The CMOC C-like 6809-targeting cross-compiler

The CMOC C-like 6809-targeting cross-compiler

By Pierre Sarrazin

Copyright © 2003-2018

Distributed under the GNU General Public License, version 3 or later (see the License section).

Version of CMOC covered by this manual: 0.1.48

Date of this manual: 2018-02-16

Table of Contents


CMOC is a Unix-based program that compiles a C-like language, generates Motorola 6809 assembly language programs and assembles them into executables for the Color Computer's Disk Basic environment. (As of version 0.1.18, targeting the Vectrex video game console is also supported.)

CMOC is not meant to be able to compile existing C programs. It is a tool to develop new programs for the Color Computer (CoCo).

The efficiency of the generated machine language is modest, but the resulting machine language will be faster than the equivalent interpreted Color Basic program. This was the initial goal of the CMOC project.

The main features missing from C are const and bit fields.

CMOC itself is written in C++ for a Unix-like system. The C++ source code is somewhat complicated because of the anarchic development process that gave rise to CMOC. Years ago, development was able to take off because having a working product had higher priority than having a clean but theoretical design.


Unsupported features

Supported C and C99 features

Differences with C

CMOC is sometimes more strict than C on some points. For instance, it is an error to call an undeclared function.

If a large discrepancy exists between C and CMOC, a bug report should be filed.

Installing CMOC

The following instructions assume a Unix-like system. A C++ compiler is necessary to compile CMOC.

The compiler

Deploy the .tar.gz source archive in a directory, go to this directory, read the generic installation instructions in the INSTALL file, then (typically) give these commands:

tar -xzf cmoc-0.1.XX.tar.gz
cd cmoc-0.1.XX
./configure --prefix=/some/installation/dir
make install

The sudo prefix may be needed to do a make install to a system directory like /usr/local/bin.

The "check" makefile target runs several unit tests.

The compiler executable is called "cmoc".

To generate the HTML documentation (this document), do make html, which will create the file doc/cmoc-manual.html.

Running CMOC

Compiling a C program

The following must be in the search path:

To compile a program written in the file foo.c, run this command:

cmoc foo.c

By default, the resulting .bin file will load at address $2800 on a CoCo (see the #pragma org directive elsewhere in this manual). The generated machine language is position-independent, so the .bin file can be loaded with an offset, by passing a second argument to Disk Basic's LOADM command.

Pass --help to cmoc to see the available options.

CMOC comes with a writecocofile command that can copy a .bin file to a 35-track Disk Basic diskette image file. For example:

writecocofile testing.dsk foo.bin

Pass --help to writecocofile for the available options.

For more information on running a CoCo emulator on a GNU/Linux system, see

Disabling some optimizations

By default, all peephole optimizations are enabled.

The peephole optimizer can be disabled by passing the -O0 option to the compiler. This can be useful when one suspects that the compiler may have generated invalid code.

Several peephole optimizations were contributed by Jamie Cho in early 2016 for version 0.1.20. They can be disabled by passing -O1.

Option -O2 is equivalent to using the default (full) optimization level.

Generated files

From the .c file passed to cmoc, the compiler generates these files:

When linking, these files are generated:

When distributing the .bin file to CoCo users, it is not necessary or useful to accompany it with the other generated files.

These files are deleted automatically when the compiler finishes, except for the executable. To keep those intermediate files, pass --intermediate (or -i) on the command line.

The Motorola S-record (SREC) format

Pass the --srec option to cmoc to make it generate an executable in Motorola S-record format. The executable file will have the .srec extension.

Since version 0.1.43 of the compiler, this format is the default one when targetting the USim 6809 simulator (with the --usim option).

The Vectrex video game console

The doc subdirectory of the source distribution contains cmoc-vectrex.markdown, which gives Vectrex-specific instructions. Giving the make html command will generate cmoc-vectrex.html, which can be viewed in a browser.

Note that questions regarding Vectrex-specific issues should be addressed to Johan Van den Brande at

Modular compilation and linking

Since version 0.1.43, CMOC has supported modular (separate) compilation as well as linking several modules (object files) together.

In other words, the user can now separate a large program into several .c files, compile each of these files to a .o object file, then have CMOC link them together into a final executable.

LWTOOLS' lwasm and lwlink commands must be in the search path, otherwise, the user can use the --lwasm= and --lwlink= command-line options to specify the full path of these commands.

Creating libraries

Object files (.o) can be grouped into a library by using LWTOOLS' lwar command:

lwar -c libstuff.a foo.o bar.o baz.o

Library filenames should have the lib prefix and be given the .a extension, so that they will be compatible with the -l option that is used to specify a library to link with, e.g.:

cmoc -o prog.bin prog.c -lstuff

In this case, only the .o files of libstuff.a that are actually needed will be copied to the executable.

It is possible to specify libstuff.a on the CMOC command line, but then all of the library's object files get copied to the executable.

The -l option must be specified after the source or object file(s).

The -L option can be used to specify a directory in which to search for libraries specified by -l. The -L option can be specified before or after the source or object file(s).

An object file is indivisible as far as the linker is concerned, and therefore, so is a C file. This means that either all of the functions and globals in a C file will end up in the executable, or none of them will. When designing a library, it may be desirable to put each function in its own C file, so that only the functions used by the parent program will be copied into the executable.

Multiple definitions

Because of the way the linker works, there is no error message when the same function or global variable is defined with external linkage by two modules. Only a warning is issued. Such a warning should be viewed as an error and the duplication should be resolved.

For similar reasons, warnings will be issued when two modules define static functions or globals with the same name, e.g., static int n;. Such warnings may be safely ignored, because the symbols are static, thus not exported, and will not actually clash.

These ambiguous diagnostic messages may be fixed by future versions of CMOC and LWTOOLS.

Specifying code and data addresses

In a modular program, the address at which code and data must be positioned must not be specify with the #pragma org and #pragma data directives. Those addresses must be specified with the --org and --data command-line options when invoking CMOC to perform linking:

cmoc -c mod1.c
cmoc -c mod2.asm
cmoc -o prog.bin --org=C000 --data=3000 mod1.o mod2.o

The same goes for #pragma limit: the --limit option must be used instead.

Starting with version 0.1.43, these pragrams can only be used when: * compiling directly from a C file to an executable (i.e., the -c option is not used); * using the --monolith option.

Uncalled functions

Non-static functions that are not called are still emitted, because they might be invoked by another module. This cannot be determined before the linking phase.

Assembly language modules

An object file typically comes from a C file, but an assembly language file can also be passed to CMOC, which will invoke the assembler on it. That file's extension must be .s or .asm.

There are conventions to be observed however:

        SECTION code  
_func   LDD     #42  
        SECTION initgl
        LDD     #1000
        STD     _someGlobal,PCR

Importing symbols used by inline assembly

If an inline assembly block uses a global variable that is provided by another module, an import directive must be included:

void f()
_foo    IMPORT
        ldd     #42
        std     _foo

Generating Prerequisites Automatically

A multi-module project typically uses a makefile to manage the build process.

CMOC can automatically generate a dependencies file, with a .d extension, that lists the source files that were encountered during the compilation of a C file. The dependency file will be compatible with GNU Make.

This is done by passing either the --deps-only or --deps option to the compiler. A .d file will be generated with a makefile rule like this:

main.o main.d : main.c foo.h bar.h baz.h

Option --deps-only stops right after generating this dependency file. Option --deps can be used with -c to have the compiler both compile the C file and generate the dependency file.

The Generating Prerequisites Automatically section of the GNU Make manual should be consulted for details on how to write a makefile that uses this mechanism. GNU Make's -M is similar to --deps-only, while -MMD is similar to --deps. Note that the sed command mentioned in that manual is not necessary with CMOC, which automatically includes the .d filename in the generated rule.

The name of the .d file is formed from the name of the .o file to be generated.

This option is not available in monolithic mode and it has no effect when also specifying the -E option (which prints the preprocessor output and stops).

Legacy monolithic mode

The legacy "monolithic" mode is available when the --monolith option is passed. In that mode, only one .c file may be specified. It will be compiled directly to an executable. During that compilation the CMOC_MONOLITH preprocessor identifier is defined. This can be used to alter the code depending on the compilation mode.

Monolithic mode is deprecated and may be removed in future versions of CMOC.

As of version 0.1.43, OS-9 and Vectrex support is only available in the monolithic mode.

Programming for CMOC

Signedness of integers

The compiler generally produces more efficient code for unsigned arithmetic. Therefore, types unsigned and unsigned char should be preferred to int and char where possible.

CMOC considers that the signedness of an additive or multiplicative operation is the signedness of the left operand.

CMOC can issue a warning when comparison operators <, <=, > or >= are used with operands of different signedness. This is obtained by passing the -Wsign-compare command-line option.

Pre-increment vs. post-increment

The pre-increment (++i) typically generates one instruction less than the post-increment (i++). This is because the post-increment must yield the initial value of the variable. When this initial value is not needed, it is advisable to use the pre-increment.

The same situation holds for the decrement operators.

Origin address

To specify the origin address of the program, start your program with a #pragma org directive. For example:

#pragma org 0x5C00

By default, the origin address is 0x2800. (Under Disk Basic, the Basic program normally starts at 0x2601, because four PMODE pages are reserved by default.)

End address and length of the executable

To determine the effective start and end address of the executable (assuming no relocation by LOADM), one can look up the program_start and program_end symbols in the .lst listing file generated by the assembler.

program_end is useful to check that the executable fits in the available RAM. On a 32k CoCo, the RAM ends at $8000. By defaut, Color Basic reserves 200 bytes for Basic strings, and before that, the system stack should be given 512 bytes for a typical C program. These assumptions mean that an executable should not go beyond $7D38.

If a program crashes the CoCo just by getting loaded, it could be because it is too long and overwrites the system stack. The parse-coco-bin Perl script, available on the CMOC home page, can be useful to confirm that.

The following example program prints the start and end of the memory it uses initially:

#include <cmoc.h>
int main()
    char *s, *e;
        leax    program_start,pcr
        stx     s
        leax    program_end,pcr
        stx     e
    printf("START AT %p, END AT %p.\n", s, e);
    return 0;

Note that the label addresses are taken relatively to the program counter (,pcr), so that this program will report the correct addresses even if the program is loaded at an address other than the one specified in the .bin file. (This can be done by passing a second argument to Disk Basic's LOADM command.)

This section assumes that the code and data sections form a single block. When using the #pragma data directive (see elsewhere in this manual), the writable globals will not be between program_start and program_end.

As of version 0.1.20, CMOC automatically displays the program start and end addresses, as well as the start and end of the writable data section, when the --verbose option is passed. In the following example, #pragma org and data have been used to position the code and data separately:

Notable addresses:
  program_start             $C000
  program_end               $DC39
  writable_globals_start    $3800
  writable_globals_end      $3B7A

Enforcing a limit address on the end of the program

As of version 0.1.20, CMOC accepts the --limit=X option. When it is passed, CMOC will fail if the end of the program, as indicated by the program_end listing symbol, exceeds address X (specified in hexadecimal).

For example, --limit=7800 will keep the program from finishing too close to the system stack under Disk Basic, which is not far below $8000. A limit at $7800 leaves two kilobytes for the stack and for Basic's string space.

It is not necessary to pass the --verbose option to use the --limit option.

A #pragma limit 0xNNNN directive in the source code achieves the same purpose as --limit.

Position-independent code

All 6809 code generated by CMOC is position-independent, i.e., it can be loaded at any address and will still work.

Determining that CMOC is the compiler

Since version 0.1.48, CMOC automatically defines _CMOC_VERSION_ to be a 32-bit number of the form XXYYZZZ when the version is X.Y.Z. For example, version 0.1.48 defines _CMOC_VERSION_ as 1048. This can be useful when some C code needs to do something under CMOC but something else under another compiler.

Targeting CoCo Disk Basic or the USim 6809 simulator

By default, the compiler defines the _COCO_BASIC_ preprocessor identifier. This identifier can be used to adapt a program to make it use alternative code depending on whether it will run under Disk Extended Color Basic or not.

By passing the --usim option, the compiler will target the USim 6809 simulator, which comes with CMOC. No .bin file is produced in this case. The .hex file can be executed by passing its name to src/usim-0.91-cmoc/usim.

The standard library

CMOC's standard library is small. The program must #include <cmoc.h> to use functions like printf(). See that file for a list of implemented C functions. Most are standard C functions while others are CMOC extensions.

dwtoa(), meaning "double word to ASCII", is not a standard C function. It takes an 11-character buffer and two unsigned integers. It converts the 32-bit value formed by these integers into an ASCII decimal represented stored in the buffer. The address of the first character of this decimal string is returned. For example, this code returns a pointer to a string equal to "65536".

char buf[11];
char *p = dwtoa(buf, 1, 0);

Since version 0.1.41 however, it is simpler to use the long type and to call printf() or sprintf() with format %ld or %lu to obtain an ASCII decimal representation.

readline() acts like Basic's LINE INPUT command and returns the address of the string entered.

Each assembly program generated by CMOC includes, which assembles to about 2000 bytes of machine code. In the future, a mechanism could be implemented in the compiler to reduce this overhead. In the mean time, one can remove the unneeded parts of this file, if memory is at a premium.

Size of the standard library

As of version 0.1.9, the standard library takes up almost 2048 bytes when all of its functions are called by the program.

But as of that version, only the standard library functions that are actually called are included in the final assembly program.

A program that calls none of these functions will shed about 2000 of those bytes.


CMOC's printf() function supports %u, %d, %x, %X, %p, %s, %c, %f and %%. Specifying a field width is allowed, except for %f. A left justification is only supported for strings, e.g., %-15s will work, but %-6u will not. Zero padding for an integer is supported (e.g., %04x).

The l modifier is supported for 32-bit longs (e.g., "%012ld");

%p always precedes the hexadecimal digits with $, as per the CoCo assembly language notation. %x and %X do not generate such a prefix. %p, %x and %X always print letter digits as capital letters (A to F, not a to f).

printf(), like putchar() and putstr(), sends its output one character at a time to Color Basic's PUTCHR routine, whose address is taken from $A002.

Redirecting the output of printf()

The standard library's printf() writes the characters of the formatted string by calling the routine whose address is stored in the library's CHROUT global variable (not to be confused with Color Basic's CHROUT vector at $A002). The same applies to functions sprintf(), putchar() and putstr().

By default, under Color Basic, that routine is the one found in that $A002 vector. To designate a C function as the new character output routine, first define the new routine:

void newOutputRoutine()
    char ch;
        pshs    x,b  // preserve registers used by this routine
        sta     ch

    // Insert C statements that print or otherwise process 'ch'.

        puls    b,x

This routine will receive the character to be printed in register A. It must preserve registers B, X, Y and U. It does not have to preserve A.

Install it at the appropriate time with this call:

void *oldCHROOT;

oldCHROOT = setConsoleOutHook(newOutputRoutine).

To restore the original output routine, do this:



This function is like printf(), but it writes into a memory buffer whose address is passed as the first argument, before the format string. For example:

char greeting[32];
void f(char *name)
    sprintf(greeting, "Hello, %s.", name);

Calling f("Lonnie") will write "Hello, Lonnie." in the greeting[] array, including a terminating '\0' character. A total of 15 bytes get written to the start of that array.

The caller is responsible for providing a buffer long enough to receive all the text resulting from the format string and its arguments, including the terminating '\0' character.

In this example, the longest "name" that can be safely passed to f() would be a 23-character name.

The standard C language offers snprintf(), which is safer because it requires passing the length of the destination buffer. But checking for this limit would have a performance hit that is not necessarily acceptable on a CoCo. If such a function is needed, it can be implemented using the technique described in the previous section.

Redefining a standard library function

Defining a function using the name of a standard library function is allowed. For example, a program could redefine rand() and srand() to implement an alternative random number generator. In the final assembly program, such functions replace the ones provided by the standard library.

Dynamic memory allocation with sbrk()

sbrk() is a function that dynamically allocates a region of memory of the size (in bytes) passed to it as an argument. It returns a void pointer. If the quantity of memory requested is not available, sbrk() returns (void *) -1. For example:

void *p = sbrk(100);
if (p != (void *) -1)
    memset(p, 'X', 100);

The name of the function comes from the notion of a "program break", which is the current end of the memory allocated to the program. The memory after that is presumed to be free for dynamic allocation.

In the case of the CoCo, the assumption is that the program is loaded after the Basic program and variables. This means the space that sbrk() can allocate from goes from there to the top of the stack, which is around $3F00 on a 16K CoCo and $7F00 on a 32K-or-more CoCo.

Use the --stack-space option or the #pragma stack_space directive (documented elsewhere in this manual) when the program needs more than 1024 bytes of stack space.

To determine how much of that memory is available for sbrk(), call sbrkmax(), which returns the number of bytes as a size_t (unsigned). CMOC ends the region available to sbrk() about 1024 bytes before the stack pointer, leaving those bytes to program calls and local variables.

sbrkmax() returns 0 if the program is loaded after the stack space.

Be careful when using both sbrk() and #pragma const_data (see previous section). If the data is not positioned correctly, sbrk() calls may end up allocating memory that is used by some global variables.

Inline assembly

Inline assembly text can be specified by surrouding it with asm { and }. In the text, one can refer to C variables (global, local and parameters) as well as functions. Labels can be used for branch instructions, but a label must either be unique to the whole program or comply with what lwasm considers a "local" label. Prefixing a global label with the name of the current C function is a good way to help prevent clashes. A label must appear at the very beginning of the line, without spaces or tabs in front of it.

One way of using lwasm local labels is to prefix the label name with the @ character. Such a label will be local to the current block, which will begin at the previous blank line (or start of the asm block) and end at the next blank line (or end of the asm block). Refer to the LWASM manual about its symbols for details on using local labels.

The assembler may also support $ as a local label marker, but it is not recommended to use it that way in inline assembly because it may hinder portability to OS-9, where $ is not used as a local label marker.

The following example fills array out with n copies of character ch, then returns the address that follows the region written to:

#include <cmoc.h>

char *f(char *out, char ch, unsigned char n)
    char *end;
        ldx     out         /* comments must be C style */
        lda     ch          // or C++ style
        ldb     n           ; or semi-colon comment (passed to assembler)
        sta     ,x+
        bne     f_loop
        stx     end
    return end;

int main()
    char a[10];
    a[9] = '\0';
    char *p = f(a, 'X', (unsigned char) sizeof(a) - 1);
    printf("a='%s', %p, %p\n", a, a, p);
    return 0;

Since version 0.1.21, when referring to a C function, the function name is replaced with its assembly label, possibly followed by the ,pcr suffix. This suffix is omitted if the instruction is BSR, LBSR or JSR, because these instructions do not support the ,pcr suffix and they do not need it anyway. The following example calls the same C function three different ways:

    jsr     someCFunction
    lbsr    someCFunction
    leax    someCFunction
    jsr     ,x

Note that CMOC always generates position independent code. This rule should be maintained in inline assembly if the resulting program is to be relocatable.

The BSR instruction is not recommended because it is a short branch and if the called function is too far from the call, the assembly step will fail.

Since 0.1.39, semi-colon comments are supported. They are passed verbatim to the assembler.

Note that using inline assembly is likely to make the program non portable to other C compilers.

See the Calling convention section elsewhere in this manual for the rules to observe. Namely, inline assembly must not modify U or Y. It is allowed to modify D, X and CC.

Preprocessor identifiers in inline assembly

The GNU C preprocessor can add spaces in surprising ways, which makes its use slightly problematic in inline assembly. For example:

#define PIA0 0xFF00

    stb     PIA0+2

One would expect this code to generate an stb 0xFF02 instruction, but cpp will actually expand this to stb 0xFF00 +2, because it apparently adds a space after the expansion of the PIA0 identifier.

lwasm takes this space as the start of the comment, so it ignores the +2 and assembles stb $FF00.

A workaround appears to be reverse the addition and write stb 2+PIA0. No space gets added before the identifier.

Therefore, preprocessor identifiers should be used with caution in inline assembly.

Referring to variables whose name is that of a register

Before version 0.1.31, bad assembly language text could be emitted if inline assembly referred to a C variable that has the name of a register. To help resolve this ambiguity, version 0.1.31 introduced a C variable escape character, which is a colon that is put in front of the C variable name.

For example:

char b;
    inc     :b
    ldb     :b
    leax    b,x

Here, :b refers to variable b of the C program, while the b in b,x refers to the register.

This change may break inline assembly code in programs that were written for versions of CMOC preceding 0.1.31. Adding a colon at the right places will resolve the issue.

Note that the escape character is not necessary on variable names that are not also register names.

Assembly-only functions

When a function is written entirely in assembly language for performance reasons, the stack frame may not be necessary. CMOC will not generate a stack frame for a function defined with the asm modifier, as in this example:

asm int f(int m, int n)
    // U not pushed, so 1st argument is at 2,s
        ldd 2,s     // load m
        addd 4,s    // add n, leave return value in D

Only asm { ... } statements are allowed in such a function. Typically, only one is needed. Local variables cannot be defined in that function and the function's parameters cannot be accessed by name. The assembly code is allowed to refer to global variables however.

By default, the compiler ends the function with the RTS instruction (or RTI if the function is declared with the interrupt modifier). To keep the compiler from emitting that instruction (RTI as well as RTS), add the __norts__ keyword:

asm __norts__ int f(int m, int n) { ... }

See the Calling convention section elsewhere in this manual for the rules to observe. In particular, note that byte arguments are promoted to words, which are pushed onto the stack in the big endian byte ordering.

Interrupt Service Routines

CMOC supports the interrupt function modifier, which tells the compiler to end the function with an RTI instruction instead of an RTS. For example, the following function handles the VSYNC 60 hertz interrupt:

interrupt void newCoCoIRQRoutine()
        lda     $FF03           // check for 60 Hz interrupt
        lbpl    irqISR_end      // return if 63.5 us interrupt
        lda     $FF02           // reset PIA0, port B interrupt flag

    // Do something in C.

    // Nothing here, so that next instruction is RTI.

This routine could be hooked to the IRQ vector this way on a CoCo:

unsigned char *irqVector = * (unsigned char **) 0xFFF8;
*irqVector = 0x7E;  // extended JMP extension
* (void **) (irqVector + 1) = (void *) newCoCoIRQRoutine;

Header <coco.h> defines macros disableInterrupts() and enableInterrupts(). They set and reset the F and I condition codes.

The FIRQ vector is at $FFF6.

Note that using the interrupt keyword is likely to make the program non portable to other C compilers.

Function pointers

The address of a function can be taken and stored in order to be called through that pointer later.

The following example shows that the two syntaxes used in C to call a function through a pointer are supported by CMOC:

unsigned char f(int a, char b) { ... }
int main()
    unsigned char (*pf)(int a, char b) = f;
    unsigned char c0 = (*pf)(1000, 'x');
    unsigned char c1 = pf(1001, 'y');
    return 0;

A member of a struct can point to a function. For example:

struct S
    void (*fp)();
void g() { ... }
int main()
    struct S s = { g };  // member 'fp' points to g()
    s.fp();  // call g()
    (*s.fp)();  // call g()
    return 0;

Array initializers

Local vs. global array

When only a single function needs to use a read-only array of integers, this array could be defined locally to that function, but it is more efficient, as of CMOC 0.1.10 to define the array as global. This is because the global array will be emitted as a sequence of FCB or FDB directives, while a local array will be initialized with a series of load and store instructions.

In the following example, array g will be initialized more efficiently than array l:

int g[] = { 55, 66, 77 };
void f()
    int l[] = { 22, 33, 44 };
    /* ... */

Execution of global array initializers

A global array initializer containing only integer values (not string literals) is treated specially. It is initialized at compile-time, not at run-time. This means that if the program modifies the array, quits, and then is reexecuted, the modified contents will still appear as is. The array will not be reinitialized to the values in the initializer.

For example:

#include <cmoc.h>
int a[2] = { 13, 17 };
int main() { a[0]++; printf("%d\n", a[0]); return 0; }

The first time this program is executed, a[0] starts at 13, then is incremented to 14, which is the number that is printed.

The second time this program is executed, a[0] starts at 14 because array a is not reinitialized upon entry.

This is meant to save memory by not including a second copy of the initializer just for run-time initialization purposes.

Read-only global arrays

When compiling for a CoCo ROM cartridge, global arrays that have an initializer and that are not meant to be modified should be mapped in the ROM part of the program. This is done with the #pragma const_data start and #pragma const_data end directives. See the Compiling a ROM image for a cartridge section for details.

Array sizes

One must be careful when specifying an array size using an arithmetic expression, like this:

char a[256 * 192 / 8];

This will generate an "invalid dimensions" error message. This is because the three numerical constants are of type int, which means they are signed integers. In 16-bit signed arithmetic, 256 * 192 is -16384, which gets divided by 8, which is -2048. This size is rejected by the compiler, which only accepts array sizes between 1 and 32767 bytes inclusively.

The fix is to force the expression to be evaluated with unsigned integers:

char a[256U * 192U / 8U];

This will be 6144 bytes, as intended.

Compiling a ROM image for a cartridge

To help support the ROM cartridge format, CMOC supports directives that allow telling it to assemble the code, the string literals and the read-only global variables at the typical cartridge ROM address of $C000, while mapping the writable global variables at a RAM address like $3800.

This is achieved by using the four #pragma directives that appear in this example:

#pragma org $C000
#pragma data $3800

int f() { return 42; }
int writableGlobal = f();

#pragma const_data start
int readOnlyGlobal = 100;
unsigned char byteArray[3] = { 11, 22, 33 };
char text[] = "hello";
#pragma const_data end

int anotherWritableGlobal;

int main()
    anotherWritableGlobal = 99;
    return 0;

Note that writableGlobal is allowed to have a run-time initializer because it is writable.

readOnlyGlobal, byteArray and text are read-only because they appear between the const_data directives. More than one such section can appear in the same program.

In the case of text, the use of the empty brackets is necessary: declaring text as char *text would be rejected by CMOC. This is because text would then be a pointer whose initialization would have to be at run-time, because the address of a string literal depends on where the program is loaded by the user.

anotherWritableGlobal is writable because it is not in a const_data section. It is allowed to have no initializer because it is writable.

Using sbrk() can be dangerous depending on where in memory the writable data section is positioned.

After compiling the program, the .bin file normally contains a single contiguous block of code. This block must be extracted from the .bin file and, for a test with the XRoar emulator, it must then be padded at the end with enough bytes so that the total file length is a multiple of 256. The following Perl script does this:

sysread(STDIN, $h, 5) == 5 or die;
sysread(STDIN, $rom, 0xFFFF) > 0 or die;
my $romLen = length($rom) - 5;
binmode STDOUT or die;
print substr($rom, 0, $romLen);
my $extra = $romLen % 256;
print chr(0) x (256 - $extra) if $extra;

The script, in a file called, can be used this way:

perl < foo.bin > foo.rom

This ROM image can be tested in the XRoar emulator this way:

xroar -machine cocous -cart-autorun -cart foo.rom

Note that XRoar requires the image file to have the .rom extension.

In a program written as above, the IRQ interrupt is not enabled, so Basic's TIMER counter (at $0112) does not get incremented. To enable the IRQ in your catridge-based program, put this at the beginning of the main() function:

    // We come here from a JMP $C000 in Color Basic (normally at $A10A in v1.2).
    // At this point, the 60 Hz interrupt has not been enabled yet, so enable it.
    lda     $ff03   // get control register of PIA0, port B
    ora     #1
    sta     $ff03   // enable 60 Hz interrupt

    // Unmask interrupts to allow the timer IRQ to be processed.
    andcc   #$af

Enumerations (enum)

Enumerations are supported, with the restriction that an enum must be declared at the global scope. In particular, an enum with an enumerator list (e.g., enum { A, B }) cannot be used in the formal parameter of a function, nor as its return type.

An enum can be anonymous (e.g., enum { A, B }) or be given a name (e.g., enum E { A, B }).

Each enumerated name can have a specified value, e.g., enum { SCREEN = 0x0400, BIT5 = 1 << 5 }. Such a value must be a constant expression.

Floating-point arithmetic

The float and double keywords and floating-point numeric literals (e.g., 1.2f or -3.5e-17) have been supported since version 0.1.40, but only under the Color Computer's Disk Basic environment.

A warning is issued when the double keyword is encountered, so the user knows not to expect double-precision. There is also a warning when a numeric literal does not use the f suffix, which specifies that the literal is single-precision. There too, double-precision must not be expected. It is recommended to code programs using the float type and the f suffix.

The compiler will fail to compile a program that uses floating-point arithmetic when a platform other than Disk Basic is targeted.

CMOC's printf() function supports the %f placeholder, but as of version 0.1.40, it does not support the width or precision parameteres of %f (e.g., "%7.3f").

The <cmoc.h> header file provides functions strtof() and atoff() to convert an ASCII decimal representation of a floating-pointer number into a float value, as well as ftoa(), to convert a float value into an ASCII decimal representation.

Function Names as Strings

The constants __FUNCTION__ and __func__ can be used to refer to the current function's name:

printf("Now executing %s().\n", __func__);

In the global namespace, these identifiers give the empty string.

Detecting null pointer accesses at run time

Accessing a value through a null pointer is a common bug. To help detect such accesses, CMOC has the --check-null command-line option, which adds run-time checks before every use of a pointer and every array element access.

By default, the handler that is invoked when a null pointer access is detected is a no-op. The program should start by calling the set_null_ptr_handler() function to set up a handler that will receive the address of the failed check as an argument. For example:

#include <cmoc.h>

struct S { int n; };

void nullPointerHandler(char *addressOfFailedCheck)
    printf("[FAIL: %p]\n", addressOfFailedCheck);

int main()
    struct S *p = 0;
    p->n = 42;
    return 0;

This program will fail and display an address. One can then look up this address in the .lst listing file generated by CMOC to determine in which function that null pointer was detected.

Using this option incurs a performance cost, so it is only recommended during debugging. An alternative is to define an assert() macro that expands to nothing when NDEBUG is defined.

Detecting stack overflows at run time

Runaway recursion or excessively long local arrays can cause corruption that is difficult to investigate. To help detect stack overflows, CMOC has the --check-stack command-line option, which adds run-time checks at the beginning of each function body.

When passing --check-stack to CMOC, the program should start with a call to set_stack_overflow_handler() that installs a handler. This handler receives two arguments: the address of the failed check and the out-of-range stack pointer. The handler must not return. For example:

void stackOverflowHandler(char *addressOfFailedCheck, char *stackRegister)
    printf("[FAIL: %p, %p]\n", addressOfFailedCheck, stackRegister);
void recurse() { recurse(); }
int main()
    return 0;

This program will fail and display two addresses. One can look up the first one in the .lst listing file generated by CMOC to determine in which function that stack overflow was detected.

Using this option incurs a performance cost, so it is only recommended during debugging.

The preprocessor identifier _CMOC_CHECK_STACK_OVERFLOW_ is defined by CMOC when --check-stack is used. This identifier can be used to exclude stack check code when not needed.

By default, CMOC allocates 1024 bytes for the stack on a CoCo (256 bytes on a Vectrex). In a program that is compiled with the default layout, the stack is preceded by the memory managed by sbrk() (see elsewhere in this manual).

Note that this feature is not usable under OS-9, where stack checking is automatic and uses a different mechanism.

Specifying the space allocated to the system stack

Option --stack-space=N (with N > 0) can be used to specify the number of bytes (in decimal) to be reserved to the stack. This affects the quantity of memory available to sbrk() and the stack checking done by --check-stack. Use this option when the program uses lots of stack space. Note that this option is not permitted when targeting the Vectrex.

The stack space can also be specified with #pragma stack_space N. The command-line option has precedence over this pragma.

Specifying the stack space must be done when compiling the C file that defines main(). It has no effect when compiling a file that does not define main(), nor when calling CMOC to perform linking on a list of object files.

Note that this feature is not usable under OS-9, where stack checking is automatic and uses a different mechanism.

OS-9 stack checking

When targeting OS-9, by default, a stack check is performed upon entering each function, unless that function is assembly-only.

The stack check verifies that there will be at least 64 bytes of free stack space once the function will have allocated stack space for its local variables. If this check fails, a "stack overflow" error message is printed in the standard output and the program exits.

To change this 64 to another (non-zero) value, use the command-line --function-stack option. Pass 0 to this option to disable the stack check.

Single-execution programs

If a program to be run under Disk Basic is intended to be executable only once, i.e., that reloading it or rebooting the CoCo is required to run it again, then the code that initializes the writable global variables is not needed upon entry to function main().

This routine is assembled at the very end of the code section. When the program specifies #pragma exec_once, then the "program break" used by sbrk() (see the previous section) is placed at the start of the routine. This makes the memory of that routine available to be allocated by sbrk().

Calling convention

CMOC follows the C convention of passing the parameters in the stack in the reverse order.

The caller pops them off the stack after the call.

An argument of type char, which is signed, is promoted to int.

An argument of type unsigned char is promoted to unsigned int.

When the return value is requested to be an r-value, is expected in the B register for a byte-size type or D for a word-size type. For an l-value, the return value must be left in X.

The body of a function must preserve registers U and S. It is allowed to modify A, B, X and CC.

Under OS-9, CMOC uses Y to refer to the data section of the current process. Any code that needs to use Y must preserve its value and restore it when finished. For portability, this rule should also be observed on platforms other than OS-9.

The compiler's low-level optimizer may emit code that uses Y when targeting a platform other than OS-9.

The called function does not have to return any particular condition codes.

Register DP is not used or modified by CMOC-generated code.

A CMOC-generated function uses a stack frame if the function receives one or more parameters or has local variables, and is not defined with the asm keyword (see the Assembly-only functions section elsewhere in this manual). Register U is used as the stack frame pointer. Such a function starts with these instructions:

    PSHS    U       
    LEAU    ,S

an ends with these:

    LEAS    ,U
    PULS    U,PC

Calling a program as a DEF USR routine

Typically, such a program gets loaded at the end of Basic's RAM area, i.e., &H7A00. Pass --org to the compiler to position the program at such an address. In the Basic program, reserve some high RAM to your CMOC program, i.e., CLEAR 200,&H7A00. Then use LOADM, then DEF USR to define a user routine that points to your CMOC program, e.g., DEF USR5=&H7A00. (The 5 in this example can be a number from 0 to 9.) Finally, call the routine using the routine number and an argument, e.g., R=USR5(1000).

Compiling a program executable by the DOS command

Disk Basic 1.1 has a DOS command that loads track 34 at $2600 and if the contents start with the bytes 'O' and 'S', the interpreter jumps at $2602.

CMOC can accommodate this with its --dos option. It implies --coco, it uses $2600 as the code origin, and it generates the "OS" marker at that address.

The output is still a .bin file. CMOC comes with a command-line tool called install-coco-boot-loader that takes this file and copies it to track 34 of a DECB disk image. For example:

cmoc --dos prog.c
install-coco-boot-loader foo.dsk prog.bin

This tool will fail if track 34 is already occupied. It has a --force option that will install the boot leader despite that.

To uninstall the boot loader, use the --uninstall option:

install-coco-boot-loader --uninstall foo.dsk

C Preprocessor portability issues

For maximum portability, a CMOC program should not use the C Preprocessor's "stringification" operator (#), e.g.:

#define f(x) #x

When this feature is supported by cpp, the expression f(foobar) yields "foobar", including the double quotes.

When this feature is not supported, the pound sign is left as is, so CMOC will see #foobar, which will typically give a syntax error. This has been observed on the Macintosh (Darwin) as of May 2017.

Also to be avoided is the use of C++-style comments on a line to be processed by cpp, e.g.:

#define foo 42  // blargh

Some preprocessors may leave the comment instead of stripping it away, so the following should be used instead:

#define foo 42  /* blargh */


EXCEPT FOR THE FILES UNDER src/usim-0.91-cmoc in the CMOC source archive, CMOC is distributed under the following license:

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see