bug compiler ?

bug compiler ?

[bbaborum]

Hello everyone,

I tried to compile this 

procedure bug is
	type my_int is new integer;
	type access_function_my_int is access function (n : my_int) return my_int;
	function f (n : my_int) return my_int is (n);
	a : access_function_my_int :=f'access;
begin
	null;
end bug;

with the command line : gcc -c -gnat2012 bug.adb

and then I get this message from the compiler:


raised STORAGE_ERROR : stack overflow or erroneous memory access

If I modify the type "my_int" by writing subtype my_int is integer;
then I dont get any error from the compiler.

More interesting, if in the first program, the function f is no longer written inline, then I dont get any error.

Does anyone know what's happeing ?	

Re: bug compiler ?

[Lucretia]

On Friday, 30 June 2017 09:49:59 UTC+1, bbab...@gmail.com  wrote:
> Hello everyone,
> 
> I tried to compile this 
> 
> procedure bug is
> 	type my_int is new integer;
> 	type access_function_my_int is access function (n : my_int) return my_int;
> 	function f (n : my_int) return my_int is (n);
> 	a : access_function_my_int :=f'access;
> begin
> 	null;
> end bug;
> 
> with the command line : gcc -c -gnat2012 bug.adb
> 
> and then I get this message from the compiler:
> 
> 
> raised STORAGE_ERROR : stack overflow or erroneous memory access
> 
> If I modify the type "my_int" by writing subtype my_int is integer;
> then I dont get any error from the compiler.
> 
> More interesting, if in the first program, the function f is no longer written inline, then I dont get any error.
> 
> Does anyone know what's happeing ?

You can't create a local pointer to f or anything there, you have to define it in a package. You're essentially trying to create a pointer on the stack to something else on the stack. There is a way around it, I think unchecked_conversion, not sure now.

Luke.

Re: bug compiler ?

[Anh Vo]

On Friday, June 30, 2017 at 4:09:31 AM UTC-7, Lucretia wrote:
> On Friday, 30 June 2017 09:49:59 UTC+1, bbab...@gmail.com  wrote:
> > Hello everyone,
> > 
> > I tried to compile this 
> > 
> > procedure bug is
> > 	type my_int is new integer;
> > 	type access_function_my_int is access function (n : my_int) return my_int;
> > 	function f (n : my_int) return my_int is (n);
> > 	a : access_function_my_int :=f'access;
> > begin
> > 	null;
> > end bug;
> > 
> > with the command line : gcc -c -gnat2012 bug.adb
> > 
> > and then I get this message from the compiler:
> > 
> > 
> > raised STORAGE_ERROR : stack overflow or erroneous memory access
> > 
> > If I modify the type "my_int" by writing subtype my_int is integer;
> > then I dont get any error from the compiler.
> > 
> > More interesting, if in the first program, the function f is no longer written inline, then I dont get any error.
> > 
> > Does anyone know what's happeing ?
> 
> You can't create a local pointer to f or anything there, you have to define it in a package. You're essentially trying to create a pointer on the stack to something else on the stack. There is a way around it, I think unchecked_conversion, not sure now.
> 

The compiler should reject it if it does not compile with syntax rule. In this case, it does compile. In fact, if the function expression is replaced by normal function, it compiles fine. 

Therefore, a bug report should be sent to gnat@adacore.com.

Anh Vo

Re: bug compiler ?

[Anh Vo]

On Friday, June 30, 2017 at 7:49:12 AM UTC-7, Anh Vo wrote:
> On Friday, June 30, 2017 at 4:09:31 AM UTC-7, Lucretia wrote:
> > On Friday, 30 June 2017 09:49:59 UTC+1, bbab...@gmail.com  wrote:
> > > Hello everyone,
> > > 
> > > I tried to compile this 
> > > 
> > > procedure bug is
> > > 	type my_int is new integer;
> > > 	type access_function_my_int is access function (n : my_int) return my_int;
> > > 	function f (n : my_int) return my_int is (n);
> > > 	a : access_function_my_int :=f'access;
> > > begin
> > > 	null;
> > > end bug;
> > > 
> > > with the command line : gcc -c -gnat2012 bug.adb
> > > 
> > > and then I get this message from the compiler:
> > > 
> > > 
> > > raised STORAGE_ERROR : stack overflow or erroneous memory access
> > > 
> > > If I modify the type "my_int" by writing subtype my_int is integer;
> > > then I dont get any error from the compiler.
> > > 
> > > More interesting, if in the first program, the function f is no longer written inline, then I dont get any error.
> > > 
> > > Does anyone know what's happeing ?
> > 
> > You can't create a local pointer to f or anything there, you have to define it in a package. You're essentially trying to create a pointer on the stack to something else on the stack. There is a way around it, I think unchecked_conversion, not sure now.
> > 
> 
> The compiler should reject it if it does not compile with syntax rule. In this case, it does compile. In fact, if the function expression is replaced by normal function, it compiles fine. 
> 
> Therefore, a bug report should be sent to gnat@adacore.com.
> 
> Anh Vo

Correcting my typo - "it does compile" was meant "it does comply"

Re: bug compiler ?

[AdaMagica]

Just a guess: Maybe the compiler treats this simple statement function as having convention intrinsic. Intrinsic functions have no address and therefore no access is possible.
Whether this is legal - check the RM - I'm too lazy.

Re: bug compiler ?

[Randy Brukardt]

"AdaMagica" <christ-usch.grein@t-online.de> wrote in message 
news:c47d5433-8eb5-4dde-99f9-173f7dded62c@googlegroups.com...
> Just a guess: Maybe the compiler treats this simple statement function as 
> having convention intrinsic. Intrinsic functions have no address and 
> therefore no access is possible.
> Whether this is legal - check the RM - I'm too lazy.

It's legal. Expression functions aren't Intrinsic, they're just a short-hand 
for a usual function call. So this should be fine, and it seems that 
whatever is going on is not fine, so it is a compiler bug.

                            Randy.

Re: bug compiler ?

[Simon Wright]

Anh Vo <anhvofrcaus@gmail.com> writes:

> Therefore, a bug report should be sent to gnat@adacore.com.

report@adacore.com

Re: bug compiler ?

[Anh Vo]

On Friday, June 30, 2017 at 11:47:24 AM UTC-7, Simon Wright wrote:
> Anh Vo <anhvofrcaus@gmail.com> writes:
> 
> > Therefore, a bug report should be sent to gnat@adacore.com.
> 
> report@adacore.com

Thanks Simon for your correction.

Re: bug compiler ?

[bbaborum]

Le vendredi 30 juin 2017 13:09:31 UTC+2, Lucretia a écrit :
> On Friday, 30 June 2017 09:49:59 UTC+1, bbab...@gmail.com  wrote:
> > Hello everyone,
> > 
> > I tried to compile this 
> > 
> > procedure bug is
> > 	type my_int is new integer;
> > 	type access_function_my_int is access function (n : my_int) return my_int;
> > 	function f (n : my_int) return my_int is (n);
> > 	a : access_function_my_int :=f'access;
> > begin
> > 	null;
> > end bug;
> > 
> > with the command line : gcc -c -gnat2012 bug.adb
> > 
> > and then I get this message from the compiler:
> > 
> > 
> > raised STORAGE_ERROR : stack overflow or erroneous memory access
> > 
> > If I modify the type "my_int" by writing subtype my_int is integer;
> > then I dont get any error from the compiler.
> > 
> > More interesting, if in the first program, the function f is no longer written inline, then I dont get any error.
> > 
> > Does anyone know what's happeing ?
> 
> You can't create a local pointer to f or anything there, you have to define it in a package. You're essentially trying to create a pointer on the stack to something else on the stack. There is a way around it, I think unchecked_conversion, not sure now.
> 
> Luke.

No.
My access function is perfectly legal in this case.
If it were a problem, the compiler would produce an error message. Instead it simply crashes.

You didn't read the end of my post.

Re: bug compiler ?

[bbaborum]

Le vendredi 30 juin 2017 20:47:24 UTC+2, Simon Wright a écrit :
> Anh Vo <anhvofrcaus@gmail.com> writes:
> 
> > Therefore, a bug report should be sent to gnat@adacore.com.
> 
> report@adacore.com

Thank you.
I just reported it.