Summary
In M._setup_blocking_diff (lua/claudecode/diff.lua), when diff_opts.open_in_new_tab = true, display_terminal_in_new_tab() creates a new tab early in setup. If setup then throws before M._register_diff_state runs, the post-pcall error handler does not close that tab, leaving the user with a stranded extra tab.
The state-based cleanup (if active_diffs[tab_name] then M._cleanup_diff_state(...)) can't fire because the diff was never registered, and the pre-registration else branch (added in #260 for the terminal-only fallback) only cleans up fallback_window and new_buffer. new_tab_handle is declared inside the pcall closure, so the error handler can't reach it even if it tried.
Trigger
Uncommon, but the same single-failure class accepted elsewhere: an error between tab creation and state registration — most realistically load_original_buffer's vim.cmd("edit " .. fnameescape(old_file_path)) hitting a BufReadPre/BufRead autocmd that throws, a swap-file conflict, or a filereadable-vs-:edit race.
Effect
One stranded tab per failed setup on the open_in_new_tab path (and the original tab is not refocused).
Suggested fix
Symmetric to the fallback_window / new_buffer hoists in #260: lift new_tab_handle (and the original tab id) above the pcall, and in the error handler's pre-registration branch switch back to the original tab and :tabclose the stranded one. Add a regression test that forces an error during open_in_new_tab setup and asserts no extra tab survives.
Context
Pre-existing (the open_in_new_tab path predates #260). Surfaced by a Claude review on #260 (#260) and deliberately deferred to keep that PR scoped to the #231 terminal-only fix. #260 already established the hoist-and-clean-up-on-error pattern this would follow.
Summary
In
M._setup_blocking_diff(lua/claudecode/diff.lua), whendiff_opts.open_in_new_tab = true,display_terminal_in_new_tab()creates a new tab early in setup. If setup then throws beforeM._register_diff_stateruns, the post-pcallerror handler does not close that tab, leaving the user with a stranded extra tab.The state-based cleanup (
if active_diffs[tab_name] then M._cleanup_diff_state(...)) can't fire because the diff was never registered, and the pre-registrationelsebranch (added in #260 for the terminal-only fallback) only cleans upfallback_windowandnew_buffer.new_tab_handleis declared inside thepcallclosure, so the error handler can't reach it even if it tried.Trigger
Uncommon, but the same single-failure class accepted elsewhere: an error between tab creation and state registration — most realistically
load_original_buffer'svim.cmd("edit " .. fnameescape(old_file_path))hitting aBufReadPre/BufReadautocmd that throws, a swap-file conflict, or afilereadable-vs-:editrace.Effect
One stranded tab per failed setup on the
open_in_new_tabpath (and the original tab is not refocused).Suggested fix
Symmetric to the
fallback_window/new_bufferhoists in #260: liftnew_tab_handle(and the original tab id) above thepcall, and in the error handler's pre-registration branch switch back to the original tab and:tabclosethe stranded one. Add a regression test that forces an error duringopen_in_new_tabsetup and asserts no extra tab survives.Context
Pre-existing (the
open_in_new_tabpath predates #260). Surfaced by a Claude review on #260 (#260) and deliberately deferred to keep that PR scoped to the #231 terminal-only fix. #260 already established the hoist-and-clean-up-on-error pattern this would follow.