Re: a new language, designed for safety !

["Dan'l Miller" <optikos@verizon.net>]

On Sunday, June 8, 2014 8:56:56 AM UTC-5, Robert A Duff wrote:
> Georg Bauhaus writes:
> > - "I can view this Obj as of type TA and call Op"
> >     (Ada, wrong view)
> >
> > - "I can send Op: to Obj of presumed type TA"
> >    (Objective-C, no method)
> 
> The two errors seem rather different, to me.  To get the Ada error, you
> have to do a downward type conversion, something that is well known to
> be questionable -- you need to be careful to say "if X in T3'Class...".

They seem quite different to me too.  ASIS could search for all downward-cast syntax and flag it.  One of the very few areas where C++98 and later is superior to even Ada2012 is that the permissible downcast syntaxes are obnoxiously obvious in modern C++ to facilitate easier pattern-matching via grep and its brethren:

1) dynamic_cast< Type2 >( expressionOfType1 ), which uses runtime-type-identification (RTTI) to perform an Ada-like runtime check

2) static_cast< Type2 >( expressionOfType1 ), which is an unsafe* compile-time forcing of Type2's virtual-function table instead of Type1's.  If Type2 does not conform to Type1 in this regard (e.g., unimplemented member-function), then undefined* behavior occurs (e.g., eventual crash, corruption of resources, potential attack surface for a worm to inject machine code).

Unfortunately, Ada's downcast syntax is not so obnoxiously conspicuous to the human eye or to mere regular-expression-only pattern-matchers.  To rigorously search for Ada downcasts throughout all source code, it seems that a parser with the expressive power of ASIS is needed.

* "Unsafe" here is defined using Mr. Duff's narrower definition of "undefined behavior" in an unsafe language, as established on another recent thread-branch above back along this thread.  Of course, undefined behavior in an "unsafe language" also meets my wider definition of "unsafe system" since 1) the compiler plus source code that it is compiling forms a system and 2) undefined behavior could easily be the root-cause bodily injury or death.

> The dangling dispatch error in Objective C can happen on any call,
> and there's no practical way to protect against it.

  i.e., no practical way *within* Objective-C, the way that Objective-C is defined today.  Even with an Objective-C equivalent of ASIS (if one were ever to exist, such as based on LLVM-Clang's relatively lucid C++ compiler source-code), there exist more avenues of precipitating a run-time error at time of "sending" a "message" to an object in Objective-C than only downcasting.  Hence, there is no especially-dangerous syntax to look for in Objective-C, because every single "message" "sent" to an object can be dangerous in the way that (only) downcasting is dangerous in Ada.