For example, the Ruby community uses this trick quite often so that one file can be a library if included by another source file, but can be an executable if invoked by the interpreter directly. See: http://gist.github.com/51337. If you run "ruby a.rb" you'll get something different than if your run "ruby b.rb", even though both are the result of "linking" a.rb and b.rb together.
...let's see you do that with C! ;-)
Your distinction would make Java a "scripting language," because every class can have its own main function, and still leaves x86 assembly as a "scripting language", and now it makes GNU C a "scripting language" too; and given the nondeterministic invocation order of the constructors of static objects, it makes any particular C++ implementation a "scripting language" too.
...and by the way, I would consider x86 assembly a scripting language. You're just scripting out the processor's actions...
#ifndef MAIN_ALREADY_DEFINED__
#define MAIN_ALREADY_DEFINED__
int main(){
/* file specific stuff */
}
#endif
Disclaimer: I wouldn't actually do this in practice...
Now, I don't know what you think, but these two languages seem to me to be almost perfect exemplars of the opposite extremes from how this definition would classify them. x86 assembly for .COM files has no type checking, no type coercion, no strings, no hash tables, very little in the way of runtime support (it pretty much owns the whole machine when it's running, under actual MS-DOS anyway, and you have to build everything yourself), takes months to learn enough about a computer to program in, typically is only used by fairly hard-core programmers, and there are no compatible interpreter implementations. VBScript is strongly dynamically typed, has strings, is used to "script" an existing application, runs in an interpreter, and is typically used by people who don't think of themselves as programmers.
So I suggest that Matz's definition may not really express what people are normally thinking of when they say "scripting languages".