Creates a temporary ZDOTDIR to handle zsh shell integration properly while preserving user's zsh configuration. This ensures VSCode shell integration works correctly with zsh without modifying the user's existing setup.
- Add terminalZdotdir setting (disabled by default)
- Create temporary directory with proper security (sticky bit)
- Add automatic cleanup on terminal close
- Add translations for all supported languages
User confirmed fixes:
Fixes: #2205Fixes: #2129
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Added two new terminal settings:
- terminalZshOhMy: Sets ITERM_SHELL_INTEGRATION_INSTALLED=Yes for Oh My Zsh
- terminalZshP10k: Sets POWERLEVEL9K_TERM_SHELL_INTEGRATION=true for Powerlevel10k
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Added a new configuration option 'terminalZshClearEolMark' (default: true) that
sets PROMPT_EOL_MARK='' in the terminal environment. This prevents issues with
command output interpretation when the output ends with special characters like '%'.
Added translations for all supported languages.
Fixes: #2194
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Add a new configuration option that allows users to toggle the PowerShell counter workaround. This workaround adds a counter to PowerShell commands to ensure proper command execution and output capture.
The setting is disabled by default, allowing users to enable it only when needed.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Add a new configurable setting to control command execution delays in terminals.
When set to a non-zero value, this adds a sleep delay after command execution
via PROMPT_COMMAND in bash/zsh and start-sleep in PowerShell.
The default value is 0, which disables the delay completely. This setting
replaces the previous hardcoded delay of 50ms that was added as a workaround
for VSCode bug #237208.
Fixes: #2017
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
When shell integration is unavailable, the UI would hang because the process
was never properly released. This change fixes the issue by:
- Emitting a 'completed' event with a descriptive message
- Marking the terminal as not busy
- Clearing the active stream
- Allowing the process to continue
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Replace hardcoded 3000ms timeout with configurable Terminal.shellIntegrationTimeout
in TerminalProcess.ts. This ensures consistent timeout behavior across all
terminal integration features and allows users to control both timeouts through
a single setting.
The error messages are also updated to display the dynamic timeout value,
providing clearer feedback when shell integration issues occur.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* fix: addLineNumbers handling of empty content
Empty files should not have line numbers, but non-empty files with empty content at a specific line offset should.
- If content is empty, return empty string for empty files
- If content is empty but startLine > 1, return line number for empty content at that offset
This ensures that the model does not think the file contains a single empty line.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* refactor: improve readFileTool XML output format
- Remove unnecessary XML indentation that could confuse the model
- Separate file content from notices and errors using dedicated tags
- Add line range information to content tags
- Handle empty files properly with self-closing tags
- Add comprehensive test coverage
Fixes#2278
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* fix: always show line numbers in read_file XML output
- Always display line numbers in non-range reads
- Improve XML formatting with consistent newlines for better readability
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* test: update tests to match new XML format with line numbers
- Update test expectations to match the new XML format with newlines
- Update tests to expect line numbers attribute in content tags
- Modify test assertions to check for the correct line range values
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* fix: consistent blank line handling in addLineNumbers
- Add newline to all output
- Handle trailing newlines and empty lines consistently
- Add test cases for blank lines:
- Multiple blank lines within content
- Multiple trailing blank lines
- Only blank lines with offset
- Trailing newlines
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* test: use actual addLineNumbers in read-file-xml tests
- Modified extract-text mock to preserve actual addLineNumbers implementation
- Removed mock implementation of addLineNumbers
- Updated test data to account for trailing newline
- Removed unnecessary mock verification
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* test: ensure actual addLineNumbers function is called in tests
- Replace direct mocking of addLineNumbers with spy on actual implementation
- Add verification to ensure the real function is called when appropriate
- Add skipAddLineNumbersCheck option for cases where function should not be called
- Update test cases to use appropriate verification options
- Fix numberedFileContent to include trailing newline for consistency
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* fix: modify readLines to process data directly instead of line by line
- Direct data processing provides more accurate results by preserving exact content with carriage returns
- Improved performance through minimal buffering and efficient string operations
- Use string indexes to find newlines while maintaining their original format
- Handle all edge cases correctly with preserved line endings
- Add tests for various edge cases including empty files, single lines, and different line endings
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* test: remove unused mockInputContent variable
Remove unused variable declaration to appease ellipsis-dev linter requirements.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
---------
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Co-authored-by: Eric Wheeler <roo-code@z.ewheeler.org>
Improve handling of AI-generated line numbers:
- Clamp negative startLine values to 0 instead of throwing errors
- Convert non-integer line numbers to integers automatically
- Only reject when line numbers aren't numbers at all
These changes prevent unnecessary model retries by handling imperfect
AI-generated line numbers gracefully, making the system more robust
when interacting with AI models.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Co-authored-by: Eric Wheeler <roo-code@z.ewheeler.org>
* fix: remove -p flag from test script to prevent git operation errors
The -p flag in npm-run-all was causing tests to run in parallel, which led to 'Cannot log after tests are done' errors with git operations. These errors don't appear when running test:extension alone.
The issue occurs because git-based tests create temporary directories and run async operations that can interfere with each other when executed in parallel. Running tests sequentially resolves this cleanly.
While it might increase total test time slightly, it ensures more reliable and consistent test results.
* refactor(terminal): improve mock streams and fix test issues
- Create shell-specific mock streams (bash, cmd, pwsh) with proper line ending handling
- Fix open handles in tests by properly managing timeouts
- Standardize stderr redirection across all shell implementations using stdio option
- Improve test reliability and output cleanliness
* fix(tests): add skipVerification option to PowerShell tests to debug Linux issues
* fix(tests): use explicit variable name in PowerShell test to fix Linux compatibility
* Refactor terminal tests to use purpose-based approach instead of command mapping
* Remove reference to non-existent test file
* fix: use printf instead of echo -e for more consistent behavior across platforms
* fix: use single quotes for PowerShell commands to preserve variables on Linux
* Update code-qa workflow to run tests on both Windows and Ubuntu
* fix: use platform-specific PowerShell command execution for Linux and Windows
* Fix toggleToolAlwaysAllow to handle path normalization for cross-platform compatibility
* Fix McpHub tests to handle normalized paths on Windows
* Suppress console.error messages in McpHub tests
* fix: make Bedrock ARN regex patterns Windows-compatible
Fixed an issue where AWS Bedrock tests were timing out on Windows but passing on Linux. The root cause was path separator handling in regex patterns used for model ID extraction from ARNs.
1. Updated model ID extraction regex to handle both forward slashes (Linux) and backslashes (Windows)
2. Modified ARN matching regex to be platform-independent
3. Ensured consistent region prefix handling for all supported regions
This change maintains functionality while ensuring cross-platform compatibility.
* fix: make WorkspaceTracker test cross-platform compatible
Fixed an issue where the WorkspaceTracker test 'should initialize with workspace files' was failing on Windows but passing on Linux. The problem was in the mock implementation of toRelativePath that only handled forward slashes.
- Updated the toRelativePath mock to use path.relative which properly handles platform-specific path separators
- Ensured all paths are converted to forward slashes for consistency in test assertions
- The fix maintains cross-platform compatibility while preserving the test's intent
* fix: make WorkspaceTracker tests cross-platform compatible
Fixed cross-platform compatibility issues in the WorkspaceTracker tests that were causing failures on Windows but passing on Linux:
1. Updated the toRelativePath mock implementation to:
- Use path.relative which properly handles platform-specific path separators
- Convert paths to forward slashes for consistency in test assertions
2. Enhanced the 'should not update file paths' test to be platform-agnostic by:
- Using more flexible assertions that don't depend on specific path formats
- Checking file path length and content rather than exact string matches
- Properly typed the test assertions to fix TypeScript errors
These changes preserve the test intent while ensuring they run successfully across different operating systems.
* fix: make McpHub tests cross-platform compatible
Fixed cross-platform compatibility issues in the McpHub tests that were causing failures on Windows but passing on Linux:
1. Made the toggleToolAlwaysAllow tests more platform-agnostic by:
- No longer relying on specific path formats which differ between Windows and Linux
- Using the last write call instead of searching for a specific path string
- Adding more robust assertions that verify structure instead of exact path matches
- Properly handling array existence checks
2. These tests would fail on Windows because paths are formatted with backslashes instead of
forward slashes, causing path equality checks to fail.
The changes maintain test intent while ensuring cross-platform compatibility.
* handle escaping of slash and quote
* fix: ensure consistent line endings in git fallback strategy
Fixed an issue where tests would fail on GitHub Windows runners but pass on local Windows machines due to line ending differences. The fix ensures consistent line ending handling by:
1. Normalizing CRLF to LF when reading files in the git fallback strategy
2. Disabling Git's automatic line ending conversion
3. Maintaining consistent line ending usage throughout text operations
* feat: run tests sequentially on Windows, parallel otherwise
* [clinerules] search clinerule in parent folders which make it easier to share common clinerules for a git repo
* fix indent
* Fix test
---------
Co-authored-by: Ying Liu <lying@microsoft.com>
Co-authored-by: Matt Rubens <mrubens@users.noreply.github.com>
* feat read file range
feat: add file truncation with code structure preview
Add configurable line limit for large file handling with source code structure preview. This improves performance and memory usage while maintaining code readability.
Key changes:
Add maxReadFileLine setting (default: 4500) with UI controls
Create efficient line counter utility using streams
Add parseSourceCodeDefinitionsForFile for single file parsing
Show truncation notice with code structure for large files
Add comprehensive tests for new functionality
Move file extensions to shared constant
add line number
pr comment
Return definition ranges from tree-sitter
make treesitter always add line
add line number to rg search
enhance tree siter output & fix test
update snapshot
fix test
* Add translations, tweak range
---------
Co-authored-by: Matt Rubens <mrubens@users.noreply.github.com>
* fix: clarify PowerShell command completion workaround
The command completion detection approach in PowerShell requires an output
string to allow duplicate commands to execute in some versions of code.
Update the string to explicitly indicate it is a Roo PowerShell workaround,
making it clear in terminal output that this is intentional behavior rather
than a side effect.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* cleanup: improve terminal logging and error handling
No functional changes - purely improves error handling and logging clarity.
Terminal.ts:
- Handle undefined process state in setActiveStream without throwing
- Add terminal IDs to all log messages for better traceability
- Improve error message clarity in shell integration timeout
TerminalRegistry.ts:
- Reorganize shell execution event handlers for better flow
- Log shell execution events before processing for reliable debugging
- Add detailed context to terminal not found scenarios
- Include command and execution state in error messages
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* feat: make terminal shell integration timeout configurable
Users with long shell startup times were encountering "Shell Integration Unavailable" errors due to the hard-coded 4s timeout. The timeout is now configurable through Advanced Settings (1-60s).
Thanks @filthy for troubleshooting and @kiwina for suggesting making the timeout configurable.
Fixes#1654
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* critical fix: race condition that prevents command completion
Terminal running state is now managed in TerminalRegistry instead of Terminal to prevent race between stream close and shell completion.
While this race may not trigger on current VSCode versions, newer releases with additional terminal fixes may expose the issue. This proactively prevents "Shell execution end event received, but process is not running" errors.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* fix: improve command execution path reporting
Enhance clarity of command execution context and error reporting:
- Check to see if the directory changed because of the command
- Clarify execution path message
- Add explicit message when command exits with non-zero code
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* system instructions: clarify terminal directory operations
Clear guidance for the AI system on:
- Working directory constraints
- Path handling requirements
- Tool vs terminal directory behavior
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
* test: update snapshots for system prompt working directory instructions
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
---------
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Co-authored-by: Eric Wheeler <roo-code@z.ewheeler.org>
Co-authored-by: Chris Estreich <cestreich@gmail.com>
feat: add multiple workspaces support
- Add getWorkspacePath function to centralize workspace directory path retrieval
- Use the new workspace directory retrieval logic in Cline, Mentions, ClineProvider, and WorkspaceTracker
- Update WorkspaceFile on tab switch and prevent redundant updates by checking prevWorkSpacePath
- Fix the bug that loads the contents of the previous tab when quickly switching tabs
- Optimize getWorkspacePath return value for better reliability
Co-authored-by: xiong <yueminxiong.xym@alibaba-inc.com>
Add VSCode workspace configuration mock to TerminalProcess.test.ts to handle PowerShell detection in terminal tests, matching the fix in TerminalProcessExec.test.ts
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Mock VSCode workspace configuration to handle PowerShell detection in terminal tests. This prevents TypeError when accessing getConfiguration() in TerminalProcess.run()
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
PowerShell requires special handling for command output due to two issues:
- A sleep delay is required to prevent the ]633;D marker from losing the
original output
- A counter is needed to work around a bug where identical commands are not executed
Changes:
- Add cmdCounter to Terminal class for unique command tracking
- Add PowerShell detection via platform and default shell profile
- Add sleep delay to ensure output is captured before command completion
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Update TerminalProcessExec tests to properly handle shell integration
event sequences:
- Set terminal.running=true before command execution
- Remove duplicate command execution that could trigger extra events
- Replace arbitrary timeout with event-based waiting for output
- Ensure proper event sequence (run -> start -> output -> end)
This aligns the tests with the safeguards added in 62ffa797 that
prevent spurious shell integration events from corrupting terminal
state.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Add explicit checks and error logging to handle problematic event sequence:
0. terminal.running=false
1. terminal.shellIntegration.executeCommand(command)
2. onDidEndTerminalShellExecution // from unexpected 'OSC 633 D' sequence
3. onDidStartTerminalShellExecution
4. stream begins
5. onDidEndTerminalShellExecution
The first onDidEndTerminalShellExecution (from unexpected OSC 633 D) is
ignored because terminal.running is false, preventing process=undefined
from being set prematurely. After the stream begins and sets
terminal.running to true, the second onDidEndTerminalShellExecution
proceeds normally.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Fix issue where background processes (like compilers) couldn't broadcast their
output to new tasks after the launching task was closed. Previously commit
851a4cd prevented terminals from responding to any task except the one that
started them.
The fix allows background terminals (taskId undefined) to act as broadcast
sources that can update any task through getEnvironmentDetails, while still
maintaining proper isolation for task-specific terminals. This enables common
workflows where:
1. A task launches a background compiler
2. That task is closed and a new task is started
3. The new task can still receive compiler errors when making changes
This gives us the best of both worlds:
- Task isolation: Active tasks only see their own terminal output
- Background broadcasting: Background processes can inform any task that needs
their output
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Add Terminal.compressTerminalOutput static method to apply run-length encoding
before truncating terminal output. This significantly reduces output size for
repeated lines while maintaining readability.
- Add compressTerminalOutput static method to Terminal class
- Replace all truncateOutput calls with Terminal.compressTerminalOutput
- Import required functions from extract-text
Test program demonstrating compression:
```python
def generate_repeats():
patterns = [
("A\n", 10), # 10 lines
("AA\n", 100), # 100 lines
("AAA\n", 1000), # 1K lines
("AAAA\n", 10000), # 10K lines
("AAAAA\n", 100000), # 100K lines
("AAAAAA\n", 1000000) # 1M lines
]
for text, count in patterns:
print(text * count, end="")
```
Sample output showing compression:
```
A
A
A
A
A
A
A
A
A
A
AA
<previous line repeated 99 additional times>
AAA
<previous line repeated 999 additional times>
AAAA
<previous line repeated 9999 additional times>
AAAAA
<previous line repeated 99999 additional times>
AAAAAA
<previous line repeated 999999 additional times>
```
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Implement applyRunLengthEncoding function to compress repeated lines in text output:
- Add line repetition compression with count message
- Focus on single line repetitions
- Only compress when beneficial
- Add tests for empty input and single line repetitions
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Use string indices to find line boundaries instead of splitting into array.
This avoids creating large arrays in memory when truncating big inputs.
- Replace split/join with indexOf/lastIndexOf for line counting
- Use slice to extract start/end sections directly from string
- Maintain same 20/80 ratio for before/after content
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
When VSCE start sequence (]633;C or ]133;C) is not received, but the
stream has started:
- Emit no_shell_integration event with clear error message
- Include preOutput in completed event for bug reporting
- Call continue() to ensure proper cleanup
- Return early to prevent further processing
This helps diagnose potential upstream VSCE bugs by providing more context
in the error messages and ensuring proper cleanup of terminal state.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Add timeout and error handling to terminal stream initialization to prevent
UI from freezing when a stream is unavailable or never starts. This ensures
that if the VSCE shell integration stream does not start within 3 seconds:
- The streamAvailable promise is rejected with a clear error
- Event listeners are cleaned up to prevent memory leaks
- Terminal state is properly reset
- Execution continues rather than hanging indefinitely
This fixes a potential deadlock where the UI could freeze waiting for a
stream that never becomes available.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Add descriptive messages to shell integration failures to help users
understand and resolve integration issues more effectively. This improves
the debugging experience by providing specific details about why shell
integration failed.
- Add message parameter to no_shell_integration event
- Update UI to display specific error messages
- Update troubleshooting documentation link
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
As pointed out by @cte, passing and checking terminal IDs in events is
unnecessary since a TerminalProcess instance can never be associated with a
different Terminal instance. The event handling is already properly scoped
to the specific TerminalProcess instance.
- Remove terminal ID parameter from shell_execution_complete event
- Remove terminal ID parameter from stream_available event
- Update all event handlers to remove ID checks
- Update all test cases to match new event signatures
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
@cte reported that the stream property is not used anywhere in the codebase.
The stream is passed directly to the process via event emitter and does not need to be stored.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Update the test file path in the comment to match the actual test file name,
making it easier to run the specific test file directly.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
The TerminalProcess class is a critical component for VSCode shell integration.
This documentation explains why changes must be minimal and carefully considered:
- Performance optimizations using index-based operations and zero-copy implementation
- Accuracy requirements for handling terminal output and escape sequences
- Complex integration with VSCode shell features and command execution
- Careful handling of stream data and escape sequence processing
- Backwards compatibility considerations for VSCode releases
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Notice: this comment required updating system test snapshots with new
execute_command XML schema feature `cwd`
- Add required cwd parameter to Terminal constructor calls in tests
- Use './' for TerminalProcess.test.ts
- Use '/test/path' for TerminalProcessExec.test.ts to match shellIntegration.cwd
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Forcing terminals to `cd` back to the project directory was disrupting
shell state without providing feedback to the model. This caused issues
with capturing output from subsequent commands, particularly with custom
shell prompts.
Instead of forcing directory changes, we now track terminal state
through shell integration with a fallback mechanism, and provide
explicit working directory feedback to the model. This allows terminals
to maintain their natural state while ensuring accurate command output
capture.
Changes:
- Remove forced `cd` commands that were disrupting terminal state
- Add getCurrentWorkingDirectory() method with shell integration fallback
- Add customCwd parameter to executeCommandTool for flexible directory handling
- Add requiredCwd parameter to control terminal selection behavior
- Refactor terminal selection logic for more consistent state management
- Modify environment details to include terminal working directory feedback
- Update XML schema to include optional working directory parameter in execute_command
The environment details now provide explicit feedback about terminal state:
Command executed in terminal N from '/path/to/dir'. Exit code: 0
Fixes: #1388
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
These changes ensure proper isolation by preventing terminal process
output from one Cline task appearing in another task's context when
multiple Cline instances are running in parallel.
- Add taskId parameter to TerminalRegistry.getTerminals to filter terminals by Cline task ID
- Update Cline.ts to use taskId-filtered terminals
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Fix an issue where background running terminals that complete their
execution do not report the final output of their command. Previously,
output was reported while the command was active, but after termination
the remaining output was not provided within the 'inactive terminals'
section of environment details.
- Implement terminal process queue system to track completed processes
- Store command and output retrieval state per process
- Add helper methods to manage the process queue efficiently
- Update getEnvironmentDetails to properly display output from completed processes
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
This fixes an issue where commands that wait for input (like 'cat'
without arguments) would hang indefinitely because no 'line' event was
emitted. The terminal would receive the VSCode shell integration marker
indicating command output has started, but since there was no actual
output yet, the UI would not proceed.
By emitting an empty line event when command output starts, we ensure
the UI can proceed even when a command is waiting for input, preventing
the task from hanging.
Thank you @cte for pointing this out in the PR development process.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
Redirect stderr to /dev/null when executing test commands to prevent
'command not found' messages from appearing in test output. This improves
test output readability while maintaining the same test functionality.
The test still verifies that nonexistent commands return exit code 127,
but does so without printing potentially confusing error messages.
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
- Delete TerminalManager.ts as functionality has been migrated
- Remove TerminalManager import and usage from tests
- Remove outdated TerminalManager references from comments
- Fix TypeScript types in TerminalRegistry event handlers
Signed-off-by: Eric Wheeler <roo-code@z.ewheeler.org>
