Building The Library¶
This document describes how to build Botan on Unix/POSIX and MS Windows systems. The POSIX oriented descriptions should apply to most common Unix systems (including MacOS X), along with POSIX-ish systems like BeOS, QNX, and Plan 9. Currently, systems other than Windows and POSIX (such as VMS, MacOS 9, OS/390, OS/400, ...) are not supported by the build system, primarily due to lack of access. Please contact the maintainer if you would like to build Botan on such a system.
Botan’s build is controlled by configure.py, which is a Python script. Python 2.5 or later is required.
For the impatient, this works for most systems:
$ ./configure.py [--prefix=/some/directory] $ make $ make check $ make install
Or using nmake, if you’re compiling on Windows with Visual C++. On platforms that do not understand the ‘#!’ convention for beginning script files, or that have Python installed in an unusual spot, you might need to prefix the configure.py command with python or /path/to/python:
$ python ./configure.py [arguments]
Configuring the Build¶
The first step is to run configure.py, which is a Python script that creates various directories, config files, and a Makefile for building everything. The script requires at least Python 2.5; any later version of Python 2.x should also work. Python 3.1 will also work but requires an extra step; see Configuring the Build With Python 3.1 for details.
The script will attempt to guess what kind of system you are trying to compile for (and will print messages telling you what it guessed). You can override this process by passing the options --cc, --os, and --cpu.
You can pass basically anything reasonable with --cpu: the script knows about a large number of different architectures, their sub-models, and common aliases for them. You should only select the 64-bit version of a CPU (such as “sparc64” or “mips64”) if your operating system knows how to handle 64-bit object code - a 32-bit kernel on a 64-bit CPU will generally not like 64-bit code.
By default the script tries to figure out what will work on your system, and use that. It will print a display at the end showing which algorithms have and have not been enabled. For instance on one system we might see lines like:
INFO: Skipping, by request only - bzip2 gnump openssl qt_mutex zlib INFO: Skipping, incompatible CPU - aes_intel aes_ssse3 asm_x86_64 mp_asm64 mp_x86_64 sha1_x86_64 INFO: Skipping, incompatible OS - beos_stats cryptoapi_rng win32_crit_section win32_stats INFO: Skipping, incompatible compiler - mp_msvc64 mp_x86_32_msvc
The ones that are skipped because they are ‘by request only’ have to be explicitly asked for, because they rely on third party libraries which your system might not have or that you might not want the resulting binary to depend on. For instance to enable zlib support, add --with-zlib to your invocation of configure.py.
You can control which algorithms and modules are built using the options --enable-modules=MODS and --disable-modules=MODS, for instance --enable-modules=zlib and --disable-modules=rc5,idea. Modules not listed on the command line will simply be loaded if needed or if configured to load by default. If you use --no-autoload, only the most core modules will be included; you can then explicitly enable things that you want to use with --enable-modules. This is useful for creating a minimal build targeting to a specific application, especially in conjunction with the amalgamation option; see The Amalgamation Build.
$ ./configure.py --no-autoload --enable-modules=rsa,ecdsa,eme1,emsa1,emsa4
will set up a build that only includes RSA, ECDSA, and some padding modes, along with their dependencies. A small subset of core features, including AES, SHA-2, HMAC, and the multiple precision integer library, are always loaded.
The script tries to guess what kind of makefile to generate, and it almost always guesses correctly (basically, Visual C++ uses NMAKE with Windows commands, and everything else uses Unix make with POSIX commands). Just in case, you can override it with --make-style=somestyle. The styles Botan currently knows about are ‘unix’ (normal Unix makefiles), and ‘nmake’, the make variant commonly used by Windows compilers. To add a new variant (eg, a build script for VMS), you will need to create a new template file in src/build-data/makefile.
The basic build procedure on Unix and Unix-like systems is:
$ ./configure.py [--enable-modules=<list>] [--cc=CC] $ make # You may need to set your LD_LIBRARY_PATH or equivalent for ./check to run $ make check # optional, but a good idea $ make install
On Unix systems the script will default to using GCC; use --cc if you want something else. For instance use --cc=icc for Intel C++ and --cc=clang for Clang.
The make install target has a default directory in which it will install Botan (typically /usr/local). You can override this by using the --prefix argument to configure.py, like so:
./configure.py --prefix=/opt <other arguments>
On some systems shared libraries might not be immediately visible to the runtime linker. For example, on Linux you may have to edit /etc/ld.so.conf and run ldconfig (as root) in order for new shared libraries to be picked up by the linker. An alternative is to set your LD_LIBRARY_PATH shell variable to include the directory that the Botan libraries were installed into.
On Mac OS X¶
In general the Unix instructions above should apply, however OS X does not support LD_LIBRARY_PATH. Thomas Keller suggests instead running install_name_tool between building and running the self-test program:
$ VERSION=1.10.0 # or whatever the current version is $ install_name_tool -change $(otool -X -D libbotan-$VERSION.dylib) \ $PWD/libbotan-$VERSION.dylib check
Building Universal Binaries¶
To build a universal binary for OS X, some simple modifications are required. First, create a CPU type for the target universalbinary, one that doesn’t specify any special handlers. This is done by creating an empty file in src/build-data/arch:
$ touch src/build-data/arch/universalbinary.txt
and then adding a special target for it in gcc, by adding the line:
universalbinary -> "-force_cpusubtype_ALL -mmacosx-version-min=10.4 -arch i386 -arch ppc"
in the section marked <mach_abi_linking> in src/build-data/cc/gcc.txt. Then configure with:
$ ./configure.py --cpu=universalbinary [other options here]
On MS Windows¶
If you don’t want to deal with building botan on Windows, check the website; commonly prebuilt Windows binaries with installers are available, especially for stable versions.
You need to have a copy of Python installed, and have both Python and your chosen compiler in your path. Open a command shell (or the SDK shell), and run:
> python configure.py --cc=msvc (or --cc=gcc for MinGW) [--cpu=CPU] > nmake > nmake check # optional, but recommended > nmake install
For Win95 pre OSR2, the cryptoapi_rng module will not work, because CryptoAPI didn’t exist. And all versions of NT4 lack the ToolHelp32 interface, which is how win32_stats does its slow polls, so a version of the library built with that module will not load under NT4. Later versions of Windows support both methods, so this shouldn’t be much of an issue anymore.
By default the install target will be C:\botan; you can modify this with the --prefix option.
When building your applications, all you have to do is tell the compiler to look for both include files and library files in C:\botan, and it will find both. Or you can move them to a place where they will be in the default compiler search paths (consult your documentation and/or local expert for details).
For iOS using XCode¶
To cross compile for iOS, configure with:
$ ./configure.py --cpu=armv7 --cc=clang --cc-bin="clang++ -arch armv7 -arch armv7s --sysroot=$(IOS_SYSROOT)"
Along with any additional configuration arguments. Using --no-autoload might be helpful as can substantially reduce code size.
Edit the makefile and change AR (around line 30) to:
AR = libtool -static -o
You may also want to edit LIB_OPT to use -Os to optimize for size.
Now build as normal with make check. Confirm the binary is compiled for both architectures with:
$ xcrun -sdk iphoneos lipo -info check Architectures in the fat file: check are: armv7 armv7s
Now sign the test application with:
$ codesign -fs "Your Name" check
which should allow you to run the library self tests on a jailbroken device.
It has been reported several times that the library can be built for Android using the NDK, but precise instructions of what was required have not been provided. If you successfully build the library for Android, please report the exact sequence of steps needed so this documentation can be updated.
Botan usually links in several different system libraries (such as librt and libz), depending on which modules are configured at compile time. In many environments, particularly ones using static libraries, an application has to link against the same libraries as Botan for the linking step to succeed. But how does it figure out what libraries it is linked against?
The answer is to ask the botan-config script. This basically solves the same problem all the other *-config scripts solve, and in basically the same manner.
There are 4 options:
--prefix[=DIR]: If no argument, print the prefix where Botan is installed (such as /opt or /usr/local). If an argument is specified, other options given with the same command will execute as if Botan as actually installed at DIR and not where it really is; or at least where botan-config thinks it really is. I should mention that it
--version: Print the Botan version number.
--cflags: Print options that should be passed to the compiler whenever a C++ file is compiled. Typically this is used for setting include paths.
--libs: Print options for which libraries to link to (this includes -lbotan).
Your Makefile can run botan-config and get the options necessary for getting your application to compile and link, regardless of whatever crazy libraries Botan might be linked against.
Botan also by default installs a file for pkg-config, namespaced by the major and minor versions. So it can be used, for instance, as:
$ pkg-config botan-1.10 --modversion 1.10.0 $ pkg-config botan-1.10 --cflags -I/usr/local/include $ pkg-config botan-1.10 --libs -L/usr/local/lib -lbotan -lm -lbz2 -lpthread -lrt
No special help exists for building applications on Windows. However, given that typically Windows software is distributed as binaries, this is less of a problem - only the developer needs to worry about it. As long as they can remember where they installed Botan, they just have to set the appropriate flags in their Makefile/project file.
Building the Python wrappers¶
The Python wrappers for Botan use Boost.Python, so you must have Boost installed. To build the wrappers, pass the flag --with-boost-python to configure.py. This will create a second makefile, Makefile.python, with instructions for building the Python module. After building the library, execute:
$ make -f Makefile.python
to build the module. Currently only Unix systems are supported, and the Makefile assumes that the version of Python you want to build against is the same one you used to run configure.py.
To install the module, use the install target.
See Python Bindings for more information about the binding.
Building the Perl XS wrappers¶
To build the Perl XS wrappers, change your directory to src/wrap/perl-xs and run perl Makefile.PL, then run make to build the module and make test to run the test suite:
$ perl Makefile.PL Checking if your kit is complete... Looks good Writing Makefile for Botan $ make cp Botan.pm blib/lib/Botan.pm AutoSplitting blib/lib/Botan.pm (blib/lib/auto/Botan) /usr/bin/perl5.8.8 /usr/lib64/perl5/5.8.8/ExtUtils/xsubpp [...] g++ -c -Wno-write-strings -fexceptions -g [...] Running Mkbootstrap for Botan () chmod 644 Botan.bs rm -f blib/arch/auto/Botan/Botan.so g++ -shared Botan.o -o blib/arch/auto/Botan/Botan.so \ -lbotan -lbz2 -lpthread -lrt -lz \ chmod 755 blib/arch/auto/Botan/Botan.so cp Botan.bs blib/arch/auto/Botan/Botan.bs chmod 644 blib/arch/auto/Botan/Botan.bs Manifying blib/man3/Botan.3pm $ make test PERL_DL_NONLAZY=1 /usr/bin/perl5.8.8 [...] t/base64......ok t/filt........ok t/hex.........ok t/oid.........ok t/pipe........ok t/x509cert....ok All tests successful. Files=6, Tests=83, 0 wallclock secs ( 0.08 cusr + 0.02 csys = 0.10 CPU)