Re: [xml-dev] Semantics and the Web: An Awkward History

[Michael Kay <mike@saxonica.com>]

> 
> I can't agree with this last point, though.  Dynamically typed files were a tremendous liberation from the statically typed files that preceded them, in which you generally had to talk to a local wizard to take a file written by a Fortran program that contained Fortran source code (such as the output of the Ratfor preprocessor) and make it acceptable to the Fortran compiler.  The difficulty was that Ratfor had written a "formatted sequential file", possibly with carriage control in column 1, whereas the compiler expected a "stream file" or "editable file", which was something completely different.
> 

I worked for 25 years with ICL's VME operating system, where every file had a "file description" that said something about its content. Files containing executable code were labelled as such; text files that could be edited with the text editor were labelled as such, and often with a subtype such as "Cobol program", as well as the character encoding; a hashed-random file containing variable length records with a 10-character key was labelled as such. It worked extremely well - except when it came to round-tripping files to other operating systems.

File names were not only orthogonal to file content (no silly ambiguous file extensions), they were also orthogonal to location (no need to rename a file when moving it to a different disk). And despite the fact that disk space was scarce and expensive in those days, files by default carried a full version history.

Sadly, too many people think the Unix file system is the best we can do.

Michael Kay
Saxonica