases_full_buffer.

3. SelectLeft at byte 0 sets an empty anchor

Source: Various selection-edge cases. Sequence: Action::SelectLeft on a fresh buffer with cursor at byte 0. Expectation: No-op (can't select left of byte 0). Observation: Cursor stays at byte 0 but anchor becomes Some(0). The selection is non-None but empty (range 0..0). Assessment: Either a real bug (anchor should not flip from None to Some(0) for a no-op selection), or a deliberate convention (deselect-on-move depends on anchor.is_some() to decide whether to collapse). Pinned in migrated_misc::migrated_select_left_at_byte_zero_creates_empty_selection.

4. DeleteWordForward consumes following whitespace

Source: Migration of KillWord-like behavior. Sequence: Action::DeleteWordForward at byte 0 of "foo bar". Expectation: Delete "foo" ⇒ " bar". Observation: Deletes "foo " (word + following whitespace) ⇒ "bar". Assessment: Both behaviors exist in the wild (Emacs kill-word ≈ word + ws, vim dw ≈ word only). Pinning the editor's choice. Not a bug.

5. OpenLine cursor position (already fixed)

Source: tests/e2e/emacs_actions.rs::test_open_line_basic. Sequence: Action::OpenLine advances the cursor; e2e expected it to stay put (Emacs C-o semantics). Status: Found during the original Track-B migration; fixed in commit 90ef113e ("fix: OpenLine cursor advance — restore Emacs C-o semantics") before the scenario plumbing existed.

6. Save-after-external-edit refuses to overwrite

Source: migrated_persistence::migrated_external_edit_after_save_persists_until_buffer_resaves. Sequence: Save buffer, externally clobber the file with different content, save again. Expectation: Last-write-wins — second editor save overwrites the external edit. Observation: External content remains on disk after the second save attempt; the editor refuses to overwrite a changed-on-disk file (probably auto-revert / conflict-detection guard). Assessment: Conservative and probably correct behavior — overwriting an external edit silently would lose data — but the migrated test pins the external content wins outcome so future behavior changes are flagged.

7. SelectAll + SortLines preserves anchor only when buffer unchanged

Source: tests/e2e/sort_lines.rs::test_sort_lines_*. Sequence: [SelectAll, SortLines]. Expectation: Either (a) anchor always preserved (selection survives the sort) or (b) anchor always cleared (selection collapses). Observation: Asymmetric:

• Buffer changed (basic, case_sensitive, preserves_trailing_newline, with_empty_lines, with_numbers): cursor at byte len, anchor cleared (None).
• Buffer unchanged (already_sorted, single_line_no_change): cursor at byte len, anchor preserved (Some(0)).

Assessment: Probably a real asymmetry. The natural expectation is "the action either preserves or clears selection consistently." Pinned in migrated_sort_lines_full::*.

10. AddCursorAbove sticky-column drifts on cascading calls

Source: tests/e2e/multicursor.rs::test_add_cursor_above. Sequence: [MoveDocumentEnd, AddCursorAbove, AddCursorAbove] on a buffer of 3 equal-length lines "Line 1\nLine 2\nLine 3" (each 6 bytes + newline). Expectation: column-matched cursors at the same column on each line: {0, 6, 13, 20}-based positions where each line's cursor lands at the same column as the original. Observation: Cursors land at {0, 6, 20} — the secondary on Line 2 sits at byte 6 (start of Line 2), not byte 13 (column-matched end of "Line 2"). The sticky-column drifts as the primary moves up, so the second AddCursorAbove adds a cursor at the new primary's column (which has already drifted). Assessment: Probably a real asymmetry — multi-cursor "add above/below" usually preserves the column from the original cursor, not from each intermediate primary. Pinned in migrated_multicursor_full::migrated_add_cursor_above_twice_yields_three_cursors.

9. InsertTab on a selection advances both anchor and cursor

Source: tests/e2e/indent_dedent.rs::test_tab_indent_multiple_lines_spaces. Sequence: [SelectAll, InsertTab] on "Line 1\nLine 2\nLine 3". Expectation (naive): anchor stays at 0 (selection grows to include the prepended indent on the first line). Observation: anchor advances by indent_width (4), so the selection covers "Line 1\n Line 2\n Line 3" — excluding the indent prepended to the first selected line. Assessment: Probably deliberate: the user "selected this content" and InsertTab logically operates on it without expanding the selection to cover the freshly-inserted indent. But it means subsequent SelectAll-then-asserting-on-selection tests need to know about the off-by-N. Pinned in migrated_indent_dedent_full::migrated_tab_indent_multiple_lines_spaces.

8. DuplicateLine lands cursor at start of duplicate

Source: tests/e2e/duplicate_line.rs::test_duplicate_line_*. Sequence: [<navigate to line>, DuplicateLine] then optional InsertChar. Expectation: Original e2e tests asserted only buffer text; cursor was unspecified. The e2e test_duplicate_line_cursor_on_new_line proves "typing after duplicate inserts on the new line" — equivalent to "cursor on duplicated line." Observation: Cursor lands at the start of the duplicated line (byte = end-of-original-line + 1 newline). Assessment: Sensible default. Pinned across migrated_duplicate_line_full::* so a behavior change surfaces.

11. RemoveSecondaryCursors also clears active selection

Source: tests/e2e/block_selection.rs::test_block_select_then_escape. Sequence: [BlockSelectDown, RemoveSecondaryCursors] on "line1 text\nline2 text\nline3 text" (cursor at byte 0). Expectation (naive): RemoveSecondaryCursors removes secondary cursors but leaves the primary's selection untouched (so cursor at 0 with selection 0..11 should remain). Observation: Two things happen in BlockSelectDown+ RemoveSecondaryCursors:

• BlockSelectDown is implemented as an extension of the primary's selection (anchor=0, position=11), not as adding a secondary cursor.
• RemoveSecondaryCursors clears the active selection in addition to dropping secondaries — the surviving cursor is at byte 11 with anchor=None. Assessment: Likely intentional (Esc-as-clear-all in a multi-cursor block), but worth pinning so a future split between "remove secondaries" and "clear selection" doesn't silently break this. Pinned in migrated_block_selection_extras::migrated_block_select_then_remove_secondary_collapses.

12. AddCursorNextMatch is three behaviors in one action

Source: tests/e2e/issue_1697_ctrl_d_after_search.rs and tests/e2e/multicursor.rs::test_add_cursor_next_match. Observation: Action::AddCursorNextMatch (Ctrl+D) does different things depending on cursor + search state:

1. No selection, active search (after Search + PromptConfirm): selects the search match at the cursor's current position. Does NOT add a new cursor.
2. No selection, no search: selects the word at the cursor.
3. Has selection (from any source): adds a new cursor at the next occurrence of the selected text; the new cursor becomes primary, the original becomes secondary. Assessment: This three-state behavior is intentional but non-obvious. Pinned in migrated_search_modal_flows::migrated_ctrl_d_after_substring_search_selects_match_not_word (case 1) and migrated_multicursor_extras::migrated_add_cursor_next_match_with_*_selection (case 3).

---

How to add a finding

1. Reproduce the asymmetry in a scenario (BufferScenario, LayoutScenario, etc.) and pin it as a passing test.
2. Add an entry above naming the source e2e file (if any), the action sequence, what the e2e claimed, what the scenario observes, and a short assessment.
3. If you suspect a bug, file an issue and link the scenario.