Re: Gnat Problem - Freezing too soon

[russ lyttle <lyttlec@removegmail.com>]

On 3/2/19 4:35 PM, Simon Wright wrote:
> russ lyttle <lyttlec@removegmail.com> writes:
> 
>> I meant why it solved this particular problem. As far as I can tell,
>> the authors never used the "not null" in any of their books. Would
>> other compilers accept the code as written, or should the code be
>> considered erroneous?
> 
> 'not null' is clearly better style.
> 
> With regard to
> 
>     task type Sporadic (S : Any_Sporadic_State;
> 		       A : Any_Sporadic_Thread_Interface)
>     with Priority => S.pri
>     is
> 
> GNAT CE 2018 fails, GNAT GPL 2017 & earlier & GCC 9 succeed. You can see
> that making the type 'not null' would eliminate a check (check would be
> done at the call site instead), so that the internal tree generated by
> the compiler would be different, and the compiler may have been changed
> in this region for other reasons.
> 
> With regard to
> 
>     protected type Sporadic_Agent (S: Any_Sporadic_State)
>     with Priority => S.Ceiling_Prioriy
>     is
> 
> all the compilers are seriously confused:
> 
>     sporadics.ada:30:22: no selector "Ceiling_Prioriy" for type "Sporadic_State'Class" defined at line 17
>     sporadics.ada:30:23: possible misspelling of "Ceiling_Priority"
>     sporadics.ada:30:23: possible misspelling of "Ceiling_Priority"
>     sporadics.ada:30:23: possible misspelling of "Ceiling_Priority"
> 
> and you only have to look at the task equivalent above to see that it
> should have been OK. This one looks like another compiler error to me.
> 
There is a misspelling: Prioriy => Priority. Correct that misspelling 
and the idiom _sometimes_ works without the "not null". The bug report 
had an instance of where the compiler failed. Would it be helpful to add 
an instance where it passed?

>> As Simon pointed out replacing "with Priority => S.pri" with "pragma
>> Priority(S.pri)" also works. They never used that form either.
>>
>> Why would the authors prefer one form over the others?
> 
> The aspect is the newer form; the book was published in 2016, so they'd
> have gone with that.
> 
That's a good reason :)