AnyEvent::GDB - asynchronous GDB machine interface interface


   use AnyEvent::GDB;


This module is an AnyEvent user, you need to make sure that you use and run a supported event loop.

It implements the GDB MI protocol, which can be used to talk to GDB without having to parse the ever changing command syntax aimed at humans.

It properly quotes your commands and parses the data structures returned by GDB.

At the moment, it's in an early stage of development, so expect changes, and, over time, further features (such as breakpoint-specific callbacks and so on).


To get you started, here is an example program that runs /bin/ls, displaying the stopped information when hitting a breakpoint on _exit:

   use Data::Dump;
   use AnyEvent::GDB;

   our $gdb = new AnyEvent::GDB
      trace => 1,
      on_exec_stopped => sub {
         ddx $_[0];

   my $done

   ddx $gdb->cmd_sync (file_exec_and_symbols => "/bin/ls");
   ddx $gdb->cmd_sync (break_insert => "_exit");
   ddx $gdb->cmd_sync ("exec_run");



Minus vs. underscores

The MI protocol uses - to separate name components, while in Perl, you use _ for this purpose.

This module usually accepts either form as input, and always converts names with - to names with _, so the library-loaded notify might become notify_library_loaded, and the host-name result in that event is stored in the host_name hash element in Perl.

Output redirection

Unfortunately, GDB has no (portable) provision to separate GDB input/output from program input/output. Obviously, without a distinction between program I/O and GDB I/O it becomes impossible to safely control GDB.

There are two ways for you around it: redirect stdin/stdout yourself, or set a tty (eg. with the inferior_set_tty command).

Unfortunately, the MI interface does not seem to support any kind of I/O redirection, so this module helps you a bit, by setting the exec-wrapper variable with a console set commmand. That is, this module does soeQmthing like the following for you, providing proper file descriptors for your actual stdin and stdout:

   set exec-wrapper <&5 >&6

The actual I/O redirection operators are also stored in $gdb->{stdio}, so you can even do it yourself, e.g. when providing your own wrapper:

   $self->cmd_raw ("set exec-wrapper $self->{stdio}", sub { });

(You need to use a raw command, as the "correct" gdb_set MI command silently ignores any exec-wrapper setting).


$gdb = new AnyEvent::GDB key => value...

Create a new GDB object using the given named parameters.

For initial experiments, it is highly recommended to run with tracing or at least verbose enabled. And don't forget to provide an on_eof callback.

   my $gdb = new AnyEvent::GDB
      on_eof => sub {
         print "We are done.\n";
      trace => 1; # or verbose => 1, for less output
exec => $path (default: "gdb")

The path of the GDB executable.

args => [$string...] (default: ["-n"])

An optional array of parameters to pass to GDB. This should not be used to load a program executable, use the file_exec_and_symbols, target_attach or similar MI commands instead.

trace => $boolean (default: 0)

If true, then all commands sent to GDB are printed to STDOUT prefixed with "> ", and all replies received from GDB are printed to STDOUT prefixed with "< ".

verbose => $boolean (default: true if trace is enabled, false otherwise)

If true, then log output and possibly other information is printed to STDOUT.

on_xxxx => $callback->(...)

This specifies a callback for a specific event - see the EVENTS section later in this document.

$gdb->cmd_raw ($command, $cb->($class, $results, $console))

Execute a raw command: $command is sent unchanged to GDB. See cmd_ for a description of the callback arguments.

Example: execute a CLI command and print its output.

   $gdb->cmd_raw ("info sh", sub {
      print "$_[3]\n";
$gdb->cmd ($command => [$option...], $parameter..., $cb->($class, $results, $console))

Execute a MI command and invoke the callback with the results.

$command is a MI command name. The leading minus sign can be omitted, and instead of minus signs, you can use underscores, i.e. all the following command names are equivalent:

   "-break-insert"   # as documented in the GDB manual
   -break_insert     # using underscores and _ to avoid having to quote
   break_insert      # ditto, when e.g. used to the left of a =>
   "break-insert"    # no leading minus

The second argument is an optional array reference with options (i.e. it can simply be missing). Each $option is either an option name (similar rules as with command names, i.e. no initial --) or an array reference with the first element being the option name, and the remaining elements being parameters: [$option, $parameter...].

The remaining arguments, excluding the last one, are simply the parameters passed to GDB.

All options and parameters will be properly quoted.

When the command is done, the callback $cb will be invoked with $class being one of done, connected, error or exit (note: not running), $results being a has reference with all the variable=value pairs from the result list.

$console is an array reference with all the GDB console messages written while command executes (for MI commands, this should always be undef and can be ignored).

Example: #todo#

($results, $console) = $gdb->cmd_sync ($command => [$option...], $parameter...]) =item $results = $gdb->cmd_sync ($command => [$option...], $parameter...])

Like cmd, but blocks execution until the command has been executed, and returns the results if sucessful. Croaks when GDB returns with an error.

This is purely a convenience method for small scripts: since it blocks execution using a condvar, it is not suitable to be used inside callbacks or modules.

That is, unless Coro is used - with Coro, you can run multiple cmd_sync methods concurrently form multiple threads, with no issues.


AnyEvent::GDB is asynchronous in nature, as the goal of the MI interface is to be fully asynchronous. Due to this, a user of this interface must be prepared to handle various events.

When an event is produced, the GDB object will look for the following four handlers and, if found, will call each one in order with the GDB object and event name (without on_) as the first two arguments, followed by any event-specific arguments:

on_event method on the GDB object

Useful when subclassing.

on_event constructor parameter/object member

The callback specified as on_event parameter to the constructor.

on_EVENTNAME method on the GDB object

Again, mainly useful when subclassing.

on_EVENTNAME constructor parameter/object member

Any callback specified as on_EVENTNAME parameter to the constructor.

You can change callbacks dynamically by simply replacing the corresponding on_XXX member in the $gdb object:

   $gdb->{on_event} = sub {
      # new event handler

Here's the list of events with a description of their arguments.

on_eof => $cb->($gdb, "eof")

Called whenever GDB closes the connection. After this event, the object is partially destroyed and must not be accessed again.

on_target => $cb->($gdb, "target", $string)

Output received from the target. Normally, this is sent directly to STDOUT by GDB, but remote targets use this hook.

on_log => $cb->($gdb, "log", $string)

Log output from GDB. Best printed to STDOUT in interactive sessions.

on_TYPE => $cb->($gdb, "TYPE", $class, $results)

Called for GDB exec, status and notify event (TYPE is one of these three strings). $class is the class of the event, with - replaced by _ everywhere.

For each of these, the GDB object will create two events: one for TYPE, and one for TYPE_CLASS. Usuaully you should provide the more specific event (TYPE_CLASS).

on_TYPE_CLASS => $cb->($gdb, "TYPE_CLASS", $results)

Called for GDB exec, status and notify event: TYPE is one of these three strings, the class of the event (with - replaced b _s) is appended to it to form the TYPE_CLASS (e.g. exec_stopped or notify_library_loaded).


The default implementations of the event method store the thread, thread_group, recording, library and running status insid ethe $gdb object.

You can access these at any time. Specifically, the following information is available:


The thread_group member stores a hash for each existing thread group. The hash always contains the id member, but might also contain other members.


The pid member only exists while the thread group is running a program, and contaisn the PID of the program.


The exit_code member only exists after a program has finished executing, and before it is started again, and contains the exit code of the program.


The recording member only exists if recording has been previously started, and is 1 if recoridng is currently active, and 0 if it has been stopped again.


The thread member stores a hash for each existing thread. The hash always contains the id member with the thread id, and the group_id member with the corresponding thread group id.


The running member is 1 while the thread is, well, running, and is missing otherwise.


The stopped member contains the result list from the on_exec_stopped notification that caused the thread to stop, and only exists when the thread is topped.


The library member contains all results from the on_library_loaded event (such as id, target_name, host_name and potentially a thread_group.




   Marc Lehmann <>