[cfe-dev] Extending Clang with information and code from external source

[David Blaikie]

On Thu, Aug 30, 2012 at 8:25 PM, Mark Boyall <wolfeinstein at gmail.com> wrote:
> Hi all,
>
> I'm attempting to build an external tool on top of Clang which is
> approximately (but not exactly) a C++ interpreter. Unfortunately, many of
> the terms used in Clang are quite generic and not searchable- seems like
> every compiler in the world ever made uses VarDecl as a type name, and I've
> had little luck finding documented examples of the C++ API. On top of that,
> the (not exactly) parts are causing me quite some headache. To begin with,
> I've been looking at how I'd interpret a simple Hello World program. But
> I've been a little stumped by some of the Sema's functionality, and have a
> few questions about how it can be used.
>
> Particularly, there is a function ArgumentDependentLookup which I need to
> use. But it takes expressions as arguments, rather than types? I thought
> that the C++ Standard defined ADL in terms of the argument types. More
> specifically,
>
> How do I convert a VarDecl to an Expr* so that it can be passed to this
> function? Logically, I want to pass this object, not it's initializer, and
> the documentation list only appears to allow me to convert the initializer
> to an Expr*.

you might want to play around with Clang's -ast-dump functionality to
explore the form of ASTs that show you what you're interested in
trying to emulate. In this specific case I believe you'd need a
DeclRefExpr to refer to the VarDecl you have.

> More generally, I'm talking about the "approximately (but not exactly) a C++
> interpreter", and we're talking about arguments which do not originate from
> C++ source code and do not have a Clang Expr* to be used, and possibly,
> there is no C++ equivalent expression that could ever exist- although there
> is a C++-equivalent argument type. How could I perform ADL for operations
> for which there is no C++ expression?
>
> I'm also facing fun when describing external types to Clang. For example, a
> RecordType can only be constructed from a RecordDecl, even if I inherited
> from RecordType, but no RecordDecl exists, because the description of this
> type did not come from a C++ file parsed by Clang. It was created
> externally. Even if I created a RecordDecl, it presents no methods to insert
> data about the type. This isn't just POD types, but special members,
> overloaded operators, and the whole shaboodle.

Perhaps you're looking for CXXRecordDecl? (again, looking at AST dumps
of valid C++ code will help you find the right AST nodes to represent
the constructs you want - if Clang can build an AST from source code,
you can build that same AST with a lot of hand-written/explicit code
to describe a source-less construct. Indeed there are some types that
cannot be written in C++ (see certain C++11 traits and
pseudo-intrinsics like std::initializer_list for the various ways the
compiler has to interact with the library))

>  In addition, I'm talking
> about functions, variables, namespaces, and the whole shaboodle. How can I
> use Clang's API to inform it of semantic objects that did not originate from
> a file parsed by Clang?
>
> I've looked at LookupOverloadedOperatorName, but apparently the only valid
> enum is OO_None. I don't think that operator<< is None. How is it possible
> to use this function to look up an operator?
>
> Once the function has been looked up, I'm going to be needing to convert the
> function to an LLVM representation, like llvm::Function*, so that I can
> generate code which calls into these functions through LLVM. For complex
> objects, I can read up on the ABI later, but for now, I wanted to focus on
> simple objects. So far, I figured that once I look up the name, I can use
> the MangleContext to mangle it, and then request the llvm::Function* from
> the llvm::Module based on the mangled name. Then I can generate LLVM code,
> link to the library and done. But where can I obtain the llvm::Module* from?
>
> Thanks for any assistance.
>
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>