Libadalang

Libadalang is a project to build a high performance semantic engine for the Ada programming language. It is meant to provide a basis to write Ada tooling, including tools working on potentially changing and incorrect code, such as IDEs. Its goals encompass, but are not limited to:

• Full support for parsing the Ada 2012 syntax, plus SPARK extensions.
• Error tolerant parsing: the parser must be able to recover from simple errors and provide a "best-guess" tree.
• Error tolerant semantic analysis: it must be possible to create a tool that works only on syntax, and completely ignores semantic issues.
• Full symbol resolution respecting Ada 2012 – and prior Ada versions – semantics.
• Bindings to a variety of languages, including Ada, C, Python and Java, so that tools can be written from various ecosystems.
• Incremental processing of source files.

If you have problems building or using Libadalang, or want to suggest enhancements, please open a GitHub issue. We also gladly accept pull requests!

High level architecture

Libadalang is a library that can be used from Ada and Python. Most of its code is Ada code, generated from the language specification that you can find in ada/language.

WARNING: You will not find the generated code in the repository. You have to generate it yourself. We're thinking about some plans to auto-generate the code and put it in another repo/branch.

It is using the Langkit framework as a basis, and is at the time of writing the main project developped using it.

The language specification, while embedded in Python syntax, is mostly its own language, the Langkit DSL, that is used to specify the part of Ada syntax and semantics that are of interest to us.

Status of the project

Libadalang is still in development. its APIs are not stable, the shape of the abstract syntax tree is not yet completelely stable, and most of its features are either not stable or not fully implemented.

It is not yet safe to rely on the API stability of Libadalang in your projects. However, Libadalang is used internally in some AdaCore projects, so you might find it as a project dependency.

Libadalang currently:

• Is able to parse 100% of Ada 2012 syntax, and presents a well formed tree for it.

• Is able to recover some syntax errors, but is still currently behind GNAT in that regard.

• Provides some name resolution/navigation. Name resolution is the item on which most work is focused as of now. A sizable part of the language is handled, but it is not yet complete.

• Is able to handle some very simple incremental processing. Reparsing a source A and querying xref on a source B that depends on A is theoretically supported, and works in some cases, but the infrastructure is not yet general enough.

For those reasons, Libadalang is only suited today for mostly syntactic tools.

Libadalang and ASIS

ASIS is widely used for static analysis of Ada code, and is an ISO standard. It is still the go-to tool if you want to create a tool that analyses Ada code. Also, as explained above, Libadalang is not mature yet, and cannot replace ASIS in tools that require semantic analysis.

However, there are a few reasons you might eventually choose to use Libadalang instead of ASIS:

1. The ASIS standard has not yet been updated to the 2012 version of Ada. More generally, the advantages derived from ASIS being a standard also means that it will evolve very slowly.

2. Syntax only tools will derive a lot of advantages on being based on Libadalang:

   • Libadalang will be completely tolerant to semantic errors. For example, a pretty-printer based on Libadalang will work whether your code is semantically correct or not, as long as it is syntactically correct.

   • Provided you only need syntax, Libadalang will be much faster than ASIS' main implementation (AdaCore's ASIS), because ASIS always does complete analysis of the input Ada code.

3. The design of Libadalang's semantic analysis is lazy. It will only process semantic information on-demand, for specific portions of the code. It means that you can get up-to-date information for a correct portion of the code even if the file contains semantic errors.

4. Libadalang has bindings to C and Python, and its design makes it easy to bind to new languages.

5. Libadalang is suitable to write tools that work on code that is evolving dynamically. It can process code and changes to code incrementally. Thus, it is suitable as an engine for an IDE, unlike AdaCore's ASIS implementation.

6. Libadalang is not tied to a particular compiler version. This combined with its staged and error tolerant design means that you can use it to detect bugs in Ada compilers/tools.

Quick guide to use Libadalang

In order to use Libadalang, one has first to generate its code and to build it. You can read and run manually the steps in the "Setup" section below, or you can use our script to semi-automate this (please read and update this script to adapt it to your setup before running it). After this, you can either use Libadalang in Ada with the libadalang.gpr project file, or in Python just import the libadalang module.

Setup

To generate and build the library itself, you'll need to go through the following steps:

• Install the GNAT tools and compiler. You can find Community Editions on https://www.adacore.com/download

• Build and install the GNATcoll library (core, plus Iconv and GMP bindings). You can find its source release on https://www.adacore.com/download or directly on GitHub: https://github.com/AdaCore/gnatcoll-core and https://github.com/AdaCore/gnatcoll-bindings (just make sure you checkout the gpl-20** branch corresponding to your GNAT Community release).

• Install Quex version 0.65.4 - https://sourceforge.net/projects/quex/files/HISTORY/0.65/quex-0.65.4.tar.gz/download Follow the installation guide in the Quex README

• Install every Python dependency. We recommend creating a virtualenv and installing them inside of it, this way:

    $ virtualenv env
    $ source env/bin/activate
    $ pip install -r REQUIREMENTS.dev
  

To develop comfortably:

• If you want interactive debugging when code is generated, install IPython
• If you want to compute code coverage for the code generator, install coverage.py (see REQUIREMENTS.dev)
• If you want to check memory issues, the testsuite has an option to track them using Valgrind.

Building

First, let's generate code for Libadalang itself. In the top-level directory, run:

$ python ada/manage.py generate

You can also pass a -c flag in order to get a code coverage report in the "coverage" directory.

Then, let's build it:

$ python ada/manage.py build

If you satisfied all the above dependencies and if you did set QUEX_PATH and the quex executable as said in quex's README, it should build fine.

Also, while developing Libadalang you might be happy to use the

$ python ada/manage.py make

command, that will wrap the two previous commands in one, generating the code and building it in one step

Install

Once you built Libadalang, you can install the library in any place you want:

$ python ada/manage.py install $INSTALL_DIR

Testing

First, make sure you have the build/bin directory in your PATH so the test cases can access the parse program. Then, in the top-level directory, run:

$ python testsuite/testsuite.py

If you want to learn more about this test driver's options (for instance to run tests under Valgrind), add a -h flag.

Because adding build/bin to the PATH is not very convenient, ada/manage.py provides a shortcut to run the testsuite:

$ python ada/manage.py test

It runs the testsuite with the --enable-color option. It is also possible to pass other arguments to testsuite.py. For instance, if you want to run under a debugger only the factor_0 test case, execute:

$ python ada/manage.py test -- -g ada/testsuite/tests/parser/factor_0

Documentation

Libadalang comes with two separate Sphinx-based documentations: the User Manual and the Developer Manual.

The first one lies in the user_manual directory and the second one in the dev_manual directory. You can consult them as text files or build them. For instance, to generate HTML documents, run from the top directory:

$ make -C user_manual html
$ make -C dev_manual html

And then open the generated files in your favorite browser:

$ $BROWSER user_manual/_build/html/index.html
$ $BROWSER dev_manual/_build/html/index.html

Note that, as it relies on Python code introspection, you need a working Libadalang Python API in order to generate the User Manual.

Playing

Libadalang has a Python API, for easy prototyping and explorative programming. It ships with an executable named playground, that allows you to analyze Ada files and play with them in an interactive Python console.

Given the following main.adb Ada file:

with Ada.Text_IO; use Ada.Text_IO;

procedure Main is
begin
    Put_Line ("Hello World");
end Main;

You can start the playground on it:

% playground main.adb

--
-- libadalang playground
--

The file(s) passed as argument have been put into the `u` variable, or units if
there are multiple.

Enjoy!

In [1]: print u.root.text
with Ada.Text_IO; use Ada.Text_IO;

procedure Main is
begin
    Put_Line ("Hello World");
end Main;

In [2]: print u.root.findall(mdl.CallExpr)
[<CallExpr 5:5-5:29>]

In [3]: print u.root.findall(mdl.CallExpr)[0].text
Put_Line ("Hello World")

The playground embeds the IPython interactive Python console, so you have a modern interactive programming environment. You can use tab completion to explore the Libadalang API.