boringssl/INCORPORATING.md

Incorporating BoringSSL into a project

Note: if your target project is not a Google project then first read the main README about the purpose of BoringSSL.

Directory layout

Typically projects create a third_party/boringssl directory to put BoringSSL-specific files into. The source code of BoringSSL itself goes into third_party/boringssl/src, either by copying or as a submodule.

It's generally a mistake to put BoringSSL's source code into third_party/boringssl directly because pre-built files and custom build files need to go somewhere and merging these with the BoringSSL source code makes updating things more complex.

Build support

BoringSSL is designed to work with many different build systems. Currently, different projects use GYP, GN, Bazel and Make to build BoringSSL, without too much pain.

The development build system is CMake and the CMake build knows how to automatically generate the intermediate files that BoringSSL needs. However, outside of the CMake environment, these intermediates are generated once and checked into the incorporating project's source repository. This avoids incorporating projects needing to support Perl and Go in their build systems.

The script util/generate_build_files.py expects to be run from the third_party/boringssl directory and to find the BoringSSL source code in src/. You should pass it a single argument: the name of the build system that you're using. If you don't use any of the supported build systems then you should augment generate_build_files.py with support for it.

The script will pregenerate the intermediate files (see BUILDING.md for details about which tools will need to be installed) and output helper files for that build system. It doesn't generate a complete build script, just file and test lists, which change often. For example, see the file and test lists generated for GN in Chromium.

Defines

BoringSSL does not present a lot of configurability in order to reduce the number of configurations that need to be tested. But there are a couple of #defines that you may wish to set:

OPENSSL_NO_ASM prevents the use of assembly code (although it's up to you to ensure that the build system doesn't link it in if you wish to reduce binary size). This will have a significant performance impact but can be useful if you wish to use tools like AddressSanitizer that interact poorly with assembly code.

OPENSSL_SMALL removes some code that is especially large at some performance cost.

Symbols

You cannot link multiple versions of BoringSSL or OpenSSL into a single binary without dealing with symbol conflicts. If you are statically linking multiple versions together, there's not a lot that can be done because C doesn't have a module system.

If you are using multiple versions in a single binary, in different shared objects, ensure you build BoringSSL with -fvisibility=hidden and do not export any of BoringSSL's symbols. This will prevent any collisions with other verisons that may be included in other shared objects. Note that this requires that all callers of BoringSSL APIs live in the same shared object as BoringSSL.

If you require that BoringSSL APIs be used across shared object boundaries, continue to build with -fvisibility=hidden but define BORINGSSL_SHARED_LIBRARY in both BoringSSL and consumers. BoringSSL's own source files (but not consumers' source files) must also build with BORINGSSL_IMPLEMENTATION defined. This will export BoringSSL's public symbols in the resulting shared object while hiding private symbols. However note that, as with a static link, this precludes dynamically linking with another version of BoringSSL or OpenSSL.