Tutorial: Programmable Tab-Completion

Overview

As with many other shells, xonsh ships with the ability to complete partially-specified arguments upon hitting the “tab” key.

In Python-mode, pressing the “tab” key will complete based on the variable names in the current builtins, globals, and locals, as well as xonsh language keywords & operators, files & directories, and environment variable names. In subprocess-mode, xonsh additionally completes based on the names of any executable files on your $PATH, alias keys, and full Bash completion for the commands themselves.

xonsh also provides a mechanism by which the results of a tab completion can be customized (i.e., new completions can be generated, or a subset of the built-in completions can be ignored).

This page details the internal structure of xonsh’s completion system and includes instructions for implementing new tab completion functions.

Structure

xonsh’s built-in completers live in the xonsh.completers package, and they are managed through an instance of OrderedDict (__xonsh__.completers) that maps unique identifiers to completion functions.

The completers are divided to exclusive completers and non-exclusive completers. Non-exclusive completers are used for completions that are relevant but don’t cover the whole completions needed (e.g. completions for the built-in commands and/or).

When the “tab” key is pressed, xonsh loops over the completion functions in order, calling each one in turn and collecting its output until it reaches an exclusive one that returns a non-empty set of completions for the current line. The collected completions are then displayed to the user.

Listing Active Completers

A list of the active completers can be viewed by running the completer list command. This command will display names and descriptions of the currently-active completers, in the order in which they will be checked.

Writing a New Completer

Completers are implemented as Python functions that take a Completion Context object. Examples for the context object:

# ls /tmp/<TAB>
CompletionContext(
    command=CommandContext(
        args=(CommandArg(value='ls'),),
        arg_index=1, prefix='/tmp/',
        ),
    python=PythonContext(multiline_code="ls /tmp/", cursor_index=8, ctx={...})
)

# ls $(whic<TAB> "python") -l
CompletionContext(
    command=CommandContext(
        args=(CommandArg(value='python', opening_quote='"', closing_quote='"'),),
        arg_index=0, prefix='whic', subcmd_opening='$(',
    ),
    python=None
)

# echo @(sys.exe<TAB>)
CompletionContext(
    command=None,
    python=PythonContext(
        multiline_code="sys.exe", cursor_index=7,
        is_sub_expression=True, ctx={...},
    )
)

Note

Xonsh still supports legacy completers - see Legacy Completers Support. For backwards-compatibility, contextual completers need to be marked (as seen in the examples).

This function should return a python set of possible completions for command.prefix in the current context. If the completer should not be used in this case, it should return None or an empty set, which will cause xonsh to move on and try to use the next completer.

Occasionally, completers will need to return a match that does not actually start with prefix. In this case, a completer should instead return a tuple (completions, prefixlength), where completions is the set of appropriate completions, and prefixlength is the number of characters in line that should be treated as part of the completion.

Note

Further completion customizations can be made using the RichCompletion object - see Advanced Completions.

The docstring of a completer should contain a brief description of its functionality, which will be displayed by completer list.

Some simple examples follow. For more examples, see the source code of the completers xonsh actually uses, in the xonsh.completers module.

# Helper decorators for completers:
from xonsh.completers.tools import *

@contextual_completer
def dummy_completer(context):
    '''
    Completes everything with options "lou" and "carcolh",
    regardless of the value of prefix.
    '''
    return {"lou", "carcolh"}

@non_exclusive_completer
@contextual_completer
def nx_dummy_completer(context):
    '''
    Like dummy_completer but its results are ADDED to the other completions.
    '''
    return {"lou", "carcolh"}

@contextual_completer
def python_context_completer(context):
    '''
    Completes based on the names in the current Python environment
    '''
    if context.python:
        last_name = context.python.prefix.split()[-1]
        return {i for i in context.python.ctx if i.startswith(last_name)}

@contextual_completer
def unbeliever_completer(context):
    '''
    Replaces "lou carcolh" with "snail" if tab is pressed after at least
    typing the "lou " part.
    '''
    if (
        # We're completing a command
        context.command and
        # We're completing the second argument
        context.command.arg_index == 1 and
        # The first argument is 'lou'
        context.command.args[0].value == 'lou' and
        # The prefix startswith 'carcolh' (may be empty)
        'carcolh'.startswith(context.command.prefix)
    ):
        return {'snail'}, len('lou ') + len(context.command.prefix)

# Save boilerplate with this helper decorator:

@contextual_command_completer_for("lou")
def better_unbeliever_completer(command):
    """Like unbeliever_completer but with less boilerplate"""
    if command.arg_index == 1 and 'carcolh'.startswith(command.prefix):
        return {'snail'}, len('lou ') + len(command.prefix)

To understand how xonsh uses completers and their return values try to set $XONSH_TRACE_COMPLETIONS to True:

>>> $XONSH_TRACE_COMPLETIONS = True
>>> pip c<TAB>
TRACE COMPLETIONS: Getting completions with context:
CompletionContext(command=CommandContext(args=(CommandArg(value='pip', opening_quote='', closing_quote=''),), arg_index=1, prefix='c', suffix='', opening_quote='', closing_quote='', is_after_closing_quote=False, subcmd_opening=''), python=PythonContext('pip c', 5, is_sub_expression=False))
TRACE COMPLETIONS: Got 3 results from exclusive completer 'pip':
{RichCompletion('cache', append_space=True),
 RichCompletion('check', append_space=True),
 RichCompletion('config', append_space=True)}

Registering a Completer

Once you have created a completion function, you can add it to the list of active completers via the completer add command or xonsh.completers.completer.add_one_completer function:

Usage:
    completer add NAME FUNC [POS]

NAME is a unique name to use in the listing

FUNC is the name of a completer function to use.

POS (optional) is a position into the list of completers at which the new completer should be added. It can be one of the following values:

  • "start" indicates that the completer should be added to the start of the list of completers (

    it should be run before all other exclusive completers)

  • "end" indicates that the completer should be added to the end of the list of completers (it should be run after all others)

  • ">KEY", where KEY is a pre-existing name, indicates that this should be added after the completer named KEY

  • "<KEY", where KEY is a pre-existing name, indicates that this should be added before the completer named KEY

If POS is not provided, it defaults to "start".

Note

It is also possible to manipulate __xonsh__.completers directly, but this is the preferred method.

Removing a Completer

To remove a completer from the list of active completers, run completer remove NAME, where NAME is the unique identifier associated with the completer you wish to remove.

Advanced Completions

To provide further control over the completion, a completer can return a RichCompletion object. Using this class, you can:

  • Provide a specific prefix length per completion (via prefix_len)

  • Control how the completion looks in prompt-toolkit (via display, description and style) -

    use the jedi xontrib to see it in action.

  • Append a space after the completion (append_space=True)

Completing Closed String Literals

When the cursor is appending to a closed string literal (i.e. cursor at the end of ls "/usr/"), the following happens:

  1. The closing quote will be appended to all completions.

    I.e the completion /usr/bin will turn into /usr/bin". To prevent this behavior, a completer can return a RichCompletion with append_closing_quote=False.

  2. If not specified, lprefix will cover the closing prefix.

    I.e for ls "/usr/", the default lprefix will be 6 to include the closing quote. To prevent this behavior, a completer can return a different lprefix or specify it inside RichCompletion.

So if you want to change/remove the quotes from a string, the following completer can be written:

@contextual_command_completer
def remove_quotes(command):
    """
    Return a completer that will remove the quotes, i.e:
    which "python"<TAB> -> which python
    echo "hi<TAB> -> echo hi
    ls "file with spaces"<TAB> -> ls file with spaces
    """
    raw_prefix_len = len(command.raw_prefix)  # this includes the closing quote if it exists
    return {RichCompletion(command.prefix, prefix_len=raw_prefix_len, append_closing_quote=False)}

Legacy Completers Support

Before completion context was introduced, xonsh had a different readline-like completion API. While this legacy API is not recommended, xonsh still supports it.

Warning

The legacy completers are less robust than the contextual system in many situations, for example:

  • ls $(which<TAB> completes with the prefix $(which

  • ls 'a file<TAB> completes with the prefix file (instead of a file)

See Completion Context PR for more information.

Legacy completers are python functions that aren’t marked by @contextual_completer and receive the following arguments:

  • prefix: the string to be matched (the last whitespace-separated token in the current line)

  • line: a string representing the entire current line

  • begidx: the index at which prefix starts in line

  • endidx: the length of the prefix in line

  • ctx: the current Python environment, as a dictionary mapping names to values

Their return value can be any of the variations of the contextual completers’.