=head1 NAME

Tk2portableTk - how to make your B<Tk> source portable to other
interpreted languages.

=for category  C Programming

=head1 Author

Ilya Zakharevich <ilya@math.ohio-state.edu>  has contributed most of
this document. Many thanks.

=head1 DESCRIPTION

B<PortableTk> is an attempt to make B<Tk> useful from other
languages. Currently tk4.0 runs under Perl using this
approach. Below, I<Lang> is the notation for an external language to
which B<PortableTk> glues B<Tk> code.

The main problem with using the code developed for B<TCL> with
different languages is the absence of data types: almost anything is
C<char*>. It makes automatic translation hopeless. However, if you
C<typedef> several new symbols to be C<char*>, you can still use your
code in B<TCL>, I<and> it will make the automatic translation
possible.

Another problem with the approach that "everything is a string" is
impossibility to have a result that says "NotApplicable" without
setting an error. Thus different B<Tk> command return different string
values that mean "error happened", like C<"">, C<" "> or
C<"??">. Other languages can be more flexible, so in B<portableTk> you
should inform the compiler that what you want to return means "error"
(see L<Setting variables>).

Currently B<PortableTk> uses several different approachs
to simplify translation: several B<TCL> functions that are especially
dangerous to use are undefined, so you can easily find places that
need to be updated to use Language-independent functions based on
compiler warnings.  Eventually a way to use these Language-independent
functions under proper B<TCL> will be also provided.  The end of this
document provides a starting point for such a project.

=head1 Structure of B<pTk>, porting your code

B<pTk>, that is a port of B<Tk>, is very special with respect to porting
of other code to B<portableTk>. The problem is that currently there is
very little hope to merge the modifications back into B<Tk>, so a
special strategy is needed to maintain this port. Do not use this
strategy to port your own code.

B<pTk> is produced from B<Tk> via a two-step process: first, some
manual editing (the result is in the subdirectory C<mTk>), and second,
automatic conversion by the C<munge> script (written in Perl). Thus the
subdirectory C<pTk/mTk> contains code with minimal possible difference
from the virgin B<Tk> code, so it is easier to merge(1) the
differences between B<Tk> versions into modified code.

It looks like the strategy for a portable code should be exactly
opposite: starting from B<TCL>-based code, apply C<munge>, and then
hand-edit the resulting code. Probably it is also possible to target
your code to B<portableTk> from scratch, since this will make it
possible to run it under a lot of I<Lang>uages.

The only reason anyone would like to look into contents of C<pTk/mTk>
directory is to find out which constructs are not supported by
C<munge>. On the other hand, C<pTk> directory contains code that is
conformant to B<portableTk>, so you can look there to find example code.

C<munge> is the script that converts most common B<Tk> constructs to
their C<portableTk> equivalent. For your code to qualify, you should
follow B<Tk> conventions on indentation and names of variables, in
particular, the array of arguments for the C<...CmdProc> should be
called C<argv>.

For details on what C<munge> can do, see
L<Translation of some TCL functions>.

=head1 B<PortableTk> API

=head2 Checking what you are running under

B<PortableTk> provides a symbol C<????>. If this symbol is defined,
your source is compiled with it.

=head2 New types of configuration options

B<PortableTk> defines several new types of configuration options:

 TK_CONFIG_CALLBACK
 TK_CONFIG_LANGARG
 TK_CONFIG_SCALARVAR
 TK_CONFIG_HASHVAR
 TK_CONFIG_ARRAYVAR
 TK_CONFIG_IMAGE

You should use them instead of TK_CONFIG_STRING whenever
appropriate. This allows your application to receive a direct
representation of the corresponding resource instead of the string
representation, if this is possible under given language.

???? It looks like C<TK_CONFIG_IMAGE> and C<TK_CONFIG_SCALARVAR> set
variables of type C<char*>.

=head2 Language data

The following data types are defined:

=over 4

=item C<Tcl_Obj *>

is the main datatype of the language.  This is a type that your C
function gets pointers to for arguments when the corresponding I<Lang>
function is called.  The corresponding config type is
C<TK_CONFIG_LANGARG>.

This is also a type that keeps information about contents of I<Lang>
variable.

=item C<Var>

Is a substitute for a C<char *> that contains name of variable. In
I<Lang> it is an object that contains reference to another I<Lang>
variable.

=item C<LangResultSave>

????

=item C<LangCallback>

C<LangCallback*> a substitute for a C<char *> that contains command to
call. The corresponding config type is C<TK_CONFIG_CALLBACK>.

=item C<LangFreeProc>

It is the type that the C<Lang_SplitList> sets. Before you call it,
declare

    Args *args;
    LangFreeProc *freeProc = NULL;
    ...
    code = Lang_SplitList(interp, value,
	&argc, &args, &freeProc);

After you use the split values, call

    if (args != NULL && freeProc) (*freeProc)(argc,args);

It is not guaranteed that the C<args> can survive deletion of C<value>.

=back

=head2 Conversion

The following macros and functions are used for conversion between
strings and the additional types:

 LangCallback * LangMakeCallback(Tcl_Obj *)
 Tcl_Obj * LangCallbackArg(LangCallback *)
 char * LangString(Tcl_Obj *)

After you use the result of LangCallbackArg(), you should free it with
C<freeProc> C<LANG_DYNAMIC> (it is not guaranteed that any change of
C<Tcl_Obj *> will not be reflected in <LangCallback>, so you cannot do
LangSet...() in between, and you should reset it to C<NULL> if you
want to do any further assignments to this C<Tcl_Obj *>).

The following function returns the C<Tcl_Obj *> that is a reference to C<Var>:

 Tcl_Obj * LangVarArg(Var)

???? It is very anti-intuitive, I hope the name is changed.

 int LangCmpCallback(LangCallback *a,Tcl_Obj * b)

(currently only a stub), and, at last,

 LangCallback * LangCopyCallback(LangCallback *)

=head2 Callbacks

Above we have seen the new datatype C<LangCallback> and the
corresponding I<Config option>  C<TK_CONFIG_CALLBACK>. The following
functions are provided for manipulation of C<LangCallback>s:

 void LangFreeCallback(LangCallback *)
 int LangDoCallback(Tcl_Interp *,LangCallback *,
	int result,int argc, char *format,...)

The argument C<format> of C<LangDoCallback> should contain a string that is
suitable for C<sprintf> with optional arguments of C<LangDoCallback>.
C<result> should be false if result of callback is not needed.

 int LangMethodCall(Tcl_Interp *,Tcl_Obj *,char *method,
	int result,int argc,...)

????

Conceptually, C<LangCallback*> is a substitute for ubiquitous C<char *>
in B<TCL>. So you should use C<LangFreeCallback> instead of C<ckfree>
or C<free> if appropriate.

=head2 Setting variables

 void LangFreeArg (Tcl_Obj *, Tcl_FreeProc *freeProc)
 Tcl_Obj *  LangCopyArg (Tcl_Obj *);
 void Tcl_AppendArg (Tcl_Interp *interp, Tcl_Obj *)
 void LangSetString(Tcl_Obj * *, char *s)
 void LangSetDefault(Tcl_Obj * *, char *s)

These two are equivalent unless s is an empty string. In this case
C<LangSetDefault> behaves like C<LangSetString> with C<s==NULL>, i.e.,
it sets the current value of the I<Lang> variable to be false.

 void LangSetInt(Tcl_Obj * *,int)
 void LangSetDouble(Tcl_Obj * *,double)

The I<Lang> functions separate uninitialized and initialized data
comparing data with C<NULL>. So the declaration for an C<Tcl_Obj *> should
look like

 Tcl_Obj * arg = NULL;

if you want to use this C<arg> with the above functions. After you are
done, you should use C<LangFreeArg> with C<TCL_DYNAMIC> as C<freeProc>.

=head2 Language functions

Use

=over 4

=item C<int  LangNull(Tcl_Obj *)>

to check that an object is false;

=item C<int  LangStringMatch(char *string, Tcl_Obj * match)>

????

=item C<void LangExit(int)>

to make a proper shutdown;

=item C<int LangEval(Tcl_Interp *interp, char *cmd, int global)>

to call I<Lang> C<eval>;

=item C<void Lang_SetErrorCode(Tcl_Interp *interp,char *code)>

=item C<char *Lang_GetErrorCode(Tcl_Interp *interp)>

=item C<char *Lang_GetErrorInfo(Tcl_Interp *interp)>

=item C<void LangCloseHandler(Tcl_Interp *interp,Tcl_Obj * arg,FILE *f,Lang_FileCloseProc *proc)>

currently stubs only;

=item C<int LangSaveVar(Tcl_Interp *,Tcl_Obj * arg,Var *varPtr,int type)>

to save the structure C<arg> into I<Lang> variable C<*varPtr>;

=item C<void LangFreeVar(Var var)>

to free the result;

=item C<int LangEventCallback(Tcl_Interp *,LangCallback *,XEvent *,KeySym)>

????

=item C<int LangEventHook(int flags)>

=item C<void LangBadFile(int fd)>

=item C<int LangCmpConfig(char *spec, char *arg, size_t length)>

unsupported????;

=item  C<void Tcl_AppendArg (Tcl_Interp *interp, Tcl_Obj *)>

=back

Another useful construction is

 Tcl_Obj * variable = LangFindVar(interp, Tk_Window tkwin, char *name);

After using the above function, you should call

 LangFreeVar(Var variable);

???? Note discrepancy in types!

If you want to find the value of a variable (of type C<Tcl_Obj *>) given the
variable name, use C<Tcl_GetVar(interp, varName, flags)>. If you are
interested in the string value of this variable, use
C<LangString(Tcl_GetVar(...))>.

To get a B<C> array of C<Tcl_Obj *> of length C<n>, use

    Tcl_Obj * *args = LangAllocVec(n);
    ...
    LangFreeVec(n,args);

You can set the values of the C<Tcl_Obj *>s using C<LangSet...> functions,
and get string value using C<LangString>.

If you want to merge an array of C<Tcl_Obj *>s into one C<Tcl_Obj *> (that will
be an array variable), use

    result = Tcl_Merge(listLength, list);

=head2 Translation of some TCL functions

We mark items that can be dealt with by C<munge> by I<Autoconverted>.

=over 4

=item C<Tcl_AppendResult>

does not take C<(char*)NULL>, but C<NULL> as delimiter. I<Autoconverted>.

=item C<Tcl_CreateCommand>, C<Tcl_DeleteCommand>

C<Tk_CreateWidget>, C<Tk_DeleteWidget>, the second argument is the
window itself, not the pathname. I<Autoconverted>.

=item C<sprintf(interp-E<gt>result, "%d %d %d %d",...)>

C<Tcl_IntResults(interp,4,0,...)>. I<Autoconverted>.

=item C<interp-E<gt>result = "1";>

C<Tcl_SetResult(interp,"1", TCL_STATIC)>. I<Autoconverted>.

=item Reading C<interp-E<gt>result>

C<Tcl_GetResult(interp)>. I<Autoconverted>.

=item C<interp-E<gt>result = Tk_PathName(textPtr-E<gt>tkwin);>

C<Tk_WidgetResult(interp,textPtr-E<gt>tkwin)>. I<Autoconverted>.

=item Sequence C<Tcl_PrintDouble, Tcl_PrintDouble, ..., Tcl_AppendResult>

Use a single command

 void Tcl_DoubleResults(Tcl_Interp *interp, int append,
	int argc,...);

C<append> governs whether it is required to clear the result first.

A similar command for C<int> arguments is C<Tcl_IntResults>.

=item C<Tcl_SplitList>

Use C<Lang_SplitList> (see the description above).

=back

=head1 Translation back to TCL

To use your B<portableTk> program with B<TCL>, put

 #include "ptcl.h"

I<before> inclusion of C<tk.h>, and link the resulting code with
C<ptclGlue.c>.

These files currently implement the following:

=over 4

=item Additional config types:

 TK_CONFIG_CALLBACK
 TK_CONFIG_LANGARG
 TK_CONFIG_SCALARVAR
 TK_CONFIG_HASHVAR
 TK_CONFIG_ARRAYVAR
 TK_CONFIG_IMAGE

=item Types:

 Var, Tcl_Obj *, LangCallback, LangFreeProc.

=item Functions and macros:

 Lang_SplitList, LangString, LangSetString, LangSetDefault,
 LangSetInt, LangSetDouble Tcl_ArgResult, LangCallbackArg,
 LangSaveVar, LangFreeVar,
 LangFreeSplitProc, LangFreeArg, Tcl_DoubleResults, Tcl_IntResults,
 LangDoCallback, Tk_WidgetResult, Tcl_CreateCommand,
 Tcl_DeleteCommand, Tcl_GetResult.

=back

Current implementation contains enough to make it possible to compile
C<mTk/tkText*.[ch]> with the virgin B<Tk>.

=head2 New types of events ????

PortableTk defines following new types of events:

 TK_EVENTTYPE_NONE
 TK_EVENTTYPE_STRING
 TK_EVENTTYPE_NUMBER
 TK_EVENTTYPE_WINDOW
 TK_EVENTTYPE_ATOM
 TK_EVENTTYPE_DISPLAY
 TK_EVENTTYPE_DATA

and a function

 char *	Tk_EventInfo(int letter,
	    Tk_Window tkwin, XEvent *eventPtr,
 	    KeySym keySym, int *numPtr, int *isNum, int *type,
            int num_size, char *numStorage)

=head1 Checking for trouble

If you start with working TCL code, you can start conversion using
the above hints. Good indication that you are doing is OK is absence
of C<sprintf> and C<sscanf> in your code (at least in the part that is
working with interpreter).

=head1 Additional API

What is described here is not included into base B<portableTk>
distribution. Currently it is coded in B<TCL> and as Perl macros (core
is coded as functions, so theoretically you can use the same object
files with different interpreted languages).

=head2 C<ListFactory>

Dynamic arrays in B<TCL> are used for two different purposes: to
construct strings, and to construct lists. These two usages will have
separate interfaces in other languages (since list is a different type
from a string), so you should use a different interface in your code.

The type for construction of dynamic lists is C<ListFactory>. The API
below is a counterpart of the API for construction of dynamic lists
in B<TCL>:

 void ListFactoryInit(ListFactory *)
 void ListFactoryFinish(ListFactory *)
 void ListFactoryFree(ListFactory *)
 Tcl_Obj * * ListFactoryArg(ListFactory *)
 void ListFactoryAppend(ListFactory *, Tcl_Obj * *arg)
 void ListFactoryAppendCopy(ListFactory *, Tcl_Obj * *arg)
 ListFactory * ListFactoryNewLevel(ListFactory *)
 ListFactory * ListFactoryEndLevel(ListFactory *)
 void ListFactoryResult(Tcl_Interp *, ListFactory *)

The difference is that a call to C<ListFactoryFinish> should precede the
actual usage of the value of C<ListFactory>, and there are two
different ways to append an C<Tcl_Obj *> to a C<ListFactory>:
ListFactoryAppendCopy() guarantees that the value of C<arg> is copied
to the list, but ListFactoryAppend() may append to the list a
reference to the current value of C<arg>. If you are not going to change
the value of C<arg> after appending, the call to ListFactoryAppend may
be quicker.

As in B<TCL>, the call to ListFactoryFree() does not free the
C<ListFactory>, only the objects it references.

The functions ListFactoryNewLevel() and ListFactoryEndLevel() return a
pointer to a C<ListFactory> to fill. The argument of
ListFactoryEndLevel() cannot be used after a call to this function.

=head2 DStrings

Production of strings are still supported in B<portableTk>.

=head2 Accessing C<Tcl_Obj *>s

The following functions for getting a value of an C<Tcl_Obj *> I<may> be
provided:

 double LangDouble(Tcl_Obj *)
 int LangInt(Tcl_Obj *)
 long LangLong(Tcl_Obj *)
 int LangIsList(Tcl_Obj * arg)

The function LangIsList() is supported only partially under B<TCL>,
since there is no data types. It checks whether there is a space
inside the string C<arg>.

=head2 Assigning numbers to C<Tcl_Obj *>s

While LangSetDouble() and LangSetInt() are supported ways to assign
numbers to assign an integer value to a variable, for the sake of
efficiency under B<TCL> it is supposed that the destination of these
commands was massaged before the call so it contains a long enough
string to sprintf() the numbers inside it. If you are going to
immediately use the resulting C<Tcl_Obj *>, the best way to do this is to
declare a buffer in the beginning of a block by

   dArgBuffer;

and assign this buffer to the C<Tcl_Obj *> by

   void LangSetDefaultBuffer(Tcl_Obj * *)

You can also create the buffer(s) manually and assign them using

   void LangSetBuffer(Tcl_Obj * *, char *)

This is the only choice if you need to assign numeric values to
several C<Tcl_Obj *>s simultaneously. The advantage of the first approach is
that the above declarations can be made C<nop>s in different languages.

Note that if you apply C<LangSetDefaultBuffer> to an C<Tcl_Obj *> that
contains some value, you can create a leak if you do not free that
C<Tcl_Obj *> first. This is a non-problem in real languages, but can be a
trouble in C<TCL>, unless you use only the above API.

=head2 Creating new C<Tcl_Obj *>s

The API for creating a new C<Tcl_Obj *> is

 void LangNewArg(Tcl_Obj * *, LangFreeProc *)

The API for creating a new C<Tcl_Obj *> is absent. Just initialize C<Tcl_Obj *> to
be C<NULL>, and apply one of C<LangSet...> methods.

After you use this C<Tcl_Obj *>, it should be freed thusly:

C<LangFreeArg(arg, freeProc)>.

=head2 Evaluating a list

Use

 int LangArgEval(Tcl_Interp *, Tcl_Obj * arg)

Here C<arg> should be a list to evaluate, in particular, the first
element should be a C<LangCallback> massaged to be an C<Tcl_Obj *>. The
arguments can be send to the subroutine by reference or by value in
different languages.

=head2 Getting result as C<Tcl_Obj *>

Use C<Tcl_ArgResult>. It is not guaranteed that result survives this
operation, so the C<Tcl_Obj *> you get should be the only mean to access the
data from this moment on. After you use this C<Tcl_Obj *>, you should free
it with C<freeProc> C<LANG_DYNAMIC> (you can do LangSet...() in between).

=cut