[Home]
[Search]
[CTG]
[RTL]
[UGR]
Switching to Digital Mars C++
Digital Mars C++ offers numerous advantages over other compilers,
both in programmer productivity and in the speed and robustness of
generated code. Switching to Digital Mars C++ is well worth the effort.
This chapter outlines general considerations for converting existing code.
What's in This Chapter
Portable Programming Practices
Some of the programming practices that help make code portable
and reusable also help make it compatible with Digital Mars C++. Your
code will be easier to convert if you:
- Use ANSI standard library functions wherever possible
and avoid compiler-specific extensions.
- Do not rely on the location, alignment, or size of objects
in memory. In addition to the inevitable implementation
dependencies of memory models, the Digital Mars C++
compiler assigns different widths to some of the standard
data types.
- Do not rely on objects of the same size being treated as if
they had the same type. Digital Mars C++ enforces stricter
type checking than either Microsoft C++ or Borland C++,
and will generate warnings or errors in these instances. In
addition, loosely typed code will often fail under 32-bit
compilation. For example, Digital Mars C++ defines the
WORD type as an unsigned short, and the UINT type
as an unsigned int. If these types are treated
identically in your code, you can use the -Jm compiler
option to relax type checking.
- Use manifest constants wherever possible, instead of
relying on explicit (and possibly compiler or environment
dependent) values.
- Write function prototypes for all your user-defined
functions. See the Chapter 3, "Digital Mars C++ Language
Implementation" for more information.
- Can avoid linking files compiled with different compilers.
Before you begin conversion, delete all .obj and .lib
files created with your old compiler for which you have
the source code. Also, be sure that the INCLUDE and LIB
environment variables reference the Digital Mars C++
directories so you do not accidentally compile with one
vendor's headers and link with the other's libraries.
Object Level Compatibility
Digital Mars C++ object modules are "broadly compatible" with
Microsoft and Borland objects. However, there are some differences:
- The Digital Mars C++ compiler's "helper" functions (long
multiply, float, and so on) are different than their Borland
or Microsoft counterparts. If you do not want to recode
calls to these functions, remove the other vendor's helper
functions from their libraries and include them in your
converted code.
- Digital Mars' floating point libraries are reentrant; our
object code is therefore different than Microsoft's or
Borland's for functions with C linkage. You can work
around this problem by rewriting affected functions to
pass a pointer to their return value as a parameter, or by
using C++, FORTRAN, or Pascal linkage instead of C
linkage.
- The layout of the struct_iob in stdio. h is different
for Digital Mars C++. Therefore, you cannot link buffered
I/ O functions compiled with Digital Mars' stdio. h with
buffered I/ O functions compiled with a different
stdio. h. To avoid problems, compile all modules in a
program with the same version of stdio. h.
- By default, Digital Mars C++ aligns structure members on
16-bit boundaries in 16-bit compilations; Microsoft and
Borland C++ do not. For binary compatibility, compile
with the -a option to suppress structure member
alignment.
Using Third-Party Libraries
If you must use third-party libraries written for Microsoft or Borland
C++ with the Digital Mars C++ compiler, there is no simple way to
determine which features are compatible. Your code could even link
correctly and still contain obscure errors.
If you have the source to the third-party library, try recompiling it
with Digital Mars C++. A better solution is to obtain the Digital Mars
version of the library from the vendor.
Recompiling for Digital Mars C++ In many cases, converting programs written for another compiler to
Digital Mars C++ can be as easy as recompiling them. Here are some
steps you can take to ease recompilation:
- Compile without the global optimizer until your program
runs without errors. A working program can occasionally
fail when optimized. For more information, see Chapter
25, "Optimizing Code."
- In Digital Mars C++, character data is signed by default. If
your code depends on char data being unsigned,
compile with either the -J or -Ju options. -J treats
chars as unsigned, not sign extended. -Ju treats chars as
unsigned, sign extended.
- If your code does not follow strict ANSI type checking
rules, use the -Jm option, which relaxes type checking.
Compile time warnings
By default, the Digital Mars C++ compiler generates warnings in
response to more conditions than do most other compilers. If you
find these unfamiliar warnings annoying, compile with the -w
(warning level) option. Each instance of -w on the SC command line
turns off a specific warning. For example, the command:
sc -w2 -w6 myfile.cpp
turns off warnings number 2 and 6. For information on warnings and
their numbers, see Chapter 2, "Compiling Code."
Converting from Microsoft
This chapter describes differences you should consider when
converting code written for Microsoft Visual C++ Version 1.5
to Digital Mars C++.
What's in This Chapter
This chapter discusses compatibility issues between Microsoft Visual
C++ and Digital Mars C++ with respect to these issues:
- Keywords
- Predefined macros
- Header files
- Libraries
- MFC
- Memory model support
- Assembly language interface
- Compiler options
Keywords
Digital Mars C++ supports most of Microsoft C's nonstandard
keywords. The following table lists Microsoft's keyword extensions
and their Digital Mars counterparts:
Table 23-1 Support for Microsoft keywords
Unsupported Microsoft keywords
__fastcall
__fortran
__saveregs
__segment
__segname
__self
Predefined Macros
Digital Mars C++ supports most of Microsoft C's nonstandard macros.
The following table lists Microsoft's extended macros and their
Digital Mars counterparts:
Table 23-2 Support for Microsoft C nonstandard macros
Unsupported Microsoft macros Digital Mars counterparts
_CHAR_UNSIGNED _CHAR_UNSIGNED
_FAST No support
_MSC_VER __DMC__ (works the same, but
version numbers are for SC)
_MSDOS
MSDOS Recommended use is as a
command line argument
_PCODE No support
_QC No support
M_I286 M_I286
M_I386 M_I386
M_I8086 M_I8086
M_I86 M_I86
M_I86CM M_I86CM
M_I86HM No support
M_I86LM M_I86LM
M_I86MM M_I86MM
M_I86SM M_I86SM
M_I86TM M_I86TM
Header Files
The Digital Mars C++ run-time library includes all the Microsoft
nonstandard headers. Many of these headers simply reference the
Digital Mars header file that contains equivalent functions. Digital Mars
C++ issues a warning whenever your code makes use of such a
header file. In general, we recommend switching to ANSI-compliant
headers wherever possible.
Library Functions
Digital Mars C++ supports the great majority of Microsoft Visual C++
Version 1.5 library functions. The table below lists the only Microsoft
functions that the Digital Mars C++ run-time library does not support:
Table 23-3 Unsupported Microsoft library functions
Unsupported Microsoft Comments
functions header
Graphics functions graph.h Digital Mars C++ does not
support the DOS graphics
system.
Charting functions pgchart.h
Virtual memory vmemory.h Digital Mars C++ no longer
functions supports Virtual Code
Management (VCM).
QuickWin functions io.h
_bexpand malloc.h Use farrealloc instead.
_bfree malloc.h
_bfreeseg malloc.h
_bheapadd malloc.h Not needed with Digital Mars
C++ heap management.
_bheapchk malloc.h
_bheapmin malloc.h
_bheapseg malloc.h
_bheapset malloc.h
_bheapwalk malloc.h
_bmalloc malloc.h
_bmsize malloc.h
_brealloc malloc.h
_dieeetomsbin math.h
_dmsbintoieee math.h
_expand malloc.h Use realloc instead.
_fexpand malloc.h Use farrealloc instead.
_fheapchk malloc.h
_fheapmin malloc.h
_fheapset malloc.h
_fheapwalk malloc.h
_fieeetomsbin math.h
_fmsbintoieee math.h
_heapadd malloc.h Not needed with Digital Mars
C++ heap management.
_heapchk malloc.h
_heapmin malloc.h
_heapset malloc.h
_heapwalk malloc.h
_j0, _j1, _jnl math.h
_j0l, _j1l, _jn math.h
_nexpand malloc.h Use realloc instead.
_nheapadd malloc.h Not needed with Digital Mars
C++ heap management.
_nheapchk malloc.h
_nheapmin malloc.h
_nheapset malloc.h
_nheapwalk malloc.h
_setbnew_handler new.h
_sethnew_handler new.h
_y0, _y1, _ynl math.h
_y0l, _y1l, _yn math.h
Microsoft Foundation Classes
Digital Mars C++ for Windows and DOS includes the Microsoft
Foundation Classes Version 2.5, for the Large memory model only.
Digital Mars C++ for Win32 includes 32-bit MFC 3.0.
Digital Mars' versions of the Microsoft Foundation Class Library are
licenced from Microsoft Corporation.
When compiling code that uses the Microsoft
Foundation Classes, use the -gf option (see
Chapter 2, "Compiling Code").
Memory Models
The Digital Mars C++ implementation of those 16-bit memory models
also supported by Microsoft C (the Small, Medium, Compact, and
Large) is similar, but not exactly the same. The significant differences
are:
- In the Digital Mars Compact and Large models, unlike their
Microsoft counterparts, the stack segment and data
segment are not the same. Near pointers in these models
are relative to the data segment and cannot be used to
access automatic or parameter variables. Check your
assembly language source code to ensure that SS is not
assumed to be equal to DS.
- Microsoft C only creates a far data segment if the amount
of data exceeds 32,767 bytes, or exceeds the value
specified via the /Gt option to CL (Digital Mars C++ has a
corresponding compiler option, -GT). While data is
always assigned to the near data segment by default,
functions are near by default in the Compact model and
far by default in the Large model in Digital Mars C++.
- Digital Mars' malloc() functions take an unsigned int
as an argument for all 16-bit memory models and take an
unsigned long for 32-bit models. Therefore, the 16-bit
models limit the size of a data object to 64KB blocks. To
allocate a block greater than 64KB, call farmalloc,
which takes an unsigned long.
- At program startup, a Microsoft C program's heap
management run-time allocates _amblksiz bytes of far
heap space. Additional heap is allocated in _amblksiz
blocks as needed. In contrast, Digital Mars C++ makes a call
to DOS for each request for far heap space.
- Unlike Microsoft C, Digital Mars C++ does not allocate heap
blocks of zero size if a program requests them. NULL is
returned instead.
- Unlike Microsoft C, Digital Mars C++ cleans up free heap
space when a program releases a block of memory.
- Digital Mars C++ does not support the Microsoft C
_heapadd, _bheapadd, and _nheapadd functions
because its heap manager calls DOS each time a program
requests far heap space.
- Digital Mars C++ does not support the Microsoft _expand,
_bexpand, _fexpand, and _nexpand routines. Use
realloc in place of _expand or _nexpand calls (or
wherever possible), and farrealloc in place of
_bexpand or _fexpand. However, be aware that
realloc and farrealloc, unlike _expand, can move
the block to a new location to satisfy the reallocation
request, in which case they will return the new location of
the reallocated block.
- Digital Mars C++ does not support these additional non-standard
Microsoft memory management functions:
_heapchk, _bheapchk, _fheapchk, _nheapchk,
_heapwalk, _bheapwalk, _fheapwalk,
_nheapwalk, _heapset, _bheapset, _fheapset,
_nheapset, _heapmin, _bheapmin, _fheapmin,
_nheapmin.
Huge memory model support
Digital Mars C++ does not support the Huge memory model. That is, a
pointer whose type is unspecified cannot be made huge by default.
However, the compiler supports huge pointers and huge data (via
the __huge modifier).
Digital Mars C++ also supports the Microsoft halloc and hfree
functions.
Support for based pointers
Digital Mars C++ does not support a based heap. No support for based
pointers is available except for:
__based(__segname("_CODE"));
__based(__segname("_DATA"));
__based(__segname("_STACK"));
If your code makes use of based pointers, convert
them to far pointers. Based pointers are a nonstandard construct and
the code generated using them is not significantly better than for far
pointers.
Assembly Language Interface
The Digital Mars C++ assembly language interface is very similar to
Microsoft's. The only differences are:
- DS is not equal to SS for the Compact and Large memory
models.
- For C functions (those declared as __cdecl or extern
"C"), double and float values are returned in the
registers, rather than in a static memory locaton. For C++,
FORTRAN, and Pascal functions, double and float
values are returned on the stack in a manner compatible
with Microsoft C.
- In Digital Mars C++, structs, floats, and doubles are
returned by allocating a temporary variable on the stack
and passing a hidden pointer to it. The called function
copies the return values into this variable and returns a
pointer to it.
Compiler Options
Digital Mars C++ provides a wide range of compiler command line
options, many of which perform operations similar or identical to a
Microsoft compiler option.
To convert your old Microsoft compiler command lines to the most
similar Digital Mars command lines possible, use the
CL utility. CL
automatically converts Microsoft command lines to SC command
lines, and then runs SC.
Converting from Borland
This chapter describes the differences you should consider when
converting code written for Borland C++ Version 4.0 to Digital Mars C++.
- Keywords
- Predefined macros
- Header files
- Libraries
- Memory model support
- Assembly language interface
- Compiler options
Keywords
Digital Mars C++ supports many of Borland C's nonstandard keywords.
The following table lists Borland's extended keywords and their
Digital Mars counterparts:
Table 24-1 Support for Borland nonstandard keywords
Unsupported Borland keywords Digital Mars counterparts
_cs __cs
_ds No equivalent
_es No equivalent
_saveregs No equivalent
_seg No equivalent
_ss __ss
Predefined Macros
Digital Mars C++ supports many of Borland C's nonstandard macros.
The following table lists Borland's extended macros and their
Digital Mars counterparts:
Table 24-2 Support for Borland C++ nonstandard macros
Unsupported Borland macros Digital Mars counterparts
__BORLANDC__ __DMC__ (works the same, but
version numbers are for SC)
__BCPLUSPLUS__ __DMC__ (works the same, but
version numbers are for SC)
__CDECL__ No equivalent
__COMPACT__ __COMPACT__
__DLL__ _WINDLL
__HUGE__ No equivalent
__LARGE__ __LARGE__
__MEDIUM__ __MEDIUM__
__MSDOS__ No equivalent
__PASCAL__ No equivalent
__OVERLAY__ No equivalent
__SMALL__ __SMALL__
__TEMPLATES__ No equivalent
__TINY__ __TINY__
__TCPLUSPLUS__ __DMC__ (works the same, but
version numbers are for SC
__TURBOC__ __DMC__ (works the same, but
version numbers are for SC)
_Windows _WINDOWS
Header Files
The Digital Mars C++ run-time library includes all the Borland non-standard
headers. Many of these headers simply reference the
Digital Mars header file that contains equivalent functions. Digital Mars
C++ issues a warning whenever code uses a header file. In general,
we recommend switching to ANSI-compliant headers wherever
possible.
Library Functions
Digital Mars C++ supports most Borland library functions. The table
below lists those Borland functions that the Digital Mars C++ run-time
library does not support:
Table 24-3 Unsupported Borland library functions
Unsupported Borland header Comments
functions
Graphics functions graphics.h
Text display functions conio.h
_dos_getvect dos.h
_dos_setvect dos.h
_harderr dos.h
_hardresume dos.h
_hardreturn dos.h
_OvrInitEms dos.h
_OvrInitExt dos.h
_setcursortype conio.h Use display package functions
cgets conio.h Use display package functions
cprintf conio.h Use display package functions
cputs conio.h Use display package functions
closedir dirent.h
cscanf conio.h
dostounix dos.h
farheapcheck alloc.h
farheapcheckfree alloc.h
farheapchecknode alloc.h
farheapfillfree alloc.h
farheapwalk alloc.h
getpass conio.h
getvect dos.h
harderr dos.h
hardresume dos.h
hardreturn dos.h
heapcheck alloc.h
heapcheckfree alloc.h
heapchecknode alloc.h
heapfillfree alloc.h
heapwalk alloc.h
import dos.h
importb dos.h
intr dos.h
opendir dirent.h
outport dos.h
outportb dos.h
readdir dirent.h
rewinddir dirent.h
setvect dos.h
unixtodos dos.h
Some Digital Mars C++ run-time library functions have
different types or numbers of arguments from their
Borland counterparts, return different values, or
support different defined constants. For information
on specific functions see the Run-Time Library
Reference.
Borland Graphics Interface
Digital Mars C++ does not support any of the functions, fonts, or
drivers in the Borland Graphics Interface.
Memory Models
The Digital Mars C++ implementations of those 16-bit memory models
also supported by Borland C++ (the Tiny, Small, Medium, Compact,
and Large models) are similar. Their significant differences include:
- In the Digital Mars Compact and Large models, unlike their
Borland counterparts, the stack and data segments are not
the same. Near pointers in these models are relative to the
data segment and cannot access automatic or parameter
variables. Check your assembly language source code to
ensure that SS is not assumed to be equal to DS.
- Digital Mars' malloc() functions take an unsigned int
as an argument for 16-bit memory models and take an
unsigned long for 32-bit models. Therefore, the 16-bit
models limit the size of a data object to 64KB blocks. To
allocate a block greater than 64KB, call farmalloc,
which takes an unsigned long.
- At program startup, Borland's memory-management
routine allocates all available DOS memory as one block
of far heap space. In contrast, Digital Mars C++ makes a call
to DOS for each request for far heap space.
- Digital Mars C++ does not support these additional non-standard
Borland memory-management functions:
heapcheck, farheapcheck, heapfillfree,
farheapfillfree, heapwalk, farheapwalk,
coreleft.
Huge memory model support
Digital Mars C++ does not support the Huge memory model. That is, a
pointer whose type is unspecified cannot be made huge by default.
However, the compiler supports huge pointers and huge data (via
the huge modifier).
Digital Mars C++ does not support the Borland halloc and hfree
functions. Convert calls to these functions to the corresponding
Digital Mars far calls (which have unsigned long argument types
and can, therefore, allocate huge blocks) or store the affected data in
a different structure (a huge array, for example).
Assembly Language Interface
The Digital Mars C++ assembly language interface is very similar to
Borland's. The only differences are:
- DS is not equal to SS for the Compact and Large memory
models.
- For C functions (those declared as __cdecl and extern
"C"), double and float values are returned in registers,
rather than in a static memory location. For C++,
FORTRAN, and Pascal functions, double and float
values are returned on the stack in a manner compatible
with Borland C.
- In Digital Mars C++, structs, floats, and doubles are
returned by allocating a temporary variable on the stack
and passing a hidden pointer to it. The called function
copies the return values into this variable and returns a
pointer to it.
- Digital Mars C++ does not support the _FLAGS register
variable.
- Digital Mars C++ does not store long double values as
10-byte quantities; they are the same as doubles (8 bytes).
Compiler Options
Digital Mars C++ provides a wide range of compiler command line
options, many of which perform operations similar or identical to a
Borland C++ compiler option.
To convert Borland compiler command lines to the most similar
Digital Mars command lines possible, use the
BCC utility. BCC
automatically converts Borland C++ command lines to SC command
lines and then runs SC.
Copyright © 1995-2001 Digital Mars. All Rights Reserved.