11 Frequently Asked Questions

11.1 Questions about installation

• I downloaded Blitz++, but when I try to gunzip it, I get “invalid compressed data–crc error”

You forgot to set binary download mode in ftp. Do so with the “binary” command.

• After accessing blitz from the cvs repository, the “autoreconf -fiv” command seems to fail mysteriously.

The current blitz autoconf build system uses libtool to manage building either static or shared versions of the library. Mac OS X systems have their own version of libtool and libtoolize that conflict with the GNU versions, so these have been renamed glibtool and glibtoolize. You must set environment variables to indicate the right tools to use. Try this:

export LIBTOOL=/usr/bin/glibtool
export LIBTOOLIZE=/usr/bin/glibtoolize

• The blitz configure script fails to find a valid Fortran compiler. Why the heck do I need a Fortran compiler for Blitz++ anyway?

A Fortran compiler is only needed to compile the Fortran portions of the blitz benchmark codes in the banchmarks subdirectory, which compare the speed of blitz code to that of raw Fortran 77 or Fortran 90 arrays. Many Darwin systems do not come with a Fortran compiler installed by default, and it can be difficult to obtain a GNU Fortran compiler for Mac OS X that is completely compatible with the default C/C++ compiler. Therefore, the blitz configure script now provides the option –disable-fortran, which will skip over Fortran configuration. This will render the benchmark codes unusable, but will allow you to build and install the blitz library.

• The linker complains about undefined references when compiling my blitz application code.

Although almost all of blitz consists of inlined templated code provided in header files via <blitz/array.h>, there are a few static global objects whose definitions are provided in a compiled blitz library. So always remember to include the appropriate -L flag and -lblitz on your linker command line in order to link your code against the blitz library.

• The compiler complains that there is no match for the TinyVector unary or binary math operator I have invoked, even though I’ve included <blitz/tinyvec.h>.

In versions prior to blitz 0.8, the tinyvec.h header automatically included all of the blitz support for expressions involving TinyVectors. Because this code is intimately linked with the expression template support for Vector, VectorPick and other Vector-like classes, this turns out to be a large amount of code. The blitz Array class uses TinyVector to represent Array shapes, and thus array.h must include tinyvec.h. This was creating a large amount of compile-time overhead, so it was decided to separate the TinyVector expression template support and put this in a new header file tinyvec-et.h. Therefore, your code must include <blitz/tinyvec-et.h> explicitly if it uses expressions with TinyVectors.

• The compiler complains that there is no Array class, even though I’ve included <blitz.h>.

You need to have the line:

using namespace blitz;

after including <blitz.h>.

• I can’t use gcc on my elderly PC because it requires 45–150Mb to compile with Blitz++

Unfortunately this is true. If this problem is ever fixed, it will be by the gcc developers, so my best suggestion is to post a bug report to the gcc-bugs list.

• I am using gcc under Solaris, and I get errors about “relocation against external symbol”

This problem can be fixed by installing the gnu linker and binutils. Peter Nordlund found that by using gnu-binutils-2.9.1, this problem disappeared. You can read a detailed discussion at http://oonumerics.org/blitz/support/blitz-support/archive/0029.html.

• I am using gcc under Solaris, and the assembler gives me an error that a symbol is too long.

This problem can also be fixed by installing the gnu linker and binutils. See the above question.

• DECcxx reports problems about “templates with C linkage”

This problem was caused by a problem in some versions of DECcxx’s math.h header: XOPEN_SOURCE_EXTENDED was causing an extern "C" { ... } section to have no closing brace. There is a kludge which is included in recent versions of Blitz++.

• On some platforms (especially SGI) the testsuite program minsumpow fails with the error: Template instantiation resulted in an unexpected function type of...

This is a known bug in the older versions of the EDG front end, which many C++ compilers use. There is no known fix. Most of Blitz++ will work, but you won’t be able to use some array reductions.

11.2 Questions about Blitz++ functionality

• For my problem, I need SVD, FFTs, QMRES, PLU, QR, ....

Blitz++ does not currently provide any of these. However, there are numerous C++ and C packages out there which do, and it is easy to move data back and forth between Blitz++ and other libraries. See these terms in the index: creating an array from pre-existing data, data(), stride(), extent(), fortranArray. For a list of other numerical C++ libraries, see the Object Oriented Numerics Page at http://oonumerics.org/oon/.

• Can Blitz++ be interfaced with Python?

Phil Austin has done so successfully. See a description of his setup in http://oonumerics.org/blitz/support/blitz-support/archive/0053.html.

Also see Harry Zuzan’s Python/Blitz image processing example code at http://www.stat.duke.edu/~hz/blitz_py/index.html.

• If I try to allocate an array which is too big, my program just crashes or goes into an infinite loop. Is there some way I can handle this more elegantly?

Blitz++ uses new to allocate memory for arrays. In theory, your compiler should be throwing a bad_alloc exception when you run out of memory. If it does, you can use a try/catch block to handle the out of memory exception. If your compiler does not throw bad_alloc, you can install your own new handler to handle out of memory.

Here is an excerpt from the ISO/ANSI C++ standard which describes the behaviour of new:

• Executes a loop: Within the loop, the function first attempts to allocate the requested storage. Whether the attempt involves a call to the Standard C library function malloc is unspecified.
• Returns a pointer to the allocated storage if the attempt is successful. Otherwise, if the last argument to set_new_handler() was a null pointer, throw bad_alloc.
• Otherwise, the function calls the current new_handler (lib.new.handler). If the called function returns, the loop repeats.
• The loop terminates when an attempt to allocate the requested storage is successful or when a called new_handler function does not return.

You can use set_new_handler to create a new handler which will issue an error message or throw an exception. For example:

void my_new_handler()
{
cerr << "Out of memory" << endl;
cerr.flush();
abort();
}

...

// First line in main():
set_new_handler(my_new_handler);

• When I pass arrays by value, the function which receives them can modify the array data. Why?

It’s a result of reference-counting. You have to think of array objects as being “handles” to underlying arrays. The function doesn’t receive a copy of the array data, but rather a copy of the handle. The alternative would be to copy the array data when passing by value, which would be grossly inefficient.

• Why can’t I use e.g. A >> 3 to do bitshifting on arrays?

The operators << and >> are used for input/ouput of arrays. It would cause problems with the expression templates implementation to also use them for bitshifting. However, it is easy enough to define your own bitshifting function – see User et.

• When I write TinyMatrix * TinyVector I get an error.

Try product(d2,d1). This works for matrix-matrix and matrix-vector products.