Skip to content

MANUAL.md

Latest commit

 

History

History
948 lines (777 loc) · 50.5 KB

MANUAL.md

File metadata and controls

948 lines (777 loc) · 50.5 KB

Table of Contents

Starting a New Debug Session

To start a debugging session, you will need to create a launch configuration for your program. Here's a minimal one:

{
    "name": "Launch",
    "type": "lldb",
    "request": "launch",
    "program": "${workspaceFolder}/<executable file>",
    "args": ["-arg1", "-arg2"],
}

These attributes are common to all CodeLLDB launch configurations:

attribute type
name string Required. Launch configuration name, as you want it to appear in the Run and Debug panel.
type string Required. Set to lldb.
request string Required. Session initiation method:
  • launch to create a new process,
  • attach to attach to an already running process.
    		•
    initCommands [string] LLDB commands executed upon debugger startup. Note that the target is not yet created at this point; if you need to perform an action related to the specific debugging target, prefer using preRunCommands.
    targetCreateCommands [string] LLDB commands executed to create the debug target.
    preRunCommands [string] LLDB commands executed just before launching or attaching to the debuggee.
    processCreateCommands [string] LLDB commands executed to create/attach the debuggee process.
    postRunCommands [string] LLDB commands executed just after launching or attaching to the debuggee.
    gracefulShutdown string ❘ [string] See Graceful Shutdown.
    preTerminateCommands [string] LLDB commands executed just before the debuggee is terminated or disconnected from.
    exitCommands [string] LLDB commands executed at the end of the debugging session.
    expressions string The default expression evaluator type: simple, python or native. See Expressions.
    sourceMap dictionary See Source Path Remapping.
    relativePathBase string Base directory used for resolution of relative source paths. Defaults to "${workspaceFolder}".
    breakpointMode enum Specifies how source breakpoints should be set:
  • path - Resolve locations using full source file path (default).
  • file - Resolve locations using file name only. This option may be useful in lieu of configuring sourceMap, however, note that breakpoints will be set in all files of the same name in the project. For example, Rust projects often have lots of files named "mod.rs".
    		•
    sourceLanguages [string] A list of source languages used in the program. This is used to enable language-specific debugger features.
    reverseDebugging bool Enable reverse debugging.

    Launching a New Process

    These attributes are applicable when the "launch" initiation method is selected:

    attribute type
    program string Path to the executable on the workspace machine (the host where CodeLLDB runs). Required unless you use targetCreateCommands or cargo. This is equivalent to invoking target create <program>; for advanced options such as selecting the target architecture or a remote file path, use targetCreateCommands instead.
    cargo string See Cargo support.
    args string ❘ [string] Command-line parameters. If provided as a string, they are split using shell-like syntax.
    cwd string Working directory for the debuggee.
    env dictionary Environment variables to add on top of those inherited from the parent process (unless LLDB's target.inherit-env setting is false, in which case the initial environment is empty). Reference existing variables with ${env:NAME}; for example: "PATH": "${env:HOME}/bin:${env:PATH}".
    envFile string Path to a file containing additional environment variables. Entries defined in env will override the values loaded from this file.
    stdio string ❘ [string] ❘ dictionary See Stdio Redirection.
    terminal string Destination for the debuggee's stdio streams:
    • console for DEBUG CONSOLE
    • integrated (default) for the VSCode integrated terminal
    • external for a new terminal window
    stopOnEntry boolean Whether to stop the debuggee immediately after launch.

    Launch Sequence

    • Run initCommands.
    • Create the debug target:
      • If targetCreateCommands attribute is present, this command sequence is executed. The currently selected target is assumed to have been created by these commands and will be associated with the current debugging session.
      • Otherwise create the target from program.
    • Apply configuration (args, env, cwd, stdio, etc.).
    • Create breakpoints.
    • The preRunCommands sequence is executed. These commands may alter debug target configuration.
    • Create the process:
      • If processCreateCommands are provided, run them (they must create the process).
      • Otherwise perform the default launch (like process launch).
    • Run postRunCommands.
    • Debug until the program exits or the user terminates session.
    • Attempt to shut down the debuggee gracefully, if requested.
    • Run preTerminateCommands.
    • Terminate the process (if still alive).
    • If restarting, go to preRunCommands step.
    • Run exitCommands.

    Graceful Shutdown

    If the gracefulShutdown attribute is present, CodeLLDB will handle VSCode graceful termination requests:

    • If the value is a string (e.g. "gracefulShutdown": "SIGTERM"), it is treated as a signal name to be sent to the debuggee (not supported on Windows). If the debuggee is currently stopped, it will be resumed so it can receive the signal.
    • If the value is a list of strings, it is interpreted as LLDB commands to execute.

    Stdio Redirection

    stdio is a list mapping file descriptors in order: stdin (0), stdout (1), stderr (2). Rules:

    • null: use the default session terminal (as determined by the terminal property).
    • "/some/path": redirect that stream to a file, pipe, or TTY device*.
    • Fewer than 3 entries: pad to length 3 by repeating the last specified value.
    • More than 3 entries: open additional descriptors (fd 4, 5, ...) accordingly.

    Examples:

    • "stdio": [null, "log.txt", null] – stdin and stderr go to the terminal; stdout goes to log.txt.
    • "stdio": ["input.txt", "log.txt"] – stdin from input.txt; stdout and stderr to log.txt.
    • "stdio": null – all three streams go to the default terminal.

    * Run tty command in a terminal window to find out the TTY device name.

    Attaching to a Running Process

    These attributes are applicable when the "attach" initiation method is selected:

    attribute type
    program string Path to the executable on the workspace machine (the host where CodeLLDB runs). This is equivalent to invoking target create <program>; for advanced options such as selecting the target architecture or a remote file path, use targetCreateCommands instead.
    pid number Process id to attach to. pid may be omitted, in which case debugger will attempt to locate an already running instance of the program. You may also use ${command:pickProcess} or ${command:pickMyProcess} here to choose process interactively.
    stopOnEntry boolean Whether to stop the debuggee immediately after attaching.
    waitFor boolean Wait for the process to launch.

    Attach sequence

    • The initCommands sequence is executed.
    • The debug target object is created:
      • If targetCreateCommands attribute is present, this command sequence is executed. The currently selected target is assumed to have been created by these commands and will be associated with the current debugging session.
      • Otherwise, target is created from the binary pointed to by the program attribute, if one exists.
      • Otherwise, target is created from the process specified by pid.
    • Breakpoints are created.
    • The preRunCommands sequence is executed. These commands may alter debug target configuration.
    • The debugger attaches to the specified process.
      • If processCreateCommands attribute is present, this command sequence is executed. These are expected to have attached debugger to the process corresponding to the debug target.
      • Otherwise, the default attach action is performed (equivalent to the process attach command).
    • The postRunCommands sequence is executed.
    • Debugging until the debuggee exits, or the user requests termination.
    • The preTerminateCommands sequence is executed.
    • The debuggee is detached from.
    • If restarting the debug session, go to preRunCommands step.
    • The exitCommands sequence is executed.

    OS Restrictions

    Operating systems may restrict attaching to running processes, even when the target is owned by the same user. For example, many modern Linux distributions restrict the ptrace syscall by default. You may need to adjust your system settings to enable this capability.

    Pick Process Command

    The ${command:pickProcess} or ${command:pickMyProcess} can be used directly in the configuration for an interactive list of processes running on the machine running Visual Studio Code:

    {
    "name": "Pick Process Attach",
    "type": "lldb",
    "request": "attach",
    "pid": "${command:pickProcess}" // Or pickMyProcess for only processes for the current user.
}

    The lldb.pickProcess and lldb.pickMyProcess commands provide more configuration when used with input variables. The optional initCommands arg let you specify lldb commands to configure a remote connection. The optional filter arg lets you filter the process list to those that match the specified filter.

    {
  "version": "0.2.0",
  "configurations": [
    {
      "name": "Filtered Remote Attach",
      "type": "lldb",
      "request": "attach",
      "pid": "${input:pickExampleProcess}",
      "initCommands": [ ], // Eg, platform select/connect commands.
    },
  ],
  "inputs": [
    {
      "id": "pickExampleProcess",
      "type": "command",
      "command": "lldb.pickProcess",
      "args": {
        "initCommands": [ ], // Eg., platform select/connect commands.
        "filter": "example" // RegExp to filter processes to.
      }
    }
  ]
}

    Debugging Externally Launched Code

    VSCode: URL

    Debugging sessions may be started from outside of VSCode by invoking one of these URLs:

    • vscode://vadimcn.vscode-lldb/launch?name=<configuration name>,[folder=<path>]
      This will start a new debug session using the named launch configuration. The optional folder parameter specifies the workspace folder where the launch configuration is defined. If omitted, all folders in the current workspace will be searched.
      • code --open-url "vscode://vadimcn.vscode-lldb/launch?name=Debug My Project"
    • vscode://vadimcn.vscode-lldb/launch/command?<env1>=<val1>&<env2>=<val2>&<command-line>
      The <command-line> will be split into the program name and arguments array using the usual shell command-line parsing rules.
      • code --open-url "vscode://vadimcn.vscode-lldb/launch/command?/path/filename arg1 \"arg 2\" arg3"
      • code --open-url "vscode://vadimcn.vscode-lldb/launch/command?RUST_LOG=error&/path/filename arg1 'arg 2' arg3"
    • vscode://vadimcn.vscode-lldb/launch/config?<yaml>
      This endpoint accepts a YAML snippet matching one of the above debug session initiation methods. The type and the request attributes may be omitted, and will default to "lldb" and "launch" respectively.
      • JSON-like YAML (if you are not quoting keys in mappings, remember to insert a space after the colon!):
        code --open-url "vscode://vadimcn.vscode-lldb/launch/config?{program: '/path/filename', args: ['arg1','arg 2','arg3']}"
      • Line-oriented YAML (%0A encodes the 'newline' character):
        code --open-url "vscode://vadimcn.vscode-lldb/launch/config?program: /path/filename%0Aargs:%0A- arg1%0A- arg 2%0A- arg3"

    All URLs above are subject to normal URI encoding rules, for example all literal % characters must be escaped as %25.

    VSCode URIs may also be invoked using OS-specific tools:

    • Linux: xdg-open <uri>
    • MacOS: open <uri>
    • Windows: start <uri>

    Examples:

    Attach debugger to the current process (C)
    char command[256];
snprintf(command, sizeof(command), "code --open-url \"vscode://vadimcn.vscode-lldb/launch/config?{'request':'attach','pid':%d}\"", getpid());
system(command);
sleep(1); // Wait for debugger to attach

    Attach debugger to the current process (Rust)

    Ever wanted to debug a build script?

    let url = format!("vscode://vadimcn.vscode-lldb/launch/config?{{'request':'attach','pid':{}}}", std::process::id());
std::process::Command::new("code").arg("--open-url").arg(url).output().unwrap();
std::thread::sleep_ms(1000); // Wait for debugger to attach

    Note: You may need to update your Cargo.toml to build the build script with debug:

    [profile.dev.build-override]
debug = true

    The URL method has limitations:

    • It launches debug session in the last active VSCode window.
    • It does not work with VSCode remoting.

    RPC Server

    You may also initiate debugging sessions via RPC by adding the lldb.rpcServer setting to the workspace or folder configuration. The value of the setting will be passed as the first argument to Node.js's server.listen method, to start listening for connections.

    As a rudimentary security feature, you may add a token attribute to the server options above. In this case, the submitted debug configurations must also contain a token entry with a matching value.

    "lldb.rpcServer": { "host": "127.0.0.1", "port": 12345, "token": "secret" }

    The easiest way to interact with an RPC endpoint is the codelldb-launch utility. When invoked from the integrated terminal, the RPC server address and port will be passed to it automatically.

    Note Every VS Code window hosts its own RPC server, so make sure the TCP port numbers do not collide. You may specify port 0 to let the OS choose an available port; that randomly assigned value is forwarded to codelldb-launch in the integrated terminal, but when launching from an external terminal you must provide the concrete port yourself.

    You can also connect to the RPC server manually:

    • Open a TCP connection using the target port you've configured above.
    • Write the launch configuration as a UTF-8 encoded string.
    • After writing the configuration data, the client must half-close its end of the connection.
    • Upon completion, CodeLLDB will respond with { "success": <true|false>, "message": <optional error message> }.

    Examples:

    These examples assume that you've enabled the RPC server using the configuration above.

    When using an external terminal, define the following environment variables first:

    export CODELLDB_LAUNCH_CONNECT=127.0.0.1:12345
export CODELLDB_LAUNCH_CONFIG="{ token: 'secret' }"
export PATH=<codelldb-extension-dir>/bin:$PATH  # See below
    Start debugging using codelldb-launch
    codelldb-launch /usr/bin/ls
    Start debugging using codelldb-launch with explicit endpoint address and launch configuration
    codelldb-launch --connect=127.0.0.1:12345 --config="{ token: 'secret' }" /usr/bin/ls
    Debug Rust unit tests

    See also the Cargo Support section.

    cargo test --config="target.'cfg(all())'.runner='codelldb-launch'"
    Bazel
    bazel run --run_under=codelldb-launch //<package>:<target>
    Start debugging using netcat
    echo "{ program: '/usr/bin/ls', token: 'secret' }" | netcat -N 127.0.0.1 12345

    codelldb-launch

    codelldb-launch is a CLI utility that talks to the RPC endpoint described above.

    The extension installs codelldb-launch into its bin directory and adds that directory to the PATH used by VS Code's integrated terminals. To locate the executable, run which codelldb-launch (Linux/MacOS), or where codelldb-launch (Windows).

    Usage:

    Variant 1

    Connect to the RPC endpoint, submit debug configuration and exit.

    codelldb-launch --connect=<address> --config=<launch configuration>
    • --connect specifies the address of the CodeLLDB RPC server endpoint to connect to. This address may also be provided via the CODELLDB_LAUNCH_CONNECT environment variable.
    • --config specifies the launch configuration as a YAML or JSON string. This address may also be provided via the CODELLDB_LAUNCH_CONFIG environment variable.

    Variant 2

    codelldb-launch --connect=<address> [--config=<configuration overrides>] [--clear-screen[=true|false]] [--] debuggee [--arg1 [--arg2 ...]]

    In this case the launch configuration is initialized using the debuggee path, command-line arguments, and current environment. The configuration is then updated using the value of the --config flag. The result is submitted to the RPC endpoint, after which codelldb-launch awaits the end of the debug session. The debug session will use the current controlling tty, if there is one, unless configuration overrides contain { "terminal": "integrated|external|console" }.

    • --clear-screen specifies whether to clear current terminal before launching the debuggee.

    Remote Debugging

    For general information on remote debugging please see LLDB Remote Debugging Guide.

    Connecting to lldb-server agent

    • Run lldb-server platform --server --listen *:<port> on the remote machine.
    • Create launch configuration similar to the one below.
    • Start debugging as usual. The executable identified by the program property will be automatically copied to lldb-server's current directory on the remote machine.

    If you require additional configuration of the remote system, you may use preRunCommands sequence to execute commands such as platform mkdir, platform put-file, platform shell, etc. (see help platform for a list of available platform commands).

    {
    "name": "Remote launch",
    "type": "lldb",
    "request": "launch",
    "program": "${workspaceFolder}/build/debuggee", // Local path.
    "initCommands": [
        "platform select <platform>", // For example: 'remote-linux', 'remote-macosx', 'remote-android', etc.
        "platform connect connect://<remote_host>:<port>",
    ],
    "env": {
        "PATH": "...", // See note below.
    }
}

    Connecting to a gdbserver-style agent

    This includes not just gdbserver itself, but also execution environments that implement the gdbserver protocol, such as OpenOCD, QEMU, rr, and others.

    • Start remote agent. For example, run gdbserver *:<port> <debuggee> <debuggee args> on the remote machine.
    • Create a launch configuration.
    • Start debugging.
    {
    "name": "Remote attach",
    "type": "lldb",
    "request": "attach",
    "targetCreateCommands": ["target create ${workspaceFolder}/build/debuggee"],
    "processCreateCommands": ["gdb-remote <remote_host>:<port>"]
}

    Please note that depending on protocol features implemented by the remote stub, there may be more setup needed. For example, in the case of "bare-metal" debugging (OpenOCD), the debugger may not be aware of memory locations of the debuggee modules; you may need to specify this manually:

    target modules load --file ${workspaceFolder}/build/debuggee -s <base load address>

    Debugging as a Different User

    While CodeLLDB does not natively support launching the debuggee as a different user, this can be easily achieved via remote debugging:

    • Start lldb-server under the target user account, for example sudo lldb-server platform --server --listen 127.0.0.1:12345 for root.
      (A copy of lldb-server is provided in this extension's installation directory under lldb/bin. Use the "Extensions: Open Extensions Folder" command to find where extensions are located, and look for "vadimcn.vscode-lldb".)
    • Add the following to your launch configuration:
    "initCommands": [
    "platform select remote-linux", // Replace with "remote-macosx" or "remote-windows" as appropriate
    "platform connect connect://127.0.0.1:12345"
]

    Reverse Debugging

    Also known as Time travel debugging. Provided you use a debugging backend that supports these commands, CodeLLDB can be used to control reverse execution and stepping.

    As of this writing, the only known backend that works is Mozilla's rr. The minimum supported version is 5.3.0.

    There are others mentioned here and here. QEMU reportedly supports record/replay in full system emulation mode. If you get any of them to work, please let me know!

    Example: (using rr)

    Record execution trace:

        rr record <debuggee> <arg1> ...

    Replay execution:

        rr replay -s <port>

    Launch config:

    {
    "name": "Replay",
    "type": "lldb",
    "request": "attach",
    "targetCreateCommands": ["target create ${workspaceFolder}/build/debuggee"],
    "processCreateCommands": ["gdb-remote 127.0.0.1:<port>"],
    "reverseDebugging": true
}

    Inspecting a Core Dump

    Use launch configuration with target create -c <core path> command:

    {
    "name": "Core dump",
    "type": "lldb",
    "request": "attach",
    "targetCreateCommands": ["target create -c ${workspaceFolder}/core"],
    "processCreateCommands": []
}

    Source Path Remapping

    Source path remapping is helpful when the program's source code is located in a different directory than it was at build time (for example, if a build server was used).

    A source map consists of pairs of "from" and "to" path prefixes. When the debugger encounters a source file path beginning with one of the "from" prefixes, it will substitute the corresponding "to" prefix instead. Example:

    "sourceMap": { "/build/time/source/path" : "/current/source/path" }

    This corresponds to the following LLDB command:

    settings set target.source-map '/build/time/source/path' '/current/source/path'

    Parameterized Launch Configurations

    Sometimes you'll find yourself adding the same parameters (e.g. a path of a dataset directory) to multiple launch configurations over and over again. CodeLLDB can help with configuration management in such cases: you can place common configuration values into lldb.dbgconfig section of the workspace configuration, then reference via ${dbgconfig:variable} in launch configurations.
    Example:

    // settings.json
"lldb.dbgconfig":
{
    "dataset": "dataset1",
    "datadir": "${env:HOME}/mydata/${dbgconfig:dataset}" // "dbgconfig" properties may reference each other,
                                                            // as long as there is no recursion.
}

// launch.json
{
    "name": "Debug program",
    "type": "lldb",
    "program": "${workspaceFolder}/build/bin/program",
    "cwd": "${dbgconfig:datadir}" // will be expanded to "/home/user/mydata/dataset1"
}

    Debugger Features

    VSCode Commands
    Show Disassembly... Choose when disassembly is shown. See Disassembly View.
    Toggle Disassembly Choose when disassembly is shown. See Disassembly View.
    Display Format... Choose the default variable display format. See Formatting.
    Toggle Pointee Summaries Choose whether to display pointee's summaries rather than the numeric value of the pointer itself. See Pointers.
    Display Options... Interactive configuration of the above display options.
    Attach to Process... Choose a process to attach to from the list of currently running processes.
    Generate Cargo Launch Configurations Generate all possible launch configurations (binaries, examples, unit tests) for the current Cargo.toml file. The resulting list will be opened in a new text editor, from which you can copy/paste the desired sections into launch.json.
    Command Prompt Open LLDB command prompt in a terminal, for managing installed Python packages and other maintenance tasks.
    View Memory... View raw memory starting at the specified address.
    Search Symbols... Search for a substring among the debug target's symbols.
    Use Alternate Backend... Choose alternate LLDB instance to be used instead of the bundled one. See Alternate LLDB backends.
    Run Self-Test Run diagnostic test to make sure that the debugger is functional.

    Debugger Commands

    CodeLLDB adds in-debugger commands that may be executed in the DEBUG CONSOLE panel during a debug session:
    debug_info Provides tools for investigation of debugging information. See debug_info -h for options.
    nofail nofail <other command> prevents errors in the execution of the specified command from aborting the current command sequence. For example, "postRunCommands": ["nofail process interrupt", "break set ..."] will execute the second command even if the first one fails (e.g., because the process is already stopped).

    Debug Console

    The VS Code DEBUG CONSOLE in CodeLLDB has two jobs:

    Input is treated as LLDB commands by default. To evaluate an expression instead, prefix it with ?, e.g. ?a+2. Expression-type prefixes stack on top of that, so ?/nat a.size() forces the native evaluator.

    Use the lldb.consoleMode setting to change the default behavior:

    • commands (default) – input is a command unless it starts with ?.
    • evaluate – input is evaluated as an expression; commands must be prefixed with /cmd or a backtick (`).
    • split – expressions stay in the DEBUG CONSOLE, while LLDB commands are run in a separate terminal.

    Regex Breakpoints

    Function breakpoints prefixed with '/re ', are interpreted as regular expressions. This causes a breakpoint to be set in every function matching the expression. The list of created breakpoint locations may be examined using the break list command.

    Conditional Breakpoints

    You may use any of the supported expression syntaxes to create breakpoint conditions. When a breakpoint condition evaluates to False, the breakpoint will not be stopped at. Any other value (or expression evaluation error) will cause the debugger to stop.

    Data Breakpoints

    Data breakpoints (or "watchpoints" in LLDB terms) allow monitoring memory locations for changes. You can create data breakpoints by choosing "Break When Value Changes" from context menu in the Variables panel. (To access advanced features, such as breaking on memory reads, use LLDB watch command).

    Note that data breakpoints require hardware support, and, as such, may come with restrictions, depending on CPU platform and OS support. For example, on x86_64 the restrictions are as follows:

    • The monitored memory region must be 1, 2, 4 or 8 bytes in size.
    • There may be at most 4 data watchpoints.

    Hit conditions

    Syntax:

    operator :: = '<' | '<=' | '=' | '>=' | '>' | '%'
hit_condition ::= operator number

    The '%' operator causes a stop after every number of breakpoint hits.

    Logpoints

    Expressions embedded in log messages via curly brackets may use any of the supported expression syntaxes.

    Disassembly View

    When execution steps into code for which debug info is not available, CodeLLDB will automatically switch to disassembly view. This behavior may be controlled using Show Disassembly and Toggle Disassembly commands. The former allows to choose between never, auto (the default) and always, the latter toggles between auto and always.

    While in disassembly view, 'step over' and 'step into' perform instruction-level stepping rather than source-level stepping.

    disassembly view

    Excluded Callers

    You may want to skip breakpoints when triggered from specific call paths — especially for "on throw" exception breakpoints in code that uses exceptions as part of normal control flow.

    When stopped on a breakpoint, you can right-click a frame in the CALL STACK panel and choose the "Exclude Caller" item. Afterwards, the debugger won't stop on that breakpoint location, if the excluded caller appears anywhere in the call stack. You can see and manage current exclusions in the EXCLUDED CALLERS panel.

    Formatting

    You may change the default display format of evaluation results using the Display Format command.

    When evaluating expressions in the DEBUG CONSOLE or in WATCH panel, you may control formatting of individual expressions by adding one of the suffixes listed below:

    suffix format
    c Character
    x Hex
    o Octal
    d Decimal
    u Unsigned decimal
    b Binary
    f Float (reinterprets bits, no casting is done)
    p Pointer
    s C string
    y Bytes
    Y Bytes with ASCII
    [<num>] Reinterpret as an array of <num> elements.

    For example, evaluation of var,x will display the value of var formatted as hex. It is also possible to combine number format and array specifiers like this: var,x[10].

    Pointers

    Pointer and reference values display the pointee by default. To view the raw address:

    • Use the Toggle Pointee Summaries command, or
    • Add the pointer to WATCH and apply a numeric format (e.g. ,p or ,x).

    Expressions

    CodeLLDB provides three expression evaluators: "simple", "python", and "native". They are used anywhere expressions are accepted: WATCH panel, DEBUG CONSOLE (inputs prefixed with ?), and breakpoint conditions.
    Default evaluator: simple. Override globally via the expressions launch property, or per expression with a prefix.

    Simple expressions

    Prefix: /se
    Simple expressions support arithmetic and logical operations directly on formatted views of values. Indexing an std::vector or comparing an std::string to a literal should “just work”.

    The following features are supported:

    • References to variables: all identifiers are assumed to refer to variables in the debuggee current stack frame. The identifiers may be qualified with namespaces and template parameters (e.g. std::numeric_limits<float>::digits).
    • Embedded native expressions: these must be delimited with ${ and }.
    • Literals: integers, floats, strings, booleans (true, false).
    • Binary operators: +, -, *, /, %, <<, >>, &, ^, |, ==, !=, >, >=, <, <=, &&, || with the same precedence/associativity as in C++. Note that division is the Python "true division", use // for integer division.
    • Unary operators: +, -, !, ~, * (dereference), & (address-of).
    • Member access: <expr>.<attr>, <expr>-><attr>
    • Indexing: <expr>[<expr>].
    • Pythonisms:
      • **: exponentiation (higher precedence than multiplication, right-associative)
      • // integer division (same precedence as division)
      • Aliases for C++ operators: and for &&, or for ||, not for !.
      • True and False aliases for true and false.

    Python expressions

    Prefix: /py
    Python expressions support full Python syntax. Any identifier prefixed by $ is substituted with the corresponding debuggee variable; these values can be mixed with normal Python variables. For example, /py [math.sqrt(x) for x in $arr] will evaluate to a list of square roots of the values contained in the array variable arr.

    Python execution shares the debugger’s embedded interpreter state and honors prior script ... commands. For example, in order to evaluate math.sqrt(x) above, you'll need to have imported the math package via script import math. To import Python modules on debug session startup, use "initCommands": ["script import ..."].

    Technical note
    Evaluation of Python expressions is performed as follows:

    • First, the expression is preprocessed and all tokens starting with '$' are replaced with calls to the __expr() function, For example, the expression [math.sqrt(x) for x in $arr] will be re-written as [math.sqrt(x) for x in __eval('arr')]
    • The resulting string is evaluated by the Python interpreter, with the __eval() function performing variable lookups and evaluation of native expressions, returning instances of Value.

    Native expressions

    Prefix: /nat
    Native expressions use LLDB's built-in expression evaluators. The specifics depend on source language of the current debug target (e.g. C, C++ or Swift).
    For example, the C++ expression evaluator offers many powerful features including interactive definition of new data types, instantiation of C++ classes, invocation of functions and class methods, and more.

    Note, however, that native evaluators ignore data formatters and operate on "raw" values, thus they are often not as convenient as "simple" or "python" expressions.

    Agent-assisted debugging

    CodeLLDB provides tools that allow LLM agents to interact with active debug sessions. To request agent assistance, simply mention #codelldb in your prompt:

    Debug this for me! #codelldb

    Python Scripting

    Debugger API

    CodeLLDB provides extended Python API via the codelldb module (also aliased as debugger), which is auto-imported into debugger's main script context:

    # codelldb

def get_config(name: str, default: Any = None) -> Any:
    '''Retrieve a configuration value from the adapter settings.
        name:    Dot-separated path of the setting to retrieve.  For example, `get_config('foo.bar')`,
                 will retrieve the value of `lldb.script.foo.bar` from VSCode configuration.
        default: The default value to return if the configuration value is not found.
    '''
def evaluate(expr: str, unwrap: bool = False) -> Value | lldb.SBValue:
    '''Performs dynamic evaluation of native expressions returning instances of Value or SBValue.
        expression: The expression to evaluate.
        unwrap: Whether to unwrap the result and return it as lldb.SBValue
    '''
def wrap(obj: lldb.SBValue) -> Value:
  '''Wrap an lldb.SBValue as a Value.'''
def unwrap(obj: Value) -> lldb.SBValue:
  '''Extract the underlying lldb.SBValue from a Value.'''
def create_webview(html: Optional[str] = None, title: Optional[str] = None, view_column: Optional[int] = None,
                   preserve_focus: bool = False, enable_find_widget: bool = False,
                   retain_context_when_hidden: bool = False, enable_scripts: bool = False):
    '''Create a webview panel.
        html:               HTML content to display in the webview.  May be later replaced via Webview.set_html().
        title:              Panel title.
        view_column:        Column in which to show the webview.
        preserve_focus:     Whether to preserve focus in the current editor when revealing the webview.
        enable_find_widget: Controls if the find widget is enabled in the panel.
        retain_context_when_hidden: Controls if the webview panel retains its context when it is hidden.
        enable_scripts:     Controls if scripts are enabled in the webview.
    '''
def start_debugging(request, configuration):
    '''Start a new debugging session.'''

    Webview

    A simplified interface for webview panels.

    class Webview:
    def dispose(self):
        '''Destroy webview panel.'''
    def set_html(self, html: str):
        '''Set HTML contents of the webview.'''
    def reveal(self,  view_column: Optional[int] = None, preserve_focus: bool = False):
        '''Show the webview panel in a given column.'''
    def post_message(self, message: Any):
        '''Post a message to the webview content.'''
        interface.send_message(dict(message='webviewPostMessage', id=self.id, inner=message))
    @property
    def on_did_receive_message(self) -> Event:
        '''Fired when webview content posts a new message.'''
    @property
    def on_did_dispose(self) -> Event:
        '''Fired when the webview panel is disposed (either by the user or by calling dispose())'''

    Event

    class Event:
    def add(self, listener: Callable[[Any]]):
        '''Add an event listener.'''
    def remove(self, listener: Callable[[Any]]):
        '''Remove an event listener.'''

    Value

    Value objects (source) are proxy wrappers around lldb.SBValue, which add implementations of standard Python operators.

    Installing Packages

    CodeLLDB bundles its own copy of Python, which may be different from the version of your default Python. As such, it likely won't be able to use third-party packages you've installed through pip. In order to install packages for use in CodeLLDB, you will need to use the LLDB: Command Prompt command in VSCode, followed by pip install --user <package>.

    Stdio in Python scripts

    • stdout output will be sent to the Debug Console
    • stderr output will be sent to the Output/LLDB panel

    Alternate LLDB Backends

    Even though CodeLLDB bundles a complete LLDB package for every supported platform, you may still want to point the extension at a different backend - for example, a custom LLDB build that contains Swift-specific extensions.

    Liblldb

    CodeLLDB embeds LLDB by loading its dynamic library (liblldb) and driving the debugger through the API it exports rather than spawning the lldb command-line tool (as some other extensions do). To switch to an alternate backend, point lldb.library at the desired liblldb shared library (which must be v15.0 or later).

    To make locating liblldb easier, CodeLLDB provides the Use Alternate Backend... command. It prompts for the path to the main lldb executable and then uses it to locate the corresponding shared library.

    How to find liblldb manually:

    • Linux: $LLDB_INSTALL_ROOT/lib/liblldb.so.<version>,
      <lldb install root> is wherever you've installed LLDB, or /usr, if it's a standard distro package.
    • MacOS: $LLDB_FRAMEWORK/LLDB if built as Apple framework, $LLDB_INSTALL_ROOT/lib/liblldb.<version>.dylib otherwise.
      LLDB_FRAMEWORK is typically located under /Library/Developer/<toolchain>/.../PrivateFrameworks.
    • Windows: $LLDB_INSTALL_ROOT/bin/liblldb.dll.

    LLDB Server

    LLDB always employs a client-server architecture, even for local debugging (except on Windows). LLDB Server is the binary that runs and monitors the debugged program in response to requests of the main debugger.

    In most cases LLDB can locate the server binary automatically; however, sometimes this fails and you may see errors similar to this: "Unable to locate lldb-server-<version>". In such cases you may override the default search logic via the lldb.server setting.

    Rust Language Support

    CodeLLDB will attempt to locate and load LLDB data formatters provided by the Rust toolchain. By default, the configured toolchain of your workspace root will be used, however this can be overridden via these configuration settings:

    • lldb.script.lang.rust.toolchain - override toolchain name, for example beta.
    • lldb.script.lang.rust.sysroot - set toolchain sysroot directly, for example /home/user/.rustup/toolchains/beta-x86_64-unknown-linux-gnu.

    To enable this feature, add "sourceLanguages": ["rust"] into your launch configuration.

    Cargo Support

    CodeLLDB has built-in support for Cargo workspaces, so you don’t need to manually configure the paths to binaries or test targets. To use this feature, replace the program property in your launch configuration with cargo:

    {
    "type": "lldb",
    "request": "launch",
    "cargo": {
        "args": ["test", "foo", "--", "--test-threads=3"], // Cargo command line to run the debug target
        // Optional fields:
        "env": { "RUSTFLAGS": "-Clinker=ld.mold" },        // Extra environment variables
        "cwd": "${workspaceFolder}",                       // Cargo working directory
        "problemMatcher": "$rustc",                        // Problem matcher(s) for Cargo output
    },
    "args": ["--test=test1"]  // These arguments will be appended to those passed to the debug target by Cargo
}

    If you only need to provide cargo arguments, this may be shortened to:

    {
    "type": "lldb",
    "request": "launch",
    "cargo": ["test", "foo", "--", "--test-threads=3", "--test=test1"],
}

    Settings

    Workspace Settings

    The "Workspace Settings" term used in the document refers to the combined view of VSCode User and Workspace settings, merged according to the settings precedence hierarchy.

    Launch Configurations Settings

    VSCode launch configuration settings are distinct from workspace settings and are not subject to the usual settings merging described above.

    However, since common defaults for all launch configurations in a project are often desired, the CodeLLDB extension provides this feature via lldb.launch.* setting group, which serve as defaults for the corresponding launch configuration settings. When a setting is specified in both locations, the values will be merged according to their type:

    • For lists, the resulting value will be a concatenation of values from both sources.
    • For dictionaries, the resulting value will be a combination of key-value pairs from both sources. For equal keys, the launch configuration value takes precedence.
    • For numbers and strings, the launch configuration value takes precedence.

    Variable Substitution in Launch Configurations

    Before being sent to the debug adapter, launch configuration settings undergo expansion of variable references. In addition to the standard expansions performed by VSCode, CodeLLDB also expands references to ${dbgconfig:<name>} as well as ${cargo:program}.

    LLDB Settings

    The LLDB debugger engine also has a number of internal settings, which affect its behavior. These may be changed using settings set <key> <value> command, which may be put into any of the *Commands launch configuration sequences (usually initCommands).

    The full list of LLDB settings may be obtained by executing settings list command during a debug session (or in LLDB command prompt).

    Workspace Configuration Reference

    Default Launch Configuration Settings

    These settings specify the default values for launch configuration settings of the same name.
    lldb.launch.initCommands Commands executed before initCommands of individual launch configurations.
    lldb.launch.preRunCommands Commands executed before preRunCommands of individual launch configurations.
    lldb.launch.postRunCommands Commands executed before postRunCommands of individual launch configurations.
    lldb.launch.gracefulShutdown Commands executed after gracefulShutdown of individual launch configurations.
    lldb.launch.preTerminateCommands Commands executed after preTerminateCommands of individual launch configurations.
    lldb.launch.exitCommands Commands executed after exitCommands of individual launch configurations.
    lldb.launch.env Additional environment variables that will be merged with 'env' of individual launch configurations.
    lldb.launch.envFile The default envFile path.
    lldb.launch.cwd The default program working directory.
    lldb.launch.stdio The default stdio destination.
    lldb.launch.expressions The default expression evaluator.
    lldb.launch.terminal The default terminal type.
    lldb.launch.sourceMap Additional entries that will be merged with 'sourceMap's of individual launch configurations.
    lldb.launch.breakpointMode The default breakpoint resolution mode.
    lldb.launch.relativePathBase The default base directory used for resolution of relative source paths. Defaults to "${workspaceFolder}".
    lldb.launch.sourceLanguages A list of source languages used in the program. This is used to enable language-specific debugger features.

    General
    lldb.dbgconfig See Parameterized Launch Configurations.
    lldb.evaluationTimeout Timeout for expression evaluation, in seconds (default=5s).
    lldb.displayFormat The default format for variable and expression values.
    lldb.showDisassembly When to show disassembly:
  • auto - only when source is not available.
  • never - never show.
  • always - always show, even if source is available.
    		•
    lldb.dereferencePointers Whether to show summaries of the pointees instead of numeric values of the pointers themselves.
    lldb.suppressMissingSourceFiles Suppress VSCode's messages about missing source files (when debug info refers to files not available on the local machine).
    lldb.consoleMode Controls whether the DEBUG CONSOLE input is by default treated as debugger commands or as expressions to evaluate:
  • commands - treat debug console input as debugger commands. In order to evaluate an expression, prefix it with '?' (question mark).",
  • evaluate - treat DEBUG CONSOLE input as expressions. In order to execute a debugger command, prefix it with '/cmd ' or '`' (backtick),
  • split - (experimental) use the DEBUG CONSOLE for evaluation of expressions, open a separate terminal for LLDB console.
    		•
    lldb.script Configuration settings provided to Python scripts running in the context of CodeLLDB. These may be read via get_config().

    Advanced
    lldb.verboseLogging Enables verbose logging. The log can be viewed in OUTPUT/LLDB panel.
    lldb.rpcServer See RPC server.
    lldb.library See Alternate Backends.
    lldb.server See Alternate Backends.
    lldb.adapterEnv Extra environment variables passed to the debug adapter.
    lldb.cargo Executable used for Cargo invocations. Override this if Cargo is not on the PATH or if you want to use an alternative tool/wrapper.
    lldb.evaluateForHovers Enable value preview when cursor is hovering over a variable.
    lldb.commandCompletions Enable command completions in DEBUG CONSOLE.
    lldb.useNativePDBReader Use the native reader for the PDB debug info format: LLDB includes two readers for the Microsoft PDB format: one based on the Microsoft DIA SDK and another implemented natively in LLDB. Currently, the DIA-based reader is the default because it is more complete. However, the native reader is faster and may be preferred for large binaries.