Re: GNAT proposal: note on implicit exceptions insertion

[Jesper Quorning <jesper.quorning@gmail.com>]

tirsdag den 12. februar 2019 kl. 00.41.54 UTC+1 skrev Randy Brukardt:
> This sounds like a great idea, in that many people (including me) have had 
> this one over the years. (The first time I remember hearing about it was at 
> the very first AdaTec (which later become SigAda) meeting that we went to - 
> circa 1984).

Ideas takes time to become reality.
 
> Anyway, this depends on what you mean by such a switch. A switch that simply 
> reported every place where the language semantics requires a 
> language-defined check wouldn't be useful, because the language requires a 
> check almost everywhere (for every subtype conversion, for instance, 
> including those that don't actually do anything).

1: subtype Index_Type is range 1 .. Integer'Last;
2: My_Array : array (Index_Type) of Boolean;
3: ...
4: My_Index : Integer := Get_Intege;  -- Get integrer frome somewhere
5: ...
6: My_Array (My_Index) := True;

Should yield something like:

program.adb:6: (note) CONSTRAINT_ERRORE may be raise on array lower bound check

> A switch that only reports places that the compiler inserts checks is quite 
> expensive in practice to build, as it has to be associated with expensive 
> check elimination code to be useful. That can be especially difficult if the 
> check elimination code happens well after the code that handles warnings (as 
> in Janus/Ada).

It is just at note to the user of the compiler. Low optimisation mode gives more notes than high optimisation modes.
  
> I've been playing with such a switch in Janus/Ada lately (in a *very* 
> limited set of checks), and it just tends to prove that one's compiler isn't 
> as good at eliminating checks as one previously thought. :-) There are lots 
> of cases of checks being left that are "obviously" not needed, and all of 
> those provide noise in using such a switch for quality improvement. Having 
> such an obvious look is bad if only because it encourages spending effort on 
> check elimination (because of visibility) over other possible places to 
> spend effort. (This switch will be available in the current version of 
> Janus/Ada as soon as I get it released -- which will have to wait until my 
> Ada Standard work is done -- but I'm not planning on publicizing it as it 
> has too many false positives to be useful to most at this time.)

That this switch could lead to impoved compilers; I had not thougt of.

> In the case of AdaCore, they have other tools for finding quality problems 
> in Ada code (like CodePeer), and they may very well want to put new effort 
> into those rather than into the compiler.

You will reduce the need for other tools as the promary tool, GCC just reports as notes, what it is doing anyway.


        Jesper.