Switch branches/tags
Nothing to show
Find file
Fetching contributors…
Cannot retrieve contributors at this time
173 lines (132 sloc) 7.35 KB
Better String library Porting Guide
by Paul Hsieh
The bstring library is an attempt to provide improved string processing
functionality to the C and C++ language. At the heart of the bstring library
is the management of "bstring"s which are a significant improvement over '\0'
terminated char buffers. See the accompanying documenation file bstrlib.txt
for more information.
Identifying the Compiler
Bstrlib has been tested on the following compilers:
Microsoft Visual C++
Watcom C/C++ (32 bit flat)
Intel's C/C++ compiler (on Windows)
The GNU C/C++ compiler (on Windows/Linux on x86 and PPC64)
Borland C++
Turbo C
There are slight differences in these compilers which requires slight
differences in the implementation of Bstrlib. These are accomodated in the
same sources using #ifdef/#if defined() on compiler specific macros. To
port Bstrlib to a new compiler not listed above, it is recommended that the
same strategy be followed. If you are unaware of the compiler specific
identifying preprocessor macro for your compiler you might find it here:
Note that Intel C/C++ on Windows sets the Microsoft identifier: _MSC_VER.
16-bit vs. 32-bit vs. 64-bit Systems
Bstrlib has been architected to deal with strings of length between 0 and
INT_MAX (inclusive). Since the values of int are never higher than size_t
there will be no issue here. Note that on most 64-bit systems int is 32-bit.
Dependency on The C-Library
Bstrlib uses the functions memcpy, memmove, malloc, realloc, free and
vsnprintf. Many free standing C compiler implementations that have a mode in
which the C library is not available will typically not include these
functions which will make porting Bstrlib to it onerous. Bstrlib is not
designed for such bare bones compiler environments. This usually includes
compilers that target ROM environments.
Porting Issues
Bstrlib has been written completely in ANSI/ISO C and ISO C++, however, there
are still a few porting issues. These are described below.
1. The vsnprintf () function.
Unfortunately, the earlier ANSI/ISO C standards did not include this function.
If the compiler of interest does not support this function then the
BSTRLIB_NOVSNP should be defined via something like:
#if !defined (BSTRLIB_VSNP_OK) && !defined (BSTRLIB_NOVSNP)
# if defined (__TURBOC__) || defined (__COMPILERVENDORSPECIFICMACRO__)
# endif
which appears at the top of bstrlib.h. Note that the bformat(a) functions
will not be declared or implemented if the BSTRLIB_NOVSNP macro is set. If
the compiler has renamed vsnprintf() to some other named function, then
search for the definition of the exvsnprintf macro in bstrlib.c file and be
sure its defined appropriately:
# define exvsnprintf(r,b,n,f,a) {r=__compiler_specific_vsnprintf(b,n,f,a);}
# define exvsnprintf(r,b,n,f,a) {r=vsnprintf(b,n,f,a);}
Take notice of the return value being captured in the variable r. It is
assumed that r exceeds n if and only if the underlying vsnprintf function has
determined what the true maximal output length would be for output if the
buffer were large enough to hold it. Non-modern implementations must output a
lesser number (the macro can and should be modified to ensure this).
2. Weak C++ compiler.
C++ is a much more complicated language to implement than C. This has lead
to varying quality of compiler implementations. The weaknesses isolated in
the initial ports are inclusion of the Standard Template Library,
std::iostream and exception handling. By default it is assumed that the C++
compiler supports all of these things correctly. If your compiler does not
support one or more of these define the corresponding macro:
The compiler specific detected macro should be defined at the top of
bstrwrap.h in the Configuration defines section. Note that these disabling
macros can be overrided with the associated enabling macro if a subsequent
version of the compiler gains support. (For example, its possible to rig
up STLport to provide STL support for WATCOM C/C++, so -DBSTRLIB_CAN_USE_STL
can be passed in as a compiler option.)
3. The bsafe module, and reserved words.
The bsafe module is in gross violation of the ANSI/ISO C standard in the
sense that it redefines what could be implemented as reserved words on a
given compiler. The typical problem is that a compiler may inline some of the
functions and thus not be properly overridden by the definitions in the bsafe
module. It is also possible that a compiler may prohibit the redefinitions in
the bsafe module. Compiler specific action will be required to deal with
these situations.
Platform Specific Files
The makefiles for the examples are basically setup of for particular
environments for each platform. In general these makefiles are not portable
and should be constructed as necessary from scratch for each platform.
Testing a port
To test that a port compiles correctly do the following:
1. Build a sample project that includes the bstrlib, bstraux, bstrwrap, and
bsafe modules.
2. Compile bstest against the bstrlib module.
3. Run bstest and ensure that 0 errors are reported.
4. Compile test against the bstrlib and bstrwrap modules.
5. Run test and ensure that 0 errors are reported.
6. Compile each of the examples (except for the "re" example, which may be
complicated and is not a real test of bstrlib and except for the mfcbench
example which is Windows specific.)
7. Run each of the examples.
The builds must have 0 errors, and should have the absolute minimum number of
warnings (in most cases can be reduced to 0.) The result of execution should
be essentially identical on each platform.
Different CPU and compilers have different capabilities in terms of
performance. It is possible for Bstrlib to assume performance
characteristics that a platform doesn't have (since it was primarily
developed on just one platform). The goal of Bstrlib is to provide very good
performance on all platforms regardless of this but without resorting to
extreme measures (such as using assembly language, or non-portable intrinsics
or library extensions.)
There are two performance benchmarks that can be found in the example/
directory. They are: cbench.c and cppbench.cpp. These are variations and
expansions of a benchmark for another string library. They don't cover all
string functionality, but do include the most basic functions which will be
common in most string manipulation kernels.
In all cases, you may email issues found to the primary author of Bstrlib at
the email address: