1 Programming with GNU Readline

This manual describes the interface between the GNU Readline Library and user programs. If you are a programmer, and you wish to include the features found in GNU Readline in your own programs, such as completion, line editing, and interactive history manipulation, this documentation is for you.

1.1 Default Behaviour

Many programs provide a command line interface, such as mail, ftp, and sh. For such programs, the default behaviour of Readline is sufficient. This section describes how to use Readline in the simplest way possible, perhaps to replace calls in your code to gets ().

The function readline prints a prompt and then reads and returns a single line of text from the user. The line which readline () returns is allocated with malloc (); you should free () the line when you are done with it. The declaration for readline in ANSI C is

char *readline (char *prompt);

So, one might say

char *line = readline ("Enter a line: ");

in order to read a line of text from the user.

The line which is returned has the final newline removed, so only the text of the line remains.

If readline encounters an EOF while reading the line, and the line is empty at that point, then (char *)NULL is returned. Otherwise, the line is ended just as if a newline was typed.

If you want the user to be able to get at the line later, (with <C-p> for example), you must call add_history () to save the line away in a history list of such lines.

add_history (line);

For full details on the GNU History Library, see the associated manual.

It is polite to avoid saving empty lines on the history list, since it is rare than someone has a burning need to reuse a blank line. Here is a function which usefully replaces the standard gets () library function:

/* A static variable for holding the line. */
static char *line_read = (char *)NULL;

/* Read a string, and return a pointer to it.  Returns NULL on EOF. */
char *
do_gets ()
{
  /* If the buffer has already been allocated, return the memory
     to the free pool. */
  if (line_read != (char *)NULL)
    {
      free (line_read);
      line_read = (char *)NULL;
    }

  /* Get a line from the user. */
  line_read = readline ("");

  /* If the line has any text in it, save it on the history. */
  if (line_read && *line_read)
    add_history (line_read);

  return (line_read);
}

The above code gives the user the default behaviour of <TAB> completion: completion on file names. If you do not want readline to complete on filenames, you can change the binding of the <TAB> key with rl_bind_key ().

int rl_bind_key (int key, int (*function)());

rl_bind_key () takes 2 arguments; key is the character that you want to bind, and function is the address of the function to run when key is pressed. Binding <TAB> to rl_insert () makes <TAB> just insert itself.

rl_bind_key () returns non-zero if key is not a valid ASCII character code (between 0 and 255).

rl_bind_key ('\t', rl_insert);

This code should be executed once at the start of your program; you might write a function called initialize_readline () which performs this and other desired initializations, such as installing custom completers, etc.

1.2 Custom Functions

Readline provides a great many functions for manipulating the text of the line. But it isn’t possible to anticipate the needs of all programs. This section describes the various functions and variables defined in within the Readline library which allow a user program to add customized functionality to Readline.

1.2.1 The Function Type

For the sake of readabilty, we declare a new type of object, called Function. A Function is a C language function which returns an int. The type declaration for Function is:

typedef int Function ();

The reason for declaring this new type is to make it easier to write code describing pointers to C functions. Let us say we had a variable called func which was a pointer to a function. Instead of the classic C declaration

int (*)()func;

we have

Function *func;

1.2.2 Naming a Function

The user can dynamically change the bindings of keys while using Readline. This is done by representing the function with a descriptive name. The user is able to type the descriptive name when referring to the function. Thus, in an init file, one might find

Meta-Rubout:	backward-kill-word

This binds the keystroke <Meta-Rubout> to the function descriptively named backward-kill-word. You, as the programmer, should bind the functions you write to descriptive names as well. Readline provides a function for doing that:

Function: rl_add_defun (char *name, Function *function, int key)

Add name to the list of named functions. Make function be the function that gets called. If key is not -1, then bind it to function using rl_bind_key ().

Using this function alone is sufficient for most applications. It is the recommended way to add a few functions to the default functions that Readline has built in already. If you need to do more or different things than adding a function to Readline, you may need to use the underlying functions described below.

1.2.3 Selecting a Keymap

Key bindings take place on a keymap. The keymap is the association between the keys that the user types and the functions that get run. You can make your own keymaps, copy existing keymaps, and tell Readline which keymap to use.

Function: Keymap rl_make_bare_keymap ()

Returns a new, empty keymap. The space for the keymap is allocated with malloc (); you should free () it when you are done.

Function: Keymap rl_copy_keymap (Keymap map)

Return a new keymap which is a copy of map.

Function: Keymap rl_make_keymap ()

Return a new keymap with the printing characters bound to rl_insert, the lowercase Meta characters bound to run their equivalents, and the Meta digits bound to produce numeric arguments.

1.2.4 Binding Keys

You associate keys with functions through the keymap. Here are functions for doing that.

Function: int rl_bind_key (int key, Function *function)

Binds key to function in the currently selected keymap. Returns non-zero in the case of an invalid key.

Function: int rl_bind_key_in_map (int key, Function *function, Keymap map)

Bind key to function in map. Returns non-zero in the case of an invalid key.

Function: int rl_unbind_key (int key)

Make key do nothing in the currently selected keymap. Returns non-zero in case of error.

Function: int rl_unbind_key_in_map (int key, Keymap map)

Make key be bound to the null function in map. Returns non-zero in case of error.

Function: rl_generic_bind (int type, char *keyseq, char *data, Keymap map)

Bind the key sequence represented by the string keyseq to the arbitrary pointer data. type says what kind of data is pointed to by data; right now this can be a function (ISFUNC), a macro (ISMACR), or a keymap (ISKMAP). This makes new keymaps as necessary. The initial place to do bindings is in map.

1.2.5 Writing a New Function

In order to write new functions for Readline, you need to know the calling conventions for keyboard invoked functions, and the names of the variables that describe the current state of the line gathered so far.

Variable: char *rl_line_buffer

This is the line gathered so far. You are welcome to modify the contents of this, but see Undoing, below.

Variable: int rl_point

The offset of the current cursor position in rl_line_buffer.

Variable: int rl_end

The number of characters present in rl_line_buffer. When rl_point is at the end of the line, then rl_point and rl_end are equal.

The calling sequence for a command foo looks like

foo (int count, int key)

where count is the numeric argument (or 1 if defaulted) and key is the key that invoked this function.

It is completely up to the function as to what should be done with the numeric argument; some functions use it as a repeat count, other functions as a flag, and some choose to ignore it. In general, if a function uses the numeric argument as a repeat count, it should be able to do something useful with a negative argument as well as a positive argument. At the very least, it should be aware that it can be passed a negative argument.

1.2.6 Allowing Undoing

Supporting the undo command is a painless thing to do, and makes your functions much more useful to the end user. It is certainly easy to try something if you know you can undo it. I could use an undo function for the stock market.

If your function simply inserts text once, or deletes text once, and it calls rl_insert_text () or rl_delete_text () to do it, then undoing is already done for you automatically, and you can safely skip this section.

If you do multiple insertions or multiple deletions, or any combination of these operations, you should group them together into one operation. This can be done with rl_begin_undo_group () and rl_end_undo_group ().

Function: rl_begin_undo_group ()

Begins saving undo information in a group construct. The undo information usually comes from calls to rl_insert_text () and rl_delete_text (), but they could be direct calls to rl_add_undo ().

Function: rl_end_undo_group ()

Closes the current undo group started with rl_begin_undo_group (). There should be exactly one call to rl_end_undo_group () for every call to rl_begin_undo_group ().

Finally, if you neither insert nor delete text, but directly modify the existing text (e.g. change its case), you call rl_modifying () once, just before you modify the text. You must supply the indices of the text range that you are going to modify.

Function: rl_modifying (int start, int end)

Tell Readline to save the text between start and end as a single undo unit. It is assumed that subsequent to this call you will modify that range of text in some way.

1.2.7 An Example

Here is a function which changes lowercase characters to the uppercase equivalents, and uppercase characters to the lowercase equivalents. If this function was bound to ‘M-c’, then typing ‘M-c’ would change the case of the character under point. Typing ‘10 M-c’ would change the case of the following 10 characters, leaving the cursor on the last character changed.

/* Invert the case of the COUNT following characters. */
invert_case_line (count, key)
     int count, key;
{
  register int start, end;

  start = rl_point;

  if (count < 0)
    {
      direction = -1;
      count = -count;
    }
  else
    direction = 1;
      
  /* Find the end of the range to modify. */
  end = start + (count * direction);

  /* Force it to be within range. */
  if (end > rl_end)
    end = rl_end;
  else if (end < 0)
    end = -1;

  if (start > end)
    {
      int temp = start;
      start = end;
      end = temp;
    }

  if (start == end)
    return;

  /* Tell readline that we are modifying the line, so save the undo
     information. */
  rl_modifying (start, end);

  for (; start != end; start += direction)
    {
      if (uppercase_p (rl_line_buffer[start]))
        rl_line_buffer[start] = to_lower (rl_line_buffer[start]);
      else if (lowercase_p (rl_line_buffer[start]))
        rl_line_buffer[start] = to_upper (rl_line_buffer[start]);
    }
  /* Move point to on top of the last character changed. */
  rl_point = end - direction;
}

1.3 Custom Completers

Typically, a program that reads commands from the user has a way of disambiguating commands and data. If your program is one of these, then it can provide completion for either commands, or data, or both commands and data. The following sections describe how your program and Readline cooperate to provide this service to end users.

1.3.1 How Completing Works

In order to complete some text, the full list of possible completions must be available. That is to say, it is not possible to accurately expand a partial word without knowing what all of the possible words that make sense in that context are. The GNU Readline library provides the user interface to completion, and additionally, two of the most common completion functions; filename and username. For completing other types of text, you must write your own completion function. This section describes exactly what those functions must do, and provides an example function.

There are three major functions used to perform completion:

1. The user-interface function rl_complete (). This function is called interactively with the same calling conventions as other functions in readline intended for interactive use; i.e. count, and invoking-key. It isolates the word to be completed and calls completion_matches () to generate a list of possible completions. It then either lists the possible completions or actually performs the completion, depending on which behaviour is desired.
2. The internal function completion_matches () uses your generator function to generate the list of possible matches, and then returns the array of these matches. You should place the address of your generator function in rl_completion_entry_function.
3. The generator function is called repeatedly from completion_matches (), returning a string each time. The arguments to the generator function are text and state. text is the partial word to be completed. state is zero the first time the function is called, and a positive non-zero integer for each subsequent call. When the generator function returns (char *)NULL this signals completion_matches () that there are no more possibilities left.

Function: rl_complete (int ignore, int invoking_key)

Complete the word at or before point. You have supplied the function that does the initial simple matching selection algorithm (see completion_matches ()). The default is to do filename completion.

Note that rl_complete () has the identical calling conventions as any other key-invokable function; this is because by default it is bound to the ‘TAB’ key.

Variable: Function *rl_completion_entry_function

This is a pointer to the generator function for completion_matches (). If the value of rl_completion_entry_function is (Function *)NULL then the default filename generator function is used, namely filename_entry_function ().

1.3.2 Completion Functions

Here is the complete list of callable completion functions present in Readline.

Function: rl_complete_internal (int what_to_do)

Complete the word at or before point. what_to_do says what to do with the completion. A value of ‘?’ means list the possible completions. ‘TAB’ means do standard completion. ‘*’ means insert all of the possible completions.

Function: rl_complete (int ignore, int invoking_key)

Complete the word at or before point. You have supplied the function that does the initial simple matching selection algorithm (see completion_matches ()). The default is to do filename completion. This just calls rl_complete_internal () with an argument of ‘TAB’.

Function: rl_possible_completions ()

List the possible completions. See description of rl_complete (). This just calls rl_complete_internal () with an argument of ‘?’.

Function: char **completion_matches (char *text, char *(*entry_function) ())

Returns an array of (char *) which is a list of completions for text. If there are no completions, returns (char **)NULL. The first entry in the returned array is the substitution for text. The remaining entries are the possible completions. The array is terminated with a NULL pointer.

entry_function is a function of two args, and returns a (char *). The first argument is text. The second is a state argument; it is zero on the first call, and non-zero on subsequent calls. It returns a NULL pointer to the caller when there are no more matches.

Function: char *filename_completion_function (char *text, int state)

A generator function for filename completion in the general case. Note that completion in the Bash shell is a little different because of all the pathnames that must be followed when looking up the completion for a command.

Function: char *username_completion_function (char *text, int state)

A completion generator for usernames. text contains a partial username preceded by a random character (usually ‘~’).

1.3.3 Completion Variables

Variable: Function *rl_completion_entry_function

A pointer to the generator function for completion_matches (). NULL means to use filename_entry_function (), the default filename completer.

Variable: Function *rl_attempted_completion_function

A pointer to an alternative function to create matches. The function is called with text, start, and end. start and end are indices in rl_line_buffer saying what the boundaries of text are. If this function exists and returns NULL then rl_complete () will call the value of rl_completion_entry_function to generate matches, otherwise the array of strings returned will be used.

Variable: int rl_completion_query_items

Up to this many items will be displayed in response to a possible-completions call. After that, we ask the user if she is sure she wants to see them all. The default value is 100.

Variable: char *rl_basic_word_break_characters

The basic list of characters that signal a break between words for the completer routine. The contents of this variable is what breaks words in the Bash shell, i.e. " \t\n\"\\’‘@$><=;|&{(".

Variable: char *rl_completer_word_break_characters

The list of characters that signal a break between words for rl_complete_internal (). The default list is the contents of rl_basic_word_break_characters.

Variable: char *rl_special_prefixes

The list of characters that are word break characters, but should be left in text when it is passed to the completion function. Programs can use this to help determine what kind of completing to do.

Variable: int rl_ignore_completion_duplicates

If non-zero, then disallow duplicates in the matches. Default is 1.

Variable: int rl_filename_completion_desired

Non-zero means that the results of the matches are to be treated as filenames. This is always zero on entry, and can only be changed within a completion entry generator function.

Variable: Function *rl_ignore_some_completions_function

This function, if defined, is called by the completer when real filename completion is done, after all the matching names have been generated. It is passed a NULL terminated array of (char *) known as matches in the code. The 1st element (matches[0]) is the maximal substring that is common to all matches. This function can re-arrange the list of matches as required, but each deleted element of the array must be free()’d.

1.3.4 A Short Completion Example

Here is a small application demonstrating the use of the GNU Readline library. It is called fileman, and the source code resides in ‘readline/examples/fileman.c’. This sample application provides completion of command names, line editing features, and access to the history list.

/* fileman.c -- A tiny application which demonstrates how to use the
   GNU Readline library.  This application interactively allows users
   to manipulate files and their modes. */

#include <stdio.h>
#include <readline/readline.h>
#include <readline/history.h>
#include <sys/types.h>
#include <sys/file.h>
#include <sys/stat.h>
#include <sys/errno.h>

/* The names of functions that actually do the manipulation. */
int com_list (), com_view (), com_rename (), com_stat (), com_pwd ();
int com_delete (), com_help (), com_cd (), com_quit ();

/* A structure which contains information on the commands this program
   can understand. */

typedef struct {
  char *name;                   /* User printable name of the function. */
  Function *func;               /* Function to call to do the job. */
  char *doc;                    /* Documentation for this function.  */
} COMMAND;

COMMAND commands[] = {
  { "cd", com_cd, "Change to directory DIR" },
  { "delete", com_delete, "Delete FILE" },
  { "help", com_help, "Display this text" },
  { "?", com_help, "Synonym for `help'" },
  { "list", com_list, "List files in DIR" },
  { "ls", com_list, "Synonym for `list'" },
  { "pwd", com_pwd, "Print the current working directory" },
  { "quit", com_quit, "Quit using Fileman" },
  { "rename", com_rename, "Rename FILE to NEWNAME" },
  { "stat", com_stat, "Print out statistics on FILE" },
  { "view", com_view, "View the contents of FILE" },
  { (char *)NULL, (Function *)NULL, (char *)NULL }
};

/* The name of this program, as taken from argv[0]. */
char *progname;

/* When non-zero, this global means the user is done using this program. */
int done = 0;

main (argc, argv)
     int argc;
     char **argv;
{
  progname = argv[0];

  initialize_readline ();       /* Bind our completer. */

  /* Loop reading and executing lines until the user quits. */
  while (!done)
    {
      char *line;

      line = readline ("FileMan: ");

      if (!line)
        {
          done = 1;             /* Encountered EOF at top level. */
        }
      else
        {
          /* Remove leading and trailing whitespace from the line.
             Then, if there is anything left, add it to the history list
             and execute it. */
          stripwhite (line);

          if (*line)
            {
              add_history (line);
              execute_line (line);
            }
        }

      if (line)
        free (line);
    }
  exit (0);
}

/* Execute a command line. */
execute_line (line)
     char *line;
{
  register int i;
  COMMAND *find_command (), *command;
  char *word;

  /* Isolate the command word. */
  i = 0;
  while (line[i] && !whitespace (line[i]))
    i++;

  word = line;

  if (line[i])
    line[i++] = '\0';

  command = find_command (word);

  if (!command)
    {
      fprintf (stderr, "%s: No such command for FileMan.\n", word);
      return;
    }

  /* Get argument to command, if any. */
  while (whitespace (line[i]))
    i++;

  word = line + i;

  /* Call the function. */
  (*(command->func)) (word);
}

/* Look up NAME as the name of a command, and return a pointer to that
   command.  Return a NULL pointer if NAME isn't a command name. */
COMMAND *
find_command (name)
     char *name;
{
  register int i;

  for (i = 0; commands[i].name; i++)
    if (strcmp (name, commands[i].name) == 0)
      return (&commands[i]);

  return ((COMMAND *)NULL);
}

/* Strip whitespace from the start and end of STRING. */
stripwhite (string)
     char *string;
{
  register int i = 0;

  while (whitespace (string[i]))
    i++;

  if (i)
    strcpy (string, string + i);

  i = strlen (string) - 1;

  while (i > 0 && whitespace (string[i]))
    i--;

  string[++i] = '\0';
}

/* **************************************************************** */
/*                                                                  */
/*                  Interface to Readline Completion                */
/*                                                                  */
/* **************************************************************** */

/* Tell the GNU Readline library how to complete.  We want to try to complete
   on command names if this is the first word in the line, or on filenames
   if not. */
initialize_readline ()
{
  char **fileman_completion ();

  /* Allow conditional parsing of the ~/.inputrc file. */
  rl_readline_name = "FileMan";

  /* Tell the completer that we want a crack first. */
  rl_attempted_completion_function = (Function *)fileman_completion;
}

/* Attempt to complete on the contents of TEXT.  START and END show the
   region of TEXT that contains the word to complete.  We can use the
   entire line in case we want to do some simple parsing.  Return the
   array of matches, or NULL if there aren't any. */
char **
fileman_completion (text, start, end)
     char *text;
     int start, end;
{
  char **matches;
  char *command_generator ();

  matches = (char **)NULL;

  /* If this word is at the start of the line, then it is a command
     to complete.  Otherwise it is the name of a file in the current
     directory. */
  if (start == 0)
    matches = completion_matches (text, command_generator);

  return (matches);
}

/* Generator function for command completion.  STATE lets us know whether
   to start from scratch; without any state (i.e. STATE == 0), then we
   start at the top of the list. */
char *
command_generator (text, state)
     char *text;
     int state;
{
  static int list_index, len;
  char *name;

  /* If this is a new word to complete, initialize now.  This includes
     saving the length of TEXT for efficiency, and initializing the index
     variable to 0. */
  if (!state)
    {
      list_index = 0;
      len = strlen (text);
    }

  /* Return the next name which partially matches from the command list. */
  while (name = commands[list_index].name)
    {
      list_index++;

      if (strncmp (name, text, len) == 0)
        return (name);
    }

  /* If no names matched, then return NULL. */
  return ((char *)NULL);
}

/* **************************************************************** */
/*                                                                  */
/*                       FileMan Commands                           */
/*                                                                  */
/* **************************************************************** */

/* String to pass to system ().  This is for the LIST, VIEW and RENAME
   commands. */
static char syscom[1024];

/* List the file(s) named in arg. */
com_list (arg)
     char *arg;
{
  if (!arg)
    arg = "*";

  sprintf (syscom, "ls -FClg %s", arg);
  system (syscom);
}

com_view (arg)
     char *arg;
{
  if (!valid_argument ("view", arg))
    return;

  sprintf (syscom, "cat %s | more", arg);
  system (syscom);
}

com_rename (arg)
     char *arg;
{
  too_dangerous ("rename");
}

com_stat (arg)
     char *arg;
{
  struct stat finfo;

  if (!valid_argument ("stat", arg))
    return;

  if (stat (arg, &finfo) == -1)
    {
      perror (arg);
      return;
    }

  printf ("Statistics for `%s':\n", arg);

  printf ("%s has %d link%s, and is %d bytes in length.\n", arg,
          finfo.st_nlink, (finfo.st_nlink == 1) ? "" : "s",  finfo.st_size);
  printf ("      Created on: %s", ctime (&finfo.st_ctime));
  printf ("  Last access at: %s", ctime (&finfo.st_atime));
  printf ("Last modified at: %s", ctime (&finfo.st_mtime));
}

com_delete (arg)
     char *arg;
{
  too_dangerous ("delete");
}

/* Print out help for ARG, or for all of the commands if ARG is
   not present. */
com_help (arg)
     char *arg;
{
  register int i;
  int printed = 0;

  for (i = 0; commands[i].name; i++)
    {
      if (!*arg || (strcmp (arg, commands[i].name) == 0))
        {
          printf ("%s\t\t%s.\n", commands[i].name, commands[i].doc);
          printed++;
        }
    }

  if (!printed)
    {
      printf ("No commands match `%s'.  Possibilties are:\n", arg);

      for (i = 0; commands[i].name; i++)
        {
          /* Print in six columns. */
          if (printed == 6)
            {
              printed = 0;
              printf ("\n");
            }

          printf ("%s\t", commands[i].name);
          printed++;
        }

      if (printed)
        printf ("\n");
    }
}

/* Change to the directory ARG. */
com_cd (arg)
     char *arg;
{
  if (chdir (arg) == -1)
    perror (arg);

  com_pwd ("");
}

/* Print out the current working directory. */
com_pwd (ignore)
     char *ignore;
{
  char dir[1024];

  (void) getwd (dir);

  printf ("Current directory is %s\n", dir);
}

/* The user wishes to quit using this program.  Just set DONE non-zero. */
com_quit (arg)
     char *arg;
{
  done = 1;
}

/* Function which tells you that you can't do this. */
too_dangerous (caller)
     char *caller;
{
  fprintf (stderr,
           "%s: Too dangerous for me to distribute.  Write it yourself.\n",
           caller);
}

/* Return non-zero if ARG is a valid argument for CALLER, else print
   an error message and return zero. */
int
valid_argument (caller, arg)
     char *caller, *arg;
{
  if (!arg || !*arg)
    {
      fprintf (stderr, "%s: Argument required.\n", caller);
      return (0);
    }

  return (1);
}

About This Document

This document was generated on January 29, 2023 using texi2html 5.0.

The buttons in the navigation panels have the following meaning:

where the Example assumes that the current position is at Subsubsection One-Two-Three of a document of the following structure:

• 1. Section One
  • 1.1 Subsection One-One
    • ...
  • 1.2 Subsection One-Two
    • 1.2.1 Subsubsection One-Two-One
    • 1.2.2 Subsubsection One-Two-Two
    • 1.2.3 Subsubsection One-Two-Three     <== Current Position
    • 1.2.4 Subsubsection One-Two-Four
  • 1.3 Subsection One-Three
    • ...
  • 1.4 Subsection One-Four

This document was generated on January 29, 2023 using texi2html 5.0.