clangml provides bindings for all versions of Clang, from 3.4 to
This library is a complete rewritting of the previous clangml
(clangml versions <4.0.0):
the bindings now rely on automatically generated C stubs to libclang, with
some extensions when libclang is incomplete.
Contrary to old clangml versions, the versions of clangml from 4.0.0 are
independent from the version of the Clang library:
any version of clangml from 4.0.0 can be built with any version of the
Clang library in the supported interval.
Currently, all versions of Clang, from 3.4 to 14.0.6, are supported.
However, clangml is statically linked to libclang, and clangml needs
to be rebuilt for every version of libclang to run with.
In addition, the low-level bindings are automatically generated
from libclang's header and their signature can change from one version
of libclang to another.
The high-level bindings (
Clang.Enum_constant) provide abstractions
that are essentially independent from libclang version.
These abstractions aim mainly to provide an algebraic datatype
representation of Clang abstract syntax tree (AST).
It is worth noticing that there can be some differences in the way clang
parses file from one version to another (in particular, some features of the
C/C++ languages are only supported by recent versions of clang,
see some examples in
Clang__ast module documentation).
clangml is installable via
opam. Since the library relies on external
dependencies, we suggest to use the depext plugin to install it together
with the packages needed for your system:
opam depext -i clangml
Manual installation requires a bootstrapped source directory.
Commits from branch
snapshot are bootstrapped: a new snapshot
is committed by continuous integration after every successful build from
To build clangml from snapshot or from a bootstrapped source directory,
you may either:
./configure && make && make install
(this method is recommended if you have to pass some options to configure);
opam pin add git+https://gitlab.inria.fr/memcad/clangml.git#snapshot.
To bootstrap the repository from a development branch (e.g.,
./bootstrap.sh first, then
./configure && make && make install as
configure relies on
llvm-config to find clang's library.
llvm-config is searched in
PATH, and you may
specify a path with
clangml requires some dependencies:
opam install dune refl.
Additionnally, to run
opam install ocamlcodoc pattern.
libclang and other external dependencies can be installed with opam depext
opam pin add -n git+https://gitlab.inria.fr/memcad/clangml.git#snapshot opam depext -i clangml
-n option asks
opam pin not to install clangml directly, and
opam depext to install clangml once dependencies are installed.)
Clang provides direct bindings to most of the symbols
defined by libclang. To match OCaml conventions, camel-case symbols
have been renamed to snake case (lower-case symbols with underscores), and
clang_ prefixes have been removed. Additional bindings have been defined in
libclang_extensions.h for some parts of clang's API that have
not been covered by libclang.
Clang.Ast provides a higher-level interface to clang's AST.
Clang.Ast.parse_file returns the AST from a file
Clang.Ast.parse_string returns the AST from a string.
You may try these functions in OCaml toplevel to discover the resulting data
Clang.Ast includes in particular the module
which declares the algebraic data types that represent the AST.
The documentation of most of the nodes contains examples that can be used as references
for how syntactic constructions are parsed, and that are extracted with
and serve as unit tests with
dune runtest (or, equivalently,
Moreover, the git branch
norms contains the AST corresponding to the examples
automatically extracted from C++14, C++17, and C++20 norms.
given an expression node
e : Clang.Expr.t, the type of
ecan be obtained by
t : Clang.Type.tis a
typedef, the underlying type declared for
tcan be obtained by
t : Clang.Type.tis a record (
union), the list of fields can by
C/C++ attributes are defined in a separate (auto-generated) module
mirror their non-lazy counterparts, by replacing eagerly constructed
desc fields by
lazy values, that are computed on demand. This is
useful to explore large ASTs efficiently (note that Clang parsing
itself can still be slow; the lazy part only concerns the conversion
Generating a new seed
clang__bindings.mli, are generated for each version of LLVM by the
stubgen tool (sub-directory
To generate these files for a given version of LLVM, you may run:
stubgen --llvm-config=$PATH_TO_LLVM_CONFIG $TARGET_PATH
stubgen depends on
generated by the
generate_attrs tool (sub-directory
To generate these files, you may run
generate_attrs --llvm-config=$PATH_TO_LLVM_CONFIG $PATH_TO_BOOTSTRAP_DIR.
The tool enumerates all the attributes supported by the given version of Clang,
and uses the bootstrap directory both as target path and to determine in which
version of Clang each attribute has been introduced.
with-doc & >= "1.5.1"
with-test & >= "0.2.0"
with-test & >= "1.0.1"
build & >= "1.8.0"