Fix typo.

[m17n/m17n-db.git] / FORMATS / IM-tut.txt

An input method is defined in a *.mim file with this format.

(input-method LANG NAME)

(description (_ "DESCRIPTION"))

(title "TITLE-STRING")

(map
  (MAP-NAME
    (KEYSEQ MAP-ACTION MAP-ACTION ...)        <- rule
    (KEYSEQ MAP-ACTION MAP-ACTION ...)        <- rule
    ...)
  (MAP-NAME
    (KEYSEQ MAP-ACTION MAP-ACTION ...)        <- rule
    (KEYSEQ MAP-ACTION MAP-ACTION ...)        <- rule
    ...)
  ...)

(state
  (STATE-NAME
    (MAP-NAME BRANCH-ACTION BRANCH-ACTION ...)   <- branch
    ...)
  (STATE-NAME
    (MAP-NAME BRANCH-ACTION BRANCH-ACTION ...)   <- branch
    ...)
  ...)

Lowercase letters and parentheses are literals, so they must be
written as they are.  Uppercase letters represent arbitrary strings.

KEYSEQ specifies a sequence of keys in this format:

  (SYMBOLIC-KEY BYMBOLIC-KEY ...)

where SYMBOLIC-KEY is the keysym 
For instance

  (n i)

represents a key sequence of <n> and <i>.  If all SYMBOLIC-KEYs are
ASCII characters, you can use the short form

  "ni"

instead.  Consult the file IM.txt for Non-ASCII characters.

MAP-ACTION and BRANCH-ACTION are a sequence of actions of this format:

  (ACTION ARG ARG ...)

The most common action is "insert" which can be written as this:

  (insert "TEXT")

But as it is very frequently used, you can use the short form

  "TEXT"

When TEXT contains only one character <C>, you can write it as

  (insert ?C)

or just

  ?C

Therefore the shortest notation for an action of inserting <a> is

  ?a

Here is a simple example:

---upcase.mim---------------------------------------------------------
(input-method en upcase)
(description (_ "Upcase all lowercase letters"))
(title "a->A")
(map
  (toupper ("a" "A") ("b" "B") ... ("z" "Z")))
(state
  (init (toupper)))
----------------------------------------------------------------------

When this input method is activated, it is in the initial condition of
the first state (in this case, it is "init").  In the initial condition,
no key is being processed and no action is suspended.  When it
receives a key event <a>, it searches branches in the current state
for a rule that matches with <a> and finds one in the map "toupper".
Then it executes MAP-ACTIONs (in this case, just inserting "A" in the
preedit buffer).  After all MAP-ACTIONs are executed, the input
method shifts to the initial condition of the current state.

The shift to "the initial condition of the first state" has a special
meaning; it commits all characters in the preedit buffer then clears
the preedit buffer.

As a result, "A" is given to an application program.

When a key event does not match with any rule in the current state,
that event is unhandled and given back to an application program.

A Turkish user may want to extend the above example for "İ" (U+0130:
LATIN CAPITAL LETTER I WITH DOT ABOVE).  It seems that assigning the
key sequence <i> <i> for that character is convenient.  So, he will
add this rule in "toupper".

    ("ii" "İ")

But we already have this rule too:

    ("i" "I")

What will happen when a key event <i> is sent to the input method?

No problem.  When the input method receives <i>, it inserts "I" in the
preedit buffer.  But, it knows that there is another rule that may
match with the additional key event <i>.  So, after inserting "I", it
suspends the normal behavior of shifting to the initial condition, and
waits for another key.  Thus, the user sees "I" with underline, which
indicates it is not yet committed.

When the input method receives the next <i>, it cancels the effects
done by the rule for the previous "i" (in this case, the preedit buffer is
cleared), and executes MAP-ACTIONs of the rule for "ii".  So, "İ" is
inserted in the preedit buffer.  This time, as there are no other rules
that match with an additional key, it shifts to the initial condition
of the current state, and commits "İ".

Then, what will happen when the next key event is <a> instead of <i>?

No problem, either.

The input method knows that there are no rules that match the <i> <a> key
sequence.  So, when it receives the next <a>, it executes the
suspended behavior (i.e. shifting to the initial condition), which
leads to commit "S".  Then the input method tries to handle <a> in the current
state, which leads to commit "A".

So far, we have explained MAP-ACTION, but not
BRANCH-ACTION.  The format of BRANCH-ACTION is the same as that of MAP-ACTION.
It is executed only after a matching rule was determined and the
corresponding MAP-ACTIONs were executed.  A typical use of
BRANCH-ACTION is to shift to a different state.

To see this effect, let us modify the current input method to upcase a
letter only at a beginning of a word (i.e. capitalizing).  For that purpose,
we modify the "init" state as this:

  (init
    (toupper (shift non-upcase)))

Here "(shift non-upcase)" is an action to shift to the new state
"non-upcase", which has two branches as below:

  (non-upcase
    (lower)
    (nil (shift init)))

The first branch is simple.  We can define the new map "lower" as the
following to insert lowercase letters as they are.

(map
  ...
  (lower ("a" "a") ("b" "b") ... ("z" "z")))

The second branch has a special meaning.  The map name "nil" means
that it matches with any key event that does not match any rules in the
other maps in the same state.  In addition, it does not
consume any key event.  We will show the full code of the new input
method before explaining how it works.

---titlecase.mim------------------------------------------------------
(input-method en titlecase)
(description (_ "Titlecase letters"))
(title "abc->Abc")
(map
  (toupper ("a" "A") ("b" "B") ... ("z" "Z") ("ii" "İ"))
  (lower ("a" "a") ("b" "b") ... ("z" "z")))
(state
  (init
    (toupper (shift non-upcase)))
  (non-upcase
    (lower (commit))
    (nil (shift init))))
----------------------------------------------------------------------

Let's see what happens when the user types the keys <a> <b> < >.

Upon <a>, "A" is committed and the state shifts to "non-upcase".
So, the next <b> is handled in the "non-upcase" state.  As it matches with a
rule in the map "lower", "b" is inserted in the preedit buffer and it
is committed explicitly by the "commit" command in BRANCH-ACTION.  After
that, the input method is still in "non-upcase" state.  So the next < >
is also handled in "non-upcase".  This time, as no rule in this state
matches with it, the branch "(nil (shift init))" is selected, and the
state is changed to init.  Please note that < > is not yet handled.
So, the input method tries to handle it in the "init" state.  Again no
rule matches with it.  Therefore, that event is given back to an application
program, which usually inserts a space for that.

When you type "a quick blown fox" with this input method, you get "A
Quick Blown Fox".  OK, you find a typo in "blown", it should be
"brown".  To correct it, you move the cursor after "l" and type
Backspace and <r>.  However, if the current input method is still
active, a capital "R" is inserted.  It is not a sophisticated
behavior.

To make the input method work well also in such a case, we must use
"surrounding text support".  It is a way to check characters around
the inputting spot and delete them if necessary.  Please note that
this facility is available only with Gtk+ applications and Qt
applications.  You cannot use it with applications that use XIM
to communicate with an input method.

Before explaining how to utilize "surrounding text support", you must
understand how to use variables, arithmetic comparison, and
conditional actions.

At first, any symbol (except for several preserved ones) used as ARG
of an action is treated as a variable.  For instance, the commands

  (set X 32) (insert X)

set the variable "X" to integer value 32, then insert a character
whose Unicode character code is 32 (i.e. SPACE).

The second argument of the "set" action can be an expression of this form:

  (OPERAND ARG1 [ARG2])

Both ARG1 and ARG2 can be an expression.  So,

  (set X (+ (* Y 32) Z)) --- set "X" to the value of "Y*32+Z"

We have the following arithmetic/bitwise OPERANDs (require two arguments):

  + - * / & |

these relational OPERANDs (require two arguments):

  == <= >= < >

and this logical OPERAND (require one argument):

  !

For surrounding text support, we have these preserved variables:

  @-0, @-N, @+N (N is a natural number)

The values of them are predefined as below and can not be altered.

  @-0  -- -1 if surrounding text is supported, -2 if not.
  @-N  -- The Nth previous character in the preedit buffer.  If there is only
          M (M<N) previous characters in it, the value is the (N-M)th previous
          character from the inputting spot.
  @+N  -- The Nth following character in the preedit buffer.  If there is only
          M (M<N) following characters in it, the value is the (N-M)th following
          character from the inputting spot.

So, provided that you have this context:

  ABC|def|GHI

("def" is in the preedit buffer, two "|"s indicate borders between the
preedit buffer and the surrounding text) and your current position in
the preedit buffer is between "d" and "e", we have these values:

  @-3 -- ?B
  @-2 -- ?C
  @-1 -- ?d
  @+1 -- ?e
  @+2 -- ?f
  @+3 -- ?G

Next, you have to understand a conditional action of this form:

  (cond
    (EXPR1 ACTION ACTION ...)
    (EXPR2 ACTION ACTION ...)
    ...)

where EXPRn are expressions.  When an input method executes this
action, it resolves the values of EXPRn one by one from the first branch.
If the value of EXPRn is resolved into nonzero, the corresponding
actions are executed.

Now you are ready to write a new version of the input method "Titlecase".

---titlecase2.mim------------------------------------------------------
(input-method en titlecase)
(description (_ "Titlecase letters"))
(title "abc->Abc")
(map
  (toupper ("a" "A") ("b" "B") ("c" "C") ("d" "D") ("e" "E")
           ("f" "F") ("g" "G") ("h" "H") ("i" "I") ("j" "J")
           ("k" "K") ("l" "L") ("m" "M") ("n" "N") ("o" "O")
           ("p" "P") ("q" "Q") ("r" "R") ("s" "S") ("t" "T")
           ("u" "U") ("v" "V") ("w" "W") ("x" "X") ("y" "Y")
           ("z" "Z") ("ii" "İ")))
(state
  (init
    (toupper

     ;; Now we have one character in the preedit buffer.  So, "@-2" is
     ;; the character just before the inputting spot.

     (cond ((| (& (>= @-2 ?A) (<= @-2 ?Z))
               (& (>= @-2 ?a) (<= @-2 ?z))
               (= @-2 ?İ))

            ;; If that character is A..Z, a..z, or İ, remember the
            ;; character in the preedit in X and delete it.

            (set X @-1) (delete @-)

            ;; Then insert a proper lower case version of X.

            (cond ((= X ?İ) "i") 
                  (1 (set X (+ X 32)) (insert X))))))))
----------------------------------------------------------------------

The above example contains a new action "delete".  So, it is time to
explain more about the preedit buffer.  The preedit buffer is a
temporary place to store a sequence of characters.  In this buffer, the
input method keeps one position called the "current position".  The current
position exists between two characters, at the beginning of the buffer, or at
the end of the buffer.  The "insert" action inserts characters before
the current position.  For instance, when your preedit
buffer contains "ab.c" ("." indicates the current position),

  (insert "xyz")

changes the buffer to "abxyz.c".

There are several predefined variables that represent a specific position in the
preedit buffer.  They are:

  @<, @=, @> -- the first, current, and last positions
  @-, @+     -- the previous and the next positions

The format of the "delete" action is:

  (delete POS)

where POS is a positional variable beginning with @.
The above action deletes the characters between POS and
the current position.  So, "(delete @-)" deletes one character before
the current position.  The other examples of "delete" actions are:

  (delete @+)  -- delete the next character
  (delete @<)  -- delete all the preceding characters in the buffer
  (delete @>)  -- delete all the following characters in the buffer

You can change the current position by the "move" action as below:

  (move @-)  -- move the current position to the position before the
                previous character
  (move @<)  -- move to the first position

Other positional variables work similarly.
Let's see how our new example works.  Whatever a key event is, the
input method is in its only state, "init".  Since an event of a lower letter
key is first handled by MAP-ACTIONs, every key is changed into the
corresponding uppercase and put into the preedit buffer.  Now this character
can be retrieved with @-1.

How can we determine whether the new character should be
lowercase or not?  We have to check the character before it, that is, @-2.
BRANCH-ACTIONs in the "init" state do the job.

It first checks if the character @-2 is between A and Z, or between a
and z, by the conditional below.

     (cond ((| (& (>= @-2 ?A) (<= @-2 ?Z))
               (& (>= @-2 ?a) (<= @-2 ?z))
               (= @-2 ?İ))

If not, there is nothing to do specially.  If so, our new key
should be changed back into lowercase.  The uppercase character is already
in the preedit buffer and we have to retrieve and remember it in the
variable X by

    (set X @-1)

and then delete that character by

    (delete @-)

Lastly we insert the character back in its lowercase form.  The
problem here is that "İ" must be changed into "i", so we need another
conditional.  The first branch

    ((= X ?İ) "i")

means that "if the character remembered in X is İ, "i" is inserted".

The second branch 

    (1 (set X (+ X 32)) (insert X))

starts with "1", which is always resolved into nonzero, so this branch
is catchall.  Actions in this branch add 32 to X itself and then insert
X.  In other words, they make A...Z into a...z respectively and insert the
lowercase character into the preedit buffer.  As the input method
reached the end of the BRANCH-ACTIONs, the character is commited.

This new input method always checks the character before the current
position, so "A Quick Blown Fox" will be successfully fixed to "A
Quick Brown Fox" by a backspace and <r>.