App::BCVI::Plugins - Documentation about the bcvi plugin API


BCVI plugins are .pm files (Perl modules) in the user's BCVI config directory ($HOME/.config/bcvi).

Plugins can:

  • add new command-line options (and add code to process those options)

  • add new back-channel commands (and add code to implement those commands in the listener process)

  • modify almost any existing functionality of bcvi (including removing functionality)

  • modify both the server (listener) and/or the client

Ideally you should be able to customise the behaviour of bcvi in pretty much any way you want without needing to edit the bcvi script itself.


Here's a silly plugin (that no sane person would ever want to use) which overrides the 'vi' command handler and instead of launching gvim it launches gedit (the GNOME text editor) - I did warn you it was a silly example:

  package App::BCVI::Gedit;
  use strict;
  use warnings;
  sub execute_vi {
      my($self) = @_;

      my $alias = $self->calling_host();
      my @files = map { "sftp://${alias}$_" } $self->get_filenames();
      system('gedit', '--', @files);

This file should be saved as $HOME/.config/bcvi/ Let's go through it line-by-line.

Each plugin must have a unique package name. The App::BCVI namespace is there for plugins to use. By convention, the filename should match the last part of the package name, with '.pm' appended.

The use strict; and use warnings; are good practice in any Perl module.

The execute_vi subroutine was copy/pasted from the bcvi script itself and then modified to work with gedit rather than gvim.

The hook_server_class line is a method call that pushes this class onto the inheritance chain for the object class that implements the listener process. When the listener process calls execute_vi in response to a request from a client, our method is called instead of the standard method. In some plugins, it might make sense to delegate to the standard method using the syntax $self->SUPER::execute_vi(@args), but in our case we're replacing the standard method rather than augmenting it.


Plugin files are never loaded from anywhere except the user's BCVI config directory. In particular, bcvi never loads any modules from the system lib/App/BCVI directory. If you get plugin modules from CPAN, you'll need to copy the .pm files into your plugin directory (or possibly symlink to the .pm file in the system lib directory).

Some plugins enhance the listener process and therefore only need to be installed on your workstation. Other plugins enhance the client so they need to be installed on the hosts where you use bcvi. Client-side plugins can register themselves to be included in the set of files that get deployed to a host when you run bcvi --install HOSTNAME.


The BCVI application is built from four classes:


Implements the listener process as a forking server. Listens on a socket, when an incoming connection is received, a child process is forked off to handle it.


Implements the client process which establishes a TCP connection to the listener process, sends a request and waits for a response.


A base class implements common methods used by both the client and the server.


A helper class used by both the client and the server to render POD to text in response to the --help option.

A plugin can push its package name onto the inheritance chain for the server by calling:


or for the client by calling


There are currently no hook methods for either the base class or the POD class because that didn't seem very useful (just ask if you really need this).

The example plugin above had a package name of App::BCVI::Gedit and it called hook_server_class(). This has two effects:

  1. When a listener process is started, it will be an instance of the App::BCVI::Gedit class

  2. The @ISA array in the App::BCVI::Gedit package will be adjusted to point to App::BCVI::Server so that all the existing methods of the server class will be inherited

If another package calls hook_server_class() then its @ISA array will be adjusted to point to the App::BCVI::Gedit class and when the listener starts it will be an instance of the second plugin class. Usually the order of loading would not be significant, but the plugin filenames are sorted alphanumerically before loading so you can rename the .pm files to have them load in a specific order.

If your plugin calls a hook method it should not explicitly set up any other inheritance relationship (either through use base or by directly altering @ISA).

Sometimes it might not be immediately obvious whether you need to hook the client class or the server class. For example if your code modifies the behaviour of the --install option then it would not be a part of the listener process but it also might not run on a remote host. The rule in these cases is: If your code does not run in the listener then it should hook the client class.

A single plugin should not call both hook_server_class() and hook_client_class() - no good can come of that.


In addition to being able to hook into the inheritance chains, a plugin can also choose to call one of the registration methods:

register_option(key => value, ...)

This method is used to register a new command-line option. The arguments are key => value pairs, for example:

      name        => 'command',
      alias       => 'c',
      arg_spec    => '=s',
      arg_name    => '<cmnd>',
      summary     => 'command to send over back-channel',
      description => <<'END_POD'
  Use C<cmnd> as the command to send over the back-channel (default: vi).
  Recognised commands are described in L<COMMANDS> below.

The recognised keys are (*=mandatory parameter):

  *name         the long form of the option name (without the initial '--')
   alias        optional single character alias
   arg_spec     if the option takes a value use '=s' for string '=i' for int etc
   arg_name     how the option value should be rendered in the POD
   dispatch_to  name of a method to be called if this option is present
  *summary      one-line summary of the option for the synopsis
  *description  longer POD snippet providing a full description of the option

The command line options are parsed using Getopt::Long so you can refer to that module's documentation for more details (of the arg_spec in particular).

If your plugin registers a command-line option then your summary and description should be visible immediately when you run bcvi --help.

Only specify a dispatch_to method if bcvi should exit immediately after your method is called.

After you have registered a command-line option, code in your plugin methods can check the value of the option (or any other option) with:


If you are unsure about the usage of any of the parameters listed above, please refer to the numerous examples in bcvi itself.

register_command(key => value, ...)

This method is used to register a handler for a new command in the listener. The arguments are key => value pairs, for example:

      name        => 'scpd',
      description => <<'END_POD'
  Uses C<scp> to copy the specified files to the calling user's F<~/Desktop>.

The recognised keys are (*=mandatory parameter):

  *name         the 'command' name which will be sent from the client
   dispatch_to  name of the handler method
  *description  POD snippet providing a full description of the command

If you don't provide a method name as an argument to the dispatch_to parameter, then the default handler method name will be the command name with 'execute_' prepended.

See "COMMAND HANDLERS" below for details of how the handler method is called.

register_aliases(alias, ...)

This method is used to register shell alias definitions that should be added to the user's local shell startup script with bcvi --add-aliases or to the shell startup script on a remote host with bcvi --install.

One call can register a list of aliases, for example:

      'test -n "${BCVI_CONF}"  && alias vi="bcvi"',
      'test -n "${BCVI_CONF}"  && alias bcp="bcvi -c scpd"',


A client-side plugin should call this method to indicate that the plugin file is required on the remote hosts and should be copied over by bcvi --install.

This method call requires no arguments:



When the listener receives a command it looks up the registered commands to locate a handler method and then calls that method (with no arguments).

If the handler method expects a list of filenames, it can get them by calling:


Alternatively, if the handler method expects string data rather than filenames, it can call:


for non-ASCII text data you may want to decode the bytes to characters using the Encode module:

  decode('utf8', $self->read_request_body())

The handler can also access the request headers via the hashref returned by:


If for some reason the handler method needs direct read or write access to the client socket, it can get the socket filehandle with:


Response Status Codes

You probably don't need to worry about this section - usually a handler does not need to worry about returning a status code at all.

On successful completion, a command handler method should simply return (the return value is not significant). The listener process will send a 200 Success status response.

On failure, a command handler may choose to die and the message will go to the user's X Session log. The client will see the socket close and will advise the user that the "Server hung up".

There are a small number of predefined status codes that can be returned to the client (but most command handlers will never need to use them):

  200   Success
  300   Response follows
  900   Permission denied
  910   Unrecognised command

You can send a response by calling:

  $self->send_response($code)   # eg: $code = 900

There is currently no way to register additional codes, but of course a handler routine could make up its own status code, write it directly to the socket (using $self->sock->write) and then exit.

The '300' response is useful for the situation where the client sent a request and is expecting data in the body of the response. If you want to see an example of this functionality, look at the built-in 'commands_pod' message that the bcvi client uses to retrieve the POD for all commands supported by the listener. A 300 response must be followed by one or more headers - terminated by a blank line. A 'Content-length' header must be included to indicate how many bytes of data follow the headers.


For examples of plugins, look for these modules on CPAN:


Implements the client-side of the Desktop notification plugin. Registers a shell alias and registers as an installable file.


Implements the server-side of the Desktop notification plugin. Registers a new command, hooks the server class and implements a command handler.


Hooks the client class to track which hosts bcvi has been installed to (using bcvi --install). Wraps the handler for the existing --install option handler and also adds a new --update-all option.

The source of bcvi itself is also a good place to look for examples of how to register options and commands and how to implement a command handler.


Copyright 2007-2012 Grant McLean <grantm at>

This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.