Next Previous Contents

8. Programming and design issues

8.1 What are LOCAL variables and why do I care?

LOCAL variables are somewhat like PRIVATE variables in that they do not "scope up"; if you declare a PRIVATE or LOCAL variable in function B which is called from function A, then function A does not know the variable exists. Function A is even free to use the same name for a different variable entirely without risk of confusion. This is useful for writing functions that can avoid "stepping on" the variables that are floating around a larger application.

LOCAL variables, introduced with Clipper 5.0, enforce a stricter kind of modularity. A LOCAL variable must be declared, and it is visible only to the particular function in which it is declared. (Remember that PRIVATEs are created by default.) Not only does a LOCAL variable refrain from "up-scoping" but it also will not "down-scope" either.

For example:

a := "a"
x := "ABC"
y := "123"
? a, x, y        // Prints "a       ABC     123"
? a, x, y        // Prints "a       ABC     456"
PRIVATE x        // Notice y not declared PRIVATE!
x := "DEF"
y := "456"
? a, x, y        // Prints "NIL     DEF     456"

The values of (a) in the two functions are completely independent of each other, because they are indeed two separate variables. The values of (x) in the two functions are also separate and independent, because CALLEE() has declared its version of (x) PRIVATE. There is only one (y) variable, however, since CALLEE() has neglected to declare it.

LOCAL variables are useful for writing code that is guaranteed bulletproof regardless of the application environment. LOCAL variables (and their cousins, the STATICs) are immune from having their values changed by unrelated underlying functions. This cuts down on maintenance and debugging while offering some speed advantages too.

Because all uses of a local variable are known to the compiler (being all in one function) there is no need for an entry in the run-time symbol table, which does save some time and memory during program execution. A side effect is that the names of LOCAL variables in a macro expression are not found; therefore

LOCAL x, y
x := "y"
? &x

will fail unless there is a PRIVATE or PUBLIC variable named (y).

LOCALs work just like "auto" or stack variables in C and C++.

8.2 What are STATIC variables and why do I care?

STATIC variables are similar to LOCALs in that they do not "up-scope" or "down-scope." STATICs, however, keep their value throughout the duration of the program, even when they are not "visible." For example:

FOR i := 1 TO 100
        ? "Loop number ", i
? "END"
LOCAL l := 0
STATIC s := 0
? l, s

The output will look like this:

    Loop number     1
        1      1
    Loop number     2
        1      2
    Loop number     3
        1      3
    Loop number     4
        1      4
        ...    ...
    Loop number   100
        1    100

Notice that each iteration of the loop increments the value of (s) from the previous iteration, while the value of (l) starts all over again from scratch.

The most common use of STATIC variables is to maintain values that are useful to a group of functions, or to the whole application, without resorting to global PRIVATE or PUBLIC variables. (See next Question.)

STATIC variables carry the same features as LOCALs with regard to symbol table entries and macro expansion as noted above.

STATICs work just like static variables in C and C++.

Someone (whose identity I've unfortunately misplaced) wrote to add: "It is dangerous to use large numbers of STATIC variables in an application, since they must all fit into one 64K data segment. The FAQ proposes creating an array of values rather than a plethora of STATICs, but I wonder if this problem should be mentioned more prominently." So there it is.

8.3 How do I switch all my PUBLICs to STATICs?

Pull these variables in logical groups (e.g., all colors in one place) into separate source files and declare them STATIC. Then write a function or set of functions which allow you to get (and optionally set) the values of these variables. For example, COLORS1.PRG will return the current value of the cNormal variable, and will also set the variable if you pass it a value.

STATIC cNormal
FUNCTION setnorm (x)
LOCAL cRet := cNormal
IF x != NIL
 cNormal := x

If you have more than a few of these variables to get and set this may present a problem; the source and object code for these functions will quickly grow out of hand! So let's take it one step further by hiding this batch of functions with the preprocessor.

// COLORS2.PRG--Link this into your application.
// Colors: 1 Normal  2 Error Message  3 Prompt  4 F-key legend
STATIC acColor[4]
FUNCTION setstdcol (x, y)
LOCAL cRet := acColor[x]
IF y != NIL
 acColor[x] := y
// APP.CH-- #include this in all your application's source files.
#xtranslate setnorm   (<x>) => setstdcol (1, <x>)
#xtranslate seterr    (<x>) => setstdcol (2, <x>)
#xtranslate setprompt (<x>) => setstdcol (3, <x>)
#xtranslate setfkey   (<x>) => setstdcol (4, <x>)

With this set of directives and the single function setstdcol() you can keep the size of your sources and EXE down. At the same time the application programmer (you or someone else) can have a stable interface to this array of values no matter how large the application gets or how many new elements you want to add to the array.

There is a school of thought that believes that variables that are truly global in intent should just be made PUBLIC anyway.

8.4 What are the ground rules for using LOCALs and STATICs?

LOCALs and STATICs must be the first statements in a function, except possibly for the PARAMETERS statement. They may be declared with initial values (e.g. LOCAL a := 1), but STATICs may not be initialized with a value that is unknown to the compiler (e.g. STATIC b := myfunc()--where myfunc is the name of your user-defined function--is invalid). The initialization is performed every time the function is called (in the case of LOCALs) or once at the beginning of the program (in the case of STATICs).

If you want to use a file-wide STATIC variable to share information among functions within the same source module, that module must be compiled with the /n switch, and the STATIC declarations must precede the first FUNCTION or PROCEDURE.

8.5 What are INIT and EXIT procedures? Why do I care?

A procedure declared as INIT PROCEDURE <name> will always execute on program startup before the main procedure is executed. A procedure declared as EXIT PROCEDURE <name> will likewise execute just after the main procedure terminates.

INIT PROCEDUREs are useful if you have STATIC variable initializations that cannot be done inline. Both INIT and EXIT PROCEDUREs are handy for generalized setup and cleanup in a module that you want to make fully independent of the calling program.

INIT FUNCTION works the same as INIT PROCEDURE, and EXIT FUNCTION works the same as EXIT PROCEDURE. Their return values are ignored though.

Clipper makes no guarantees as to the order of execution of INIT and EXIT procedures.

On this point, Clayton Neff writes: If you have 3 or 4 INIT functions, you don't have any control over the order in which they are called. Since the RDD setup routines are also in an INIT function, Clipper won't necessarily have set up the RDD system before your INIT function is called. This would cause any data file access routines to crash. The same is true for EXIT routines as well.

The INIT and EXIT features worked fine in Clipper 5.01 but weren't documented until the 5.2 release.

8.6 What are the ANNOUNCE and REQUEST statements?

(Answer supplied by Doug Lee.)

REQUEST is used to force functions, procedures, and modules to be linked into an application whether or not they are referenced directly. REQUEST officially replaces the EXTERNAL statement provided in versions of Ca-Clipper prior to 5.2.

Functions and procedures are linked into an application automatically if they are referenced directly in the source code or if they are located in .OBJ files which are explicitly listed in the link script. However, if a function is not referenced until run time and is to be drawn from a library, it must be REQUESTed. Examples of this include INIT and EXIT procedures, functions and procedures referenced only from within index expressions or report forms, and MEMOEDIT user functions. If you have a report form which calls MEMOTRAN and your application makes no further reference to that function, you must include the line


in one of your source files.

ANNOUNCE is used to name modules for REQUESTing. By using ANNOUNCE, you can cause a whole collection of procedures and functions to be linked into an application with one REQUEST line. If you have a module containing a number of functions used exclusively by report forms, you may place a statement like


at the top of the module. You may then REQUEST REPUTILS to bring the whole module into the application.

REQUEST and ANNOUNCE are often used in conjunction with INIT and EXIT procedures to establish the environment of an application with minimal effort. For example, you could write a module similar to the following and include it in your local function library:


STATIC screen


* Preserve the screen which was active when this app was started screen := SAVESCREEN(0,0, 24,79)

* Initialize the Clipper environment SET SCOREBOARD OFF SET EXCLUSIVE OFF : :



* Restore the screen which was active when this app was started RESTSCREEN(0,0, 24,79, screen)


Then, including the line


would automatically cause the screen to be saved and restored across program execution, etc.

From: (Clayton Neff)
Subject: CA-Clipper Frequently Asked Questions
Date: Mon, 22 Apr 1996 15:38:33 GMT

A REQUEST statement tells the Clipper compiler to ask the linker to resolve an external reference. So, instead of calling an actual function, you can request the module that it resides in. This is most useful for making sure the proper RDD is linked in, since your code won't call the data access functions directly. An ANNOUNCE statement simply adds an identifier to the object code. This way you can label a particular module internally in such a way that the linker will find it, even if there is no actual function with that name. Again, this is useful with RDDs, since it is unlikely that there would be a function or procedure called DBFCDX.

8.7 Is it more efficient to use one multi-dimensional array or several single dimension arrays in a program? The array(s)would be static and called at the top of the program.

(Answer supplied by Mark Neidorff.)

Assembler programmers are very familiar with memory segments. Each segment is a 64k block of memory. Clipper uses the DS segment as a lookup table for several different types of information. In Clipper the DS segment is called DGROUP. Third-party libraries may be significant users of DGROUP. Once DGROUP is filled, a program crashes. This is normally not a concern for developers.

An entry in DGROUP is created for each static variable in a program. By minimizing DGROUP usage, a program is made more bullet-proof. Therefore, from the standpoint of DGROUP usage, a single static multi-dimensional array is more efficient than several single dimensional arrays.

From a completely different perspective, a multi-dimensional array is a model of data. Once you are comfortable with multi-dimensional arrays, you can dynamically store data in your code. A 50 record 3 field lookup table (dbf file) can be represented in memory with an array like this:

    // MyLookup[3,50]

At startup you can read the entire table into the array once and close the table saving a file handle and cutting down on disk lookups every time you browse or seek in the table. (This is especially significant on a network system) This type of modeling is not possible with single dimensional arrays. You also do not have to know the size of the array before it is created since you can dynamically resize the array with the aadd() function.

8.8 What is the meaning of the compiler warning "Ambiguous reference: GETLIST"?

GETLIST is an array variable that is created by CA-Clipper, by default, as the list of all currently used GET objects. (The @...GET command preprocesses into an AADD (Getlist, ...) call, and READ references Getlist also.)

If you don't declare GETLIST in your code the compiler won't realize that the variable will in fact exist at runtime, hence the warning. If this bothers you, you can put the statement MEMVAR GETLIST in a header file that is shared throughout the application, or even declare GETLIST as a LOCAL in functions that use GETs and READs if your program organization allows it.

8.9 I am getting lockups running my CA-Clipper application under DOS 6.0. It worked under DOS 5.0. What's going on?

If you are running Clipper 5.01 apps and DOS 6.0 and are running EMM386.EXE and using its NOEMS command-line parameter, then you should also use the NOVCPI parameter with EMM386 or include E0 in your CLIPPER environment variable.

This is due to a check that the Clipper 5.01 runtime performed to verify the existence of EMS that is not fully compatbile with the EMM386.EXE that comes with DOS 6.0.

Another option, if you really need EMS memory, is to link in the object provided in EMM501.ZIP from The Oasis. This will allow S'87 and 5.01 to correctly detect EMM memory.

If you move to CA-Clipper 5.2e or 5.3b this problem goes away.

8.10 Can you combine Borland C/C++ or Turbo C/C++ code with your CA-Clipper applications?

Last updated 2 February 1993: Contributions by Ed Boggan, Carl Nelson, Nick Pombra, Alex Matijaca, and Mark Zurier

Yes, if you follow these guidelines:

One workable set of switches seems to be -c -f- -ml -N-. For optimized C code, try -c -f- -ml -N- -Z -O -G -w.

Zurier wrote later: Well, it's a little old and really only applies to Clipper Summer '87 and 5.0 with BC++ 16-bit versions. It's up to you, but if you include it, please put that caveat in there. I neither use Clipper nor Borland anymore; I'm all Visual C++ and Java (Visual J++ for now due to the tight integration) and thank God the 32-bit world finally got rid of the stupid memory models <g>.

In fact, based on that, it seems the article is way out of date. I'd only use it if you think there are still 16-bit people out there. I don't.

8.11 Is there a way to recompile or relink the standard CA-Clipper libraries to make CA-Clipper fully compatible with the Borlandcompilers?

No. Floating point function calls are too tightly coupled with the compiler's code generation to allow that to be done.

It is possible to write BCD replacement functions for the very low-level functions that call the floating-point math libraries. (They have names such as __DVADD() and can be found in the map file output from any decent linker.) The problem with this approach is that any numeric _literals_ in the source code are already embedded into the object code as double-precision floats. There does not appear to be a good way for a replacement math library to deal with the fact that sometimes it will be handed BCD numerics and float numerics at different times.

8.12 How about C++?

Last updated 17 June 1991: contributions by Diane Lask

Any C++ function which interfaces to your CA-Clipper code must be declared consistently as extern "C" CLIPPER func (void)

This is an ANSI linkage specifier which turns off "name mangling." One disadvantage is that if a function name is overloaded of course you may only access one instance of that function name from your CA-Clipper code. (Otherwise how could CA-Clipper tell them apart?)

If your C++ code must call Extend API functions, then you should edit your copy of EXTEND.H or EXTEND.API to bracket all the function declarations with the following code:

#ifdef __cplusplus
extern "C" {
  /* Existing function declarations here. */
#ifdef __cplusplus

C++ code which is not called directly from your CA-Clipper application can be safely left "mangled." For Borland C++ users, the normal prohibitions on use of floating point math remain.

8.13 How do I compile Microsoft C 5.1 code?

The correct compiler switches are /FPa /Gs /Oals /AL. You may use floating point math without restriction.

With CA-Clipper 5.3 I am not sure. Anyone?

8.14 What about Microsoft C 6.0? And later?

Last updated 26 June 1992: contributions by Don Caton

Use the undocumented /Gh compiler switch and the LLIBCA.LIB library from Microsoft C 5.1. The library is available on the CLIPGER forum on Compuserve.

For those who do not read German fluently or who do not have regular access to Compuserve it might be easier to purchase the C 5.1 library from Microsoft as a "downgrade" from a licensed later version of their C/C++ compiler line. The number to call, in the United States, is 800-426-9400. At the recorded message press 2 to order the downgrade. It costs USD 25.00, or USD 15.00 without printed documentation. These prices only apply to registered users of Microsoft C or C++ compilers from release 6.0 forward.

CA-Clipper 5.3 was compiled with Microsoft C version 8.0, so if you are using 5.3 just get the latest Microsoft C compiler.

8.15 Can you use Borland's Turbo Debugger with the embedded C and assembler code in your CA-Clipper application?

Yes. For best results, compile the C code with the -v switch, assemble with /zi, and Turbo-link with /v. Then just fire up TD with the name of your executable and set a breakpoint at the beginning of your C or assembler function of interest.

I have found that TD286, Borland's 80286 protected-mode debugger, works as well without crowding CA-Clipper code out of memory. Still you are probably better off testing your C/assembler code from a small "shell" CA-Clipper program to save run-time memory.

It is difficult to use TD386 with a CA-Clipper program since the former cannot coexist with the EMM386 expanded memory driver, and the later runs better with expanded memory.

8.16 I want to write a C function that concatenates a list of strings whose lengths are unknown at compile-time. How can Ido this efficiently?

Last updated 30 October 1991: contributions by Brian Marasca

If you're concerned about performance, the best possible solution would be to define a maximum string length beforehand. If you can do that, this piece of code, submitted by Brian Marasca, is pretty high-performance:

#define MAXLEN 120   /* Set this to the maximum required length. */
    char *p, *q, *buf ;
    int cx = 1, parms = _parinfo ( 0 ) ;

    q = buf = _xgrab (MAXLEN + 1) ;
    while (parms--)
         p = _parc (cx++) ;
         while (*p) *q++ = *p++ ;
    *q = '\0' ;       /* very important! */
    _retc (buf) ;
    _xfree (buf) ;

8.17 I want to call my C or assembler function with a reference parameter in a string. Why can't you just use the pointer returnedby _parc() to modify a string parameter?

Last updated 23 July 1991; Contributions by Ted Means

Clipper 5.0 performed under-the-hood memory optimizations that sometimes resulted in two strings being assigned the same pointer.

For example, if you did this:

       X := "Some string"
       Y := X

In all likelihood Clipper would simply make a copy of the pointer rather than actually copying the string. It sounds great in theory, because it saves memory. But problems arose when a C or ASM routine attempted to modify X or Y by writing directly to the assigned memory. Since the same pointer was assigned to two different memvars, they both got changed, even though that was not the intended effect.

So with the release of Clipper 5.01, Nantucket graciously provided a method for accessing strings that is more convenient. If you call __StorCLen with a null pointer *before* calling __ParC, it warns Clipper that you plan to directly modify a __ParC pointer and Clipper will take whatever steps are necessary to make it safe to do so.

This gives the best of both worlds, in that Clipper can still perform its memory optimizations, but C and ASM programmers get the benefit of being able to directly modify strings. This is all documented in the Release Notes, but briefly, if you need to write directly to a string, use the following sequence:

Writing directly to a __ParC pointer without using this method is not sanctioned by Nantucket and is likely to cause problems. Note, though, that if you don't need to write to the string you can simply call __ParC by itself.

8.18 How do I call CA-Clipper code from C? Can this even be done?

Last updated 25 November 1991; contributions by Steve Larsen and Jobst Hensiek

With versions prior to 5.2, you have to set up Clip's internal EVAL stack properly with _xpushf() and _xpopf(). Afterwards, call _xdo() to execute the CA-Clipper function. The whole thing is undocumented, and it is discouraged by CA because it requires the use of internal functions which are not guaranteed to remain unchanged from one CA-Clipper version to the next. Most people spend weeks poking around the symbol tables to get this to work.

CA-Clipper 5.2e's well-documented Item API (Application Programming Interface) allows you to evaluate CA-Clipper expressions, and view or change CA-Clipper variables, directly from C code. You can request this documentation free from CA if you have purchased CA-Clipper 5.2e.

It appears to be very difficult if not impossible to call CA-Clipper code successfully from within a C "main" program. The environment must be CA-Clipper calling C calling CA-Clipper for this stunt to work.

Next Previous Contents