5.  Practical Experiences with Elk

5.1.  Elk and ISOTEXT

      In developing the document processing system ISOTEXT, Elk proved to be a major asset [Bormann 1991]. Scheme was used as the implementation language for all user interface aspects of ISOTEXT. Apart from providing extensibility to users of ISOTEXT, using Elk as the base for ISOTEXT made it possible to write the shell code in a high level language with all its amenities, e.g. automatic storage reclamation. As no recompilation and relinking is necessary, it is a quick operation to apply and test changes to the user interface.

      Elk provides for a strong ``firewall'' in the ISOTEXT system: bugs in the Scheme code give rise to errors at the Scheme level, which can easily be debugged using the (primitive, but functional) built-in debugger of Elk, while conditions such as core dumps always are the result of bugs in the ISOTEXT kernel implementation.

      All this assistance for the development of ISOTEXT could be obtained without sacrificing the performance of the ISOTEXT kernel system, which is still written in efficient C++.

      Elk also allowed us to isolate the ISOTEXT kernel from the choice of an X toolkit: the ISOTEXT kernel is unaware of the toolkit being used (``Xt'' with OSF/Motif). The Scheme code builds a user interface using the Motif library interface and provides X windows to the ISOTEXT kernel. Input is processed by the Scheme code which calls editor primitives provided by the ISOTEXT kernel and schedules redisplay operations. Replacing Xt and OSF/Motif by e.g. Xview would require no changes in the ISOTEXT kernel.

      The work on ISOTEXT clearly identified one single main problem in writing non-trivial extensions: as any request for new heap space can trigger a garbage collection, extensions must register local or temporary Scheme objects with the garbage collector to protect them from being discarded during a GC run caused by any nested procedure call. While this scheme has the advantage that maximum utilization of the available heap space is guaranteed, it imposes a strict discipline on the extension programmer. Failure to properly protect temporary Scheme objects usually results in delayed crashes of the application that are hard to trace back to the actual source of the problem. For instance, when developing the X11 extensions to Elk, most of the time spent for debugging was due to GC-related bugs.

5.2.  Elk and TELES.VISION

      Another example for using Elk and its X interface as the basis for a user interface subsystem is the TELES.VISION desktop video conferencing system [TELES 1991]. First, a somewhat generalized User Interface Management System was built in about 1500 lines of Scheme, which was then instantiated to build a number of revisions of the TELES.VISION user interface. The user interface communicates with the rest of the conferencing system via a remote procedure call C library, using Scheme continuations as a basis for a simple form of multithreading. According to the TELES.VISION implementors [Bastian 1993], Elk was a ``perfect fit'' for this application, with the single exception that its initial garbage collector placed too heavy a burden on the memory starved initial environment (where 8 MB of memory had to be shared between an operating system, various realtime device drivers, drivers for video codec hardware, and an MS-Windows emulation subsystem). This has since been remedied by adding memory. Using Elk also helped when TELES.VISION was ported to OS/2 -- in particular, its continuations ported easily. Also, Elk was used in the TELES.VISION project to build a rapid prototype of the central conference management subsystem (again using continuations to provide multithreading) within less than two weeks.

5.3.  Other Projects

      While Elk has been used in the ISOTEXT project since 1987, legal issues prevented making it publicly available until the fall of 1989. Since, Elk has gained acceptance, in fact sufficient momentum to encourage others to contribute software. Elk has been used successfully as an extension language for a hypertext database, a distributed version management system, various CAD programs, testing and simulation systems for digital circuits as well as environmental models. It also has found use simply as a Scheme programming environment, in particular for its X and Motif interface.

      The X extensions have proven useful in particular for writers of applications with graphical user interfaces based on X; Elk enables them to write their user interfaces or parts thereof in Scheme to achieve a high degree of customizability.

      Elk also has found use as a free-standing Scheme implementation. In combination with the X extensions it is well-suited for teaching X to beginners, as a tool for interactively exploring X, and as a platform for rapid prototyping of X-based applications.

      Outside of the UNIX world, we are aware of user-done ports to DOS (both 16 bit and 32 bit using DJGPP), OS/2, and MacOS.

      Users cited the following features as significant for their choice of Elk: dynamic object code loading, dumping of ready-to-run executables, Elk's performance, its legally unencumbered availability, and finally its simplicity and adaptability (and, as users say, its consistent, clean and well-structured code).

      Users are not happy with various artificial limitations still in the system (such as the static heap size which with the stop-and-copy garbage collector needs to be fixed at invocation time), with Elk's performance, and with the fact that Elk ``likes to be in control'' (i.e., supplies the main program). In addition, prospective users tend to ponder acceptance problems with their fellow workers and customers (who might not be well versed in Lisp/Scheme) before committing to Elk. Finally, for many extension language applications, Elk is ``too big'', and users have asked for versions without the more expensive Elk features such as arbitrary size number support or continuations. On the other hand, users have asked for additional features such as an inter-process communication interface, or a better debugger. Also, a port to MS-Windows has been actively sought.

Markup created by unroff 1.0,    September 24, 1996,    net@informatik.uni-bremen.de