Language Subsets

[Marco Leise]

Am Mon, 16 Apr 2012 18:10:05 +0900
schrieb Klaim - Joël Lamotte <mjklaim at gmail.com>:

> On Mon, Apr 16, 2012 at 17:53, deadalnix <deadalnix at gmail.com> wrote:
> 
> >
> > @safe only allow operation that cannot screw up your memory. @system is
> > the complete language.
> >
> > @trusted is an explicit qualifie, to allow @safe code to call @system
> > code. It is mandatory, for instance, for @safe code to call the GC (GC code
> > cannot be @safe).
> >
> > It is up to the develloper to ensure that @trusted code can really be
> > trusted, so it isn't a subset.
> >
> 
> 
> Thanks, it helps a bit.
> However, a clear list of feature -> subset would be far more useful for me.
> 
> Klaim / Joel Lamotte

Since noone answered, I don't know what you heard in the video, but from my experience @safe is the only real subset that exists. And for you as the developer it means, that you can not use I/O and operations on pointers. Also when compiling in -release mode @safe keeps array bounds checks, while @system drops those checks. I imagine @safe code as encapsulated. It cannot talk to the outside world or access unrelated parts through pointer arithmetic. (Although, you can always call @trusted code which in turn calls @system code.) It's probably comparable to programming in a VM language like Java. You can apply @safe to either single functions or right at the top of your module to cover all code like so: "@safe:".

A feature table of @safe/@system is probably overkill, since most features still work. You just lose ASM/pointer arithmetics and direct calls to @system functions (which is the default). Work is on the way to mark as much of Phobos @safe as possible, since some functions still miss that attribute.

-- 
Marco