Newer
Older
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
"http://www.w3.org/TR/html4/strict.dtd">
Kevin
committed
<html>
<META http-equiv="Content-Type" content="text/html; charset=ISO-8859-1" />
<title>Clang - Get Involved</title>
<link type="text/css" rel="stylesheet" href="menu.css" />
<link type="text/css" rel="stylesheet" href="content.css" />
<h1>Getting Involved with the Clang Project</h1>
<p>Once you have <a href="get_started.html">checked out and built</a> clang and
played around with it, you might be wondering what you can do to make it better
and contribute to its development. Alternatively, maybe you just want to follow
the development of the project to see it progress.
</p>
<h2>Follow what's going on</h2>
<p>Clang is a subproject of the <a href="http://llvm.org">LLVM Project</a>, but
has its own mailing lists because the communities have people with different
interests. The two clang lists are:</p>
<ul>
<li><a href="http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits">cfe-commits
</a> - This list is for patch submission/discussion.</li>
<li><a href="http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev">cfe-dev</a> -
This list is for everything else clang related (questions and answers, bug
reports, etc).</li>
<p>If you are interested in clang only, these two lists should be all
you need. If you are interested in the LLVM optimizer and code generator,
please consider signing up for <a
href="http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev">llvmdev</a> and <a
href="http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits">llvm-commits</a>
<p>The best way to talk with other developers on the project is through the <a
href="http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev">cfe-dev mailing
list</a>. The clang mailing list is a very friendly place and we welcome
newcomers. In addition to the cfe-dev list, a significant amount of design
discussion takes place on the <a
href="http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits">cfe-commits mailing
list</a>. All of these lists have archives, so you can browse through previous
discussions or follow the list development on the web if you prefer.</p>
<h2>Open Projects</h2>
<p>Here are a few tasks that are available for newcomers to work on, depending
on what your interests are. This list is provided to generate ideas, it is not
intended to be comprehensive. Please ask on cfe-dev for more specifics or to
verify that one of these isn't already completed. :)</p>
<li><b>Compile your favorite C/ObjC project with "clang -fsyntax-only"</b>:
the clang type checker and verifier is quite close to complete (but not bug
free!) for C and Objective C. We appreciate all reports of code that is
rejected by the front-end, and if you notice invalid code that is not rejected
by clang, that is also very important to us. For make-based projects,
the <a href="get_started.html#ccc"><code>ccc</code></a> script in clang's
<tt>utils</tt> folder might help to get you started.</li>
<li><b>Compile your favorite C project with "clang -emit-llvm"</b>:
The clang to LLVM converter is getting more mature, so you may be able to
compile it. If not, please let us know. Again,
<a href="get_started.html#ccc"><code>ccc</code></a> might help you. Once it
compiles it should run. If not, that's a bug :)</li>
<li><b>Debug Info Generation</b>: -emit-llvm doesn't fully support emission
of <a href="http://llvm.org/docs/SourceLevelDebugging.html">LLVM debug info</a>
(which the code generator turns into DWARF). The missing pieces are pretty
minor at this point.</li>
<li><b>Overflow detection</b>: an interesting project would be to add a -ftrapv
compilation mode that causes -emit-llvm to generate overflow tests for all
signed integer arithmetic operators, and call abort if they overflow. Overflow
is undefined in C and hard for people to reason about. LLVM IR also has
intrinsics for generating arithmetic with overflow checks directly.</li>
<li><b>Undefined behavior checking</b>: similar to adding -ftrapv, codegen could
insert runtime checks for all sorts of different undefined behaviors, from
reading uninitialized variables, buffer overflows, and many other things. This
checking would be expensive, but the optimizers could eliminate many of the
checks in some cases, and it would be very interesting to test code in this mode
for certain crowds of people. Because the inserted code is coming from clang,
the "abort" message could be very detailed about exactly what went wrong.</li>
<li><b>Continue work on C++ support</b>: Implementing all of C++ is a very big
job, but there are lots of little pieces that can be picked off and implemented.
See the <a href="cxx_status.html">C++ status report page</a> to find out what is
missing and what is already at least partially supported.</li>
<li><b>Improve target support</b>: The current target interfaces are heavily
stubbed out and need to be implemented fully. See the FIXME's in TargetInfo.
Additionally, the actual target implementations (instances of TargetInfoImpl)
also need to be completed. This includes defining builtin macros for linux
targets and other stuff like that.</li>
<li><b>Implement 'builtin' headers</b>: GCC provides a bunch of builtin headers,
such as stdbool.h, iso646.h, float.h, limits.h, etc. It also provides a bunch
of target-specific headers like altivec.h and xmmintrin.h. clang will
eventually need to provide its own copies of these (and there is a <a href=
"http://lists.cs.uiuc.edu/pipermail/cfe-dev/2007-December/000560.html">lot of
improvement</a> that can be made to the GCC ones!) that are clean-room
implemented to avoid GPL taint.</li>
<li><b>Implement a clang 'libgcc'</b>: As with the headers, clang (or a another
related subproject of llvm) will need to implement the features that libgcc
provides. libgcc provides a bunch of routines the code generator uses for
"fallback" when the chip doesn't support some operation (e.g. 64-bit divide on
a 32-bit chip). It also provides software floating point support and many other
things. I don't think that there is a specific licensing reason to reimplement
libgcc, but there is a lot of room for improvement in it in many
dimensions.</li>
<li><b>Implement an tool to generate code documentation</b>: Clang's
library-based design allows it to be used by a variety of tools that reason
about source code. One great application of Clang would be to build an
auto-documentation system like doxygen that generates code documentation from
source code. The advantage of using Clang for such a tool is that the tool would
use the same preprocessor/parser/ASTs as the compiler itself, giving it a very
<li><b>Use clang libraries to implement better versions of existing tools</b>:
Clang is built as a set of libraries, which means that it is possible to
implement capabilities similar to other source language tools, improving them
in various ways. Two examples are <a href="http://distcc.samba.org/">distcc</a>
and the <a href="http://delta.tigris.org/">delta testcase reduction tool</a>.
The former can be improved to scale better and be more efficient. The later
could also be faster and more efficient at reducing C-family programs if built
on the clang preprocessor.</li>
<li><b>Use clang libraries to extend Ragel with a JIT</b>: <a
href="http://research.cs.queensu.ca/~thurston/ragel/">Ragel</a> is a state
machine compiler that lets you embed C code into state machines and generate
C code. It would be relatively easy to turn this into a JIT compiler using
LLVM.</li>
<li><b>Self-testing using clang</b>: There are several neat ways to
improve the quality of clang by self-testing. Some examples:
<ul>
<li>Improve the reliability of AST printing and serialization by
ensuring that the AST produced by clang on an input doesn't change
when it is reparsed or unserialized.
<li>Improve parser reliability and error generation by automatically
or randomly changing the input checking that clang doesn't crash and
that it doesn't generate excessive errors for small input
changes. Manipulating the input at both the text and token levels is
likely to produce interesting test cases.
</ul>
</li>
<p>If you hit a bug with clang, it is very useful for us if you reduce the code
that demonstrates the problem down to something small. There are many ways to
do this; ask on cfe-dev for advice.</p>