The terminal.Stream next/nextSlice functions can now no longer fail.
All prior failure modes were fully isolated in the handler `vt`
callbacks. As such, vt callbacks are now required to not return an error
and handle their own errors somehow.
Allowing streams to be fallible before was an incorrect design. It
caused problematic scenarios like in `nextSlice` early terminating
processing due to handler errors. This should not be possible.
There is no safe way to bubble up vt errors through the stream because
if nextSlice is called and multiple errors are returned, we can't
coalesce them. We could modify that to return a partial result but its
just more work for stream that is unnecessary. The handler can do all of
this.
This work was discovered due to cleanups to prepare for more C APIs.
Less errors make C APIs easier to implement! And, it helps clean up our
Zig, too.
Adds progress-style config to control OSC 9;4 progress bar visibility.
Defaults to true, set false to hide.
Fixes#11241
AI Disclosure: Claude Code (Opus 4.6) used for codebase exploration,
code review, and testing assistance. All code written and reviewed by
hand.
Use OSC 133;P (prompt mark) instead of 133;A (fresh line + prompt mark)
inside PS1 and PS2. Readline redraws the prompt on vi mode switches,
Ctrl-L, and other events, and 133;A's fresh-line behavior would emit a
CR+LF whenever the cursor wasn't at column 0, causing visible extra
newlines.
The one-time 133;A is now emitted via printf in __ghostty_precmd, which
only runs once per prompt cycle via PROMPT_COMMAND. On SIGWINCH, bash
redraws PS1 (firing the 133;P marks) but doesn't re-run PROMPT_COMMAND,
so there's no unwanted fresh-line on resize either. The redraw=last flag
persists from the initial printf.
This is a little less optimal than our previous approach, in terms of
number of prompt marks we emit, but it produces an overall more correct
result, which is the important thing.
Because readline prints its output outside the scope of PS1, those
characters "inherit" the surrounded prompt scope. This is usually fine,
but it can sometimes get out of sync (especially during redraws). This
is inherently a limitation of the fact that it's a separate output
channel, so we just have to accept that can happen.
See: #11267
Fixes#11316
This mirrors the `prompt` actions (hence why there is no window action
here) and enables setting titles via keybind actions which importantly
lets this work via command palettes, App Intents, AppleScript, etc.
Only replace the \n prompt escape when inserting secondary prompt marks,
not literal newlines ($'\n'). Literal newlines may appear inside $(...)
or `...` command substitutions, and inserting escape sequences there
breaks the shell syntax. For example:
PS1='$(if [ $? -eq 0 ]; then echo -e "P";
else echo -e "F";
fi) $ '
The literal newlines between the if/else/fi are part of the shell syntax
inside the command substitution. The previous code replaced all literal
newlines in PS1 with newline + OSC 133 escape sequences, which injected
terminal escapes into the middle of the command substitution and caused
bash to report a syntax error when evaluating it.
The \n prompt escape is PS1-specific and safe to replace globally. This
means prompts using literal newlines for line breaks (rather than \n)
won't get per-line secondary marks, but this is the conventional form
and avoids the need for complex shell parsing.
Fixes: #11267
Replace the strip-in-preexec / re-add-in-precmd pattern for OSC 133
marks with a save/restore approach. Instead of pattern-matching marks
out of PS1 (which exposes PS1 in intermediate states to other hooks), we
save the original PS1/PS2 before adding marks and then restore them.
This also adds dynamic theme detection: if PS1 changed between cycles
(e.g., a theme rebuilt it), we skip injecting continuation marks into
newlines. This prevents breaking plugins like Pure that use pattern
matching to strip/rebuild the prompt.
Additionally, move _ghostty_precmd to the end of precmd_functions in
_ghostty_deferred_init (instead of substituting in-place) so that the
first prompt is properly marked even when other hooks were appended
after our auto-injection.
There's one scenario that we still don't complete cover:
precmd_functions+=(_test_overwrite_ps1)
_test_overwrite_ps1() {
PS1="test> "
}
... which results in the first prompt not printing its prompt marks
because _test_overwrite_ps1 becomes the last thing to run, overwriting
our marks, but this will be fixed for subsequent prompts when we move
our handler back to the last index.
Fixes: #11282
If the CLI argument `--working-directory` is not used with
`+new-window`, the current working directory that `ghostty +new-window`
is run from will be appended to the list of configuration data sent
to the main Ghostty process. If `-e` _was_ used on the CLI, the
`--working-directory` that was appended will be interpreted as part of
the command to be executed, likely causing it to fail.
Instead, insert `--working-directory` at the beginning of the list of
configuration that it sent to the main Ghostty process.
Fixes#11356
Fixes#11336
Introduce a proper WorkingDirectory tagged union type with home, inherit,
and path variants. The field is now an optional (?WorkingDirectory) where
null represents "use platform default" which is resolved during Config.finalize
to .inherit (CLI) or .home (desktop launcher).
If you have "Noto Sans Tai Tham" and/or "Noto Sans Javanese" installed
locally on Linux, three tests fail. This PR disables those tests until a
more permanent solution can be found.
If you have "Noto Sans Tai Tham" and/or "Noto Sans Javanese" installed
locally on Linux, three tests fail. This PR disables those tests until a
more permanent solution can be found.
When a grapheme expands to width 2 at the screen edge, this path can
write spacer_head before printWrap() sets row.wrap. With an active
hyperlink, printCell triggers hyperlink bookkeeping and page integrity
checks in that intermediate state, causing UnwrappedSpacerHead.
Mark row.wrap before writing spacer_head in this grapheme-wrap path to
keep the intermediate state valid.
When a grapheme expands to width 2 at the screen edge, this path can write
spacer_head before printWrap() sets row.wrap. With an active hyperlink,
printCell triggers hyperlink bookkeeping and page integrity checks in that
intermediate state, causing UnwrappedSpacerHead.
Mark row.wrap before writing spacer_head in this grapheme-wrap path to keep
the intermediate state valid.
This fixes an error if the script was sourced a second time:
bash: __ghostty_ps0: readonly variable
Because this is a non-exported variable, this would only happen if the
script was sourced multiple times in the same bash session.
If an existing PROMPT_COMMAND was a string ending in ; (and maybe some
spaces), we'd add a redundant ;, resulting in a syntax error. Now we
strip any trailing `;[[:space:]]*` characters from the original string
before add ours.
This fixes an error if the script was sourced a second time:
bash: __ghostty_ps0: readonly variable
Because this is a non-exported variable, this would only happen if the
script was sourced multiple times in the same bash session.
If you have multiple splits and start searching naturally the focus
transfers over to the search widget which would apply the unfocused
options. This could make it difficult to view your matches from
searching without re-focusing the surface.
This was discovered when I tested
https://github.com/ghostty-org/ghostty/discussions/11218 (which is a
different issue)
Add `aid=$pid` to 133;A and 133;D for nested shell tracking, and fix the
state comparison which was incorrectly using `constantly` (comparing a
string to a function, which always evaluated to true).
OSC 133;B (input start) and 133;P;k=r (right prompt) cannot be reliably
implemented at the script level because Elvish escapes control
characters in prompt function output, and writing directly to /dev/tty
has timing issues because Elvish renders its prompts on a background
thread. Full semantic prompt support requires a native implementation:
https://github.com/elves/elvish/pull/1917
See: #10523
Add `aid=$pid` to 133;A and 133;D for nested shell tracking, and fix the
state comparison which was incorrectly using `constantly` (comparing a
string to a function, which always evaluated to true).
OSC 133;B (input start) and 133;P;k=r (right prompt) cannot be reliably
implemented at the script level because Elvish escapes control
characters in prompt function output, and writing directly to /dev/tty
has timing issues because Elvish renders its prompts on a background
thread. Full semantic prompt support requires a native implementation:
https://github.com/elves/elvish/pull/1917
See: #10523
This is an update to address common agentic issues I run into,
but the `build.nu` script may be generally helpful to people using
the Nix env since `xcodebuild` is broken by default in Nix due to the
compiler/linker overrides Nix shell does.
I encountered an issue related to
https://github.com/ghostty-org/ghostty/discussions/8641 and
https://github.com/ghostty-org/ghostty/pull/8647, but in `zsh` instead
of `bash`.
One of my aliases is:
```bash
alias sudo='sudo '
```
Which causes following error when sourcing the zsh shell integrations:
```shell
source /usr/share/ghostty/shell-integration/zsh/ghostty-integration
/usr/share/ghostty/shell-integration/zsh/ghostty-integration:149: defining function based on alias `sudo'
/usr/share/ghostty/shell-integration/zsh/ghostty-integration:233: parse error near `()'
```
