Surface syntax is overrated

Many applications are specified in terms of surface syntax, for example:

• Typesetting and documentation systems such as TeX, LaTeX, Texinfo, etc.
• Assemblers such as GAS or NASM.
• Parser generators such as YACC or Bison.

Other applications are specified in terms of the interaction gestures that they accept, for example:

• Word processors such as Word or the Libre Office word processor.
• Spreadsheet programs such as Excel or the Libre Office spreadsheet program.

• People have very strong opinions about surface syntax and interaction gestures. A program can be very good, but will still be rejected by users because of these superficial characteristics.
• If input to such a program is generated by another program as opposed to an end-user, the data must be unparsed and then parsed again, which is unnecessary work. A typical example would be an assembler used as a backend for a compiler.
• Customizing the core logic of such programs is hard or impossible, simply because the way they work internally is not specified or documented.
• Frequently, concentrating on superficial characteristics in the design process can lead to badly designed data structures for the core logic.
• Designing a good surface syntax or a set of good interaction gestures is very hard.

As an alternative, we propose that programs such as those in the examples should first be specified in terms of the data structures they manipulate, and in terms of the ways that these data structures can be manipulated.

By concentrating on the core logic and specifying it, we can obtain several advantages:

• We automatically obtain an interchange format in the form of a serialized version of the graph that makes up the objects of the application.
• Customization is easier because the format of the data is clearly specified in a way that can be directly manipulated by the custom code.
• Several different applications using different surface syntax or different input gestures can co-exist for a single version of the core data structures, thus allowing easier collaboration between users with different preferences.