Migrate command execution to structured argv lists across the plugin, replacing shell-string command construction with `vim.system` argv execution. The runner API now exposes argv-based execution as the default path:
- `runner.execute_command` -> `runner.execute`
- `runner.execute_command_async` -> `runner.execute_async`
- `runner.execute_command_raw` -> `runner.execute_raw`
- `runner.execute_command_raw_async` -> `runner.execute_raw_async`
- `runner.execute_command_sync` was removed; use `runner.execute` and handle the result directly
BREAKING CHANGES:
- `cmd.new({ args = ... })` now requires `args` as `string[]` argv tokens, not a single string.
- `cmd.log({ ... })` now uses `raw_flags` as `string[]` for raw passthrough flags; `raw = "..."` is no longer supported.
- Direct users of `jj.core.runner` must migrate from the old `execute_command*` string helpers to the new argv-based `execute*` helpers.
Pass raw command arguments to the interactive floating terminal instead
of a shell-escaped string.
This fixes native `jj resolve` exiting immediately when opened without
an external tool.
Add split command that allows splitting a change into multiple revisions
from the log buffer or via :J split. Supports --parallel, --message,
--fileset, and --ignore-immutable flags. Includes immutability checks
with user confirmation and empty change detection.
A regresion was introduced by #81 where the log buffer was not being
restored when there was no modification after describing from the
log buffer. This has been reworked and now the function `on_done` is
always called not only when there are lines in the buffer giving the
responsibility to check the nilness on the `on_done` function.
There was a bug also when restoring cursor position which was being
wiped when the buffer was closed which is not actually what we want
- This required some improvements on the editor buffer logic,
introduced a `on_write` (previously on_done but it wasn't the
right meaning) hook and an `on_unload` to allow for both
describe and commit correct workflows.
- Introduced an utils function to extract the description text and
clean it post buffer write
- Refactor diff into modular architecture (native, diffview, codediff)
- Add the possibility for users to define their own diff backend
- Support external diff tools like diffview.nvim codediff.nvim
- Enable visual mode selection to diff two changes in log buffer
- Add summary tooltip integration with terminal
- Update documentation with new diff workflows
Display a summary of changed files for any revision in the log buffer
using a floating tooltip. From the tooltip, users can diff or edit
specific files directly.
- Add summary tooltip triggered by <S-k> in log buffer
- Integrate tooltip as stateful component of terminal
- Make summary keymaps configurable
- Add documentation and demo gif
- Squash: behaves like rebase, the user can select 1 or more
revisions to squash and then a destination
- Quick squash: `-r {rev} -u --ignore-immutable` flag from jj allowing
for quick visuall squashes
Now it is possible to rebase interactively from the log buffer with visual feedback. You can rebase one or many visually selected changes.
Supported target modes are: `-o` / `-A` / `-B`.
Limitation:
- As of now you cannot rebase a whole branch (-b) from the log view.
Bonus from this pr:
- Now keymaps support different modes
When closing a terminal buffer (floating or regular), a race condition
could occur where the BufWipeout/BufDelete autocmd clears the buffer
reference, but the job's on_exit handler later tries to operate on it
via vim.schedule. This caused 'Invalid buffer' errors.
Add validity checks in both floating and regular buffer on_exit
handlers to safely handle the case where the buffer has already been
cleaned up.
Add cursor position restoration with configurable timing to handle terminal
buffer rendering asynchronously. This fixes cursor column being reset to 0
when refreshing jj log commands.
Changes:
- Add `terminal.cursor_render_delay` configuration option (default: 10ms)
- Add buffer.get_cursor() and buffer.set_cursor() helpers in core/buffer
* buffer.set_cursor() uses defer_fn with configurable delay for terminal buffers
* Automatic position validation with clamping to buffer bounds
* Line number clamped to [1, line_count]
* Column clamped to [0, line_length] based on actual line content
* Validation happens inside deferred callback to check against final rendered content
- Extract clamp_cursor_position() helper to eliminate code duplication
- Refactor terminal module to use new buffer cursor helpers
- Store and restore cursor position when cycling through log commands
- Update README with terminal configuration section and example usage
The delay is necessary because nvim_open_term() + nvim_chan_send() have
asynchronous rendering in the terminal emulator layer. Setting the cursor
before rendering completes results in the column being reset to 0. Users
experiencing issues can increase the delay value if needed.
- Close terminal buffer before opening describe editor
- Capture log state before closing to detect if we should reopen
- Add on_unload callback to editor to handle log reopening
- Suppress redraws during buffer transitions with lazyredraw
- Remove duplicate hardcoded status keymaps from terminal.lua (were referencing undefined functions)
- Fix resolve_keymaps_from_specs to validate handlers exist before creating keymaps
- Add M.setup() to cmd module for configuration
- Move describe editor config to M.config.describe.editor with nested structure
- Add keymaps config with separate log, status, and close sections
- Implement resolve_keymaps_from_specs() helper to build keymap lists from config
- Extract editor.open_editor() keymaps parameter support
- Move status/log event handlers from terminal.lua to cmd.lua
- Convert log/status buffer keymaps to config-driven approach with handler specs
NicolasGB