• Introduction
• Contact Information
• Overview for the Impatient
• Running Tests
• Supported Platforms
• Directories
• Features
• Outstanding Issues
• Future Work
• Further Information
• Project Team

Babel is a language interoperability tool intended for use by the high-performance scientific computing community. Developed by the Components project ( https://software.llnl.gov/Babel/ ) at Lawrence Livermore National Laboratory, Babel supports the Scientific Interface Definition Language (SIDL) for the language- independent declaration of interfaces associated with scientific software packages.

The Babel tool, applied to a SIDL file, results in the automatic generation of the associated skeleton and stub source files. The Babel user then need only add the necessary code to the _Impl source files to complete the provision of a language-independent interface to the package described by the SIDL file. The languages currently supported by Babel are C, C++, FORTRAN 77, Fortran 90/95, Fortran 2003/2008, Java and Python.

The remainder of this document provides basic information about Babel including features, outstanding issues, and future work.

If you have any questions or concerns with the installation process or usage of Babel, feel free to contact the project team at components@llnl.gov . This project is no longer funded, so it's mainly of historical interest.

Once you have completed the installation process (see the INSTALL file in this directory), you are ready to proceed with using Babel. A complete tutorial that steps you through the process of building the ubiquitious "Hello World!" example for a single client and server is available in the user's guide.

Briefly, now that you have the Babel compiler available, you'll need to have a sidl file defined before running the compiler. Below is a skeleton of the command line required to build the stubs, skeletons, Makefile fragment, and GNUmakefile.

    $ ./<babel>/bin/babel --makefile -s<lang> <name>.sidl

where is the desired language, is the name of the SIDL file, and is the installation's top directory.

Now you'll need to add the appropriate code within the splicer pairs in the _Impl files. Once you've finished filling in the implementation details, you can simply compile and link the appropriate libraries. Again, refer to the tutorial for more information.

If you are interested in running the regression tests that are packaged in this distribution you will need to build them since they are excluded from the basic installation process. As in the example in the INSTALL file, given that the Babel distribution is in $HOME/babel, the process for building and running the regression tests is as follows:

    $ cd $HOME/babel
    $ make check

Various diagnostics and results (e.g., PASS, XFAIL) will be displayed to indicate the number and status of each set of tests. At this time, there are three sets of tests each with three instances (9 total) that are expected to fail (i.e., XFAIL) These are for arrays in Python, arrays in Java, and exceptions in Java.

The Babel compiler has been successfully run on the platform/ compiler pairs listed below.

The Fortran 2003/2008 binding requires GCC >= 4.6.1, IBM XL Fortran >= 11.1, or Intel Compiler Suite >= 12.1.032 on any platform.

The following represents the source directory tree associated with this release, relative to the top level of the installation directory, and a brief description of the contents of each:

Key features, or capabilities, of the Babel toolkit are described in this section.

• SIDL Data Types SIDL defines the following data types:

    arrays      double        interface
    bool        enum          long
    char        fcomplex      opaque
    class       float         string
    dcomplex    int
  

  All data types are currently supported in C, C++, F77, F90, F03, Python, and Java with the additional exception of long in Python. It is important, however, to note that the binding for enum in F77 and F90 is currently INTEGER due to Babel's dependence upon C for its internal object representation (IOR).

  Refer to ./regression/args for clients and/or servers in C, C++, F77, F90, F03, Python, and Java that exploit most data types. Use of arrays can be found in ./regression/arrays (including in Python), enums in ./regression/enums, and strings in ./regression/strings. A crash course in SIDL arrays can be found in the users guide.

• SIDL Data Passing Modes The following data passing modes are defined: in, inout, out, and return. Note that return values in F77 and F90 are translated to out arguments. Refer to ./regression/args for clients and servers in C, C++, F77, F90 and F03 that exploit these modes.

• SIDL Exceptions: SIDL provides support for exceptions. Clients are, however, required to check the return values. A set of macros are defined in ./runtime/sidl/SIDL_Exception.h to facilitate throwing and catching exceptions in C. Refer to ./regression/exceptions for C, C++, F77, F90, F03, and Python servers and C, C++, F77, F90, F03, Java, and Python clients.

• SIDL Inheritance: SIDL supports Java-like inheritance. Refer to ./regression/inherit for servers in C, C++, F77, F90 and F03 and clients in C, C++, F77, F90, F03, Java, and Python that exploit these features.

• SIDL Method Overloading: SIDL supports method overloading on the client and server side for object-oriented languages using an exact match of the arguments. Languages that support method overloading make use of the method name from the SIDL file; whereas, those that do not support overloading rely on the optional method name extension to build unique method names. Refer to ./regression/overload for examples of the specification and use of overloaded methods in C, C++, F77, F90 and F03 servers and C, C++, F77, F90, F03, Java, and Python clients that exploit these features.

• Makefile Generation: Babel generates a lot of code. To assist the developer in managing the code, Babel will generate fragments of makefiles. These files are called "babel.make" and define standard macros that list all the relevant source code generated in that directory.

• Code Splicing: Although Babel generates a lot of code, some code (notably) "impl" files, require hand editing to produce useful libraries. Babel preserves your previous edits in certain files. This is done with some special CodeSplicer directives embedded in comments in the generated code. All code nested between matching CodeSplicer directives will be preserved by the Babel compiler.

• Direct Access to Numerics in FORTRAN 77: "Pointers" are provided for direct access to numeric data types (i.e., dcomplex, double, fcomplex, float, int, and long). There is a potential for an alignment problem with arrays of dcomplex and double, in particular, so check the return value of the pointer before using it.

• Array Element Access in FORTRAN 77 and Fortran 90: Array elements are accessible via subroutine calls. For more information, see the users manual.

• Strings in FORTRAN 77: When implementing a method with an inout or out string, the size of the outgoing string is at least 512 characters long.

• Enums in FORTRAN 77 and Fortran 90: Due to Babel's dependence upon C for its internal object representation, the binding for an enumerated type is INTEGER. The good news is that if your Fortran and C compilers use the same word size for C int and Fortran INTEGER, everything should work fine. However, if the sizes are different, then you may have problems with using enums.

There are several aspects of the distribution that still need work. The outstanding issues that need to be addressed include features as well as documentation.

• Python 'long' Data Type: Arrays of long are treated as arrays of int32_t in Python. Outside of arrays, SIDL long is mapped to Python's indefinite precision integer data type.

• Support for Multiple Fortran Compilers In order to use SIDL with multiple Fortran compilers, there currently must be a separate Babel installation for each compiler. You will also need different versions of the runtime library.

• Build Processing A mechanism for simplifying the build process needs to be explored, especially to facilitate user builds.

• C++ Shared Libraries and libtool In general, the creation of shared libraries in C++ seems to be problematic. It was necessary to make compiler-specific modifications to libtool. Consequently, this distribution includes our modified libtool.

The following is a list of key activities that would have been explored for incorporation into subsequent releases:

• Expanded platform and compiler support
• Simplified Build processing
• Data Redistribution
• Expanded interface contract specification and enforcement capabilities

The following files are available at the top of the release directory structure provide additional information on the Babel release:

Additional background information can be found in the ./doc/papers and ./doc/talks subdirectories including the outdated specification (./doc/papers/specification.ps). More recent papers and talks can be found at our web site at our project page https://software.llnl.gov/Babel/

Additional documentation including a tutorial, command line arguments, SIDL grammar, and crash courses on SIDL arrays and Fortran can be found in ./doc/babel101.

The LLNL Components Project has ended. The people were working on it at the end were:

• Tammy Dahlgren
• Tom Epperly
• Adrian Prantl

Community code contributors:

• Boyana Norris
• Ben Allan
• Stefan Muszala

Our alpha testers are:

• Bill Bosl
• Jeff Painter
• Andy Cleary
• Steve Smith

and our alumni are:

• Melvina Blackgoat
• Nathan Dykman
• Kevin Durrenberger
• Dietmar Ebner
• Sarah Knoop
• Scott Kohn
• Gary Kumfert
• Jim Leek
• Brent Smolinski