Unconstrained Objects

Unconstrained Objects

[Eric Shulman]

Does anyone know if modern Ada compilers allocate objects of
the largest possible size for an unconstrained object.
I would appreciate the names of any compilers if possible.
I am currently using Meridian PDK 4.1.4 and GNAT 2.00.

Thank you

Eric Shulman

Re: Unconstrained Objects

[Robert Dewar]

GNAT allocates the maximum size for unconstrained variant records. We
consider this the preferable implementation, since hidden use of the3
heap seems intolerable.

Re: Unconstrained Objects

[HUMPHREY_TERRY]

In <3ls78u$nr3@gnat.cs.nyu.edu> dewar@cs.nyu.edu writes:

> GNAT allocates the maximum size for unconstrained variant records. We
> consider this the preferable implementation, since hidden use of the3
> heap seems intolerable.
> 

Does this mean that GNAT does not implicitly allocated from the heap
under any circumstances?  If the answer is no, then under what circumstances
does GNAT implicitly allocate from the heap?

Thanks in advance,
Terry Humphrey

Re: Unconstrained Objects

[Cyrille Comar]

HUMPHREY_TERRY (HUMPHREY@ASD1.JSC.NASA.GOV) wrote:

: Does this mean that GNAT does not implicitly allocated from the heap
: under any circumstances?  If the answer is no, then under what circumstances
: does GNAT implicitly allocate from the heap?

GNAT allocates outer-level objects with a non-static size on the
heap. This is the only case of implicit allocation that will remains
in the final version.  GNAT used to allocate anonymous objects (for
functions returning unconstrained types) on the heap but it is not the
case anymore, now those objects are alocated on a secondary stack. If
you find other implicit uses of the heap (that is to say, leaks), they
are considered BUGS and should be reported as such to gnat-report.

In fact, there is another case of implicit allocation for the current
implementation of the secondary-stack, if it grows too much, it
allocates a new block on the heap.
In contexts where this is not desired, it is easy to provide another
implementation of the secondary-stack (in s-secsta.ad?) which is a
regular storage pool. No change to the compiler itself is needed, you
don't even need to recompile it...

--
------------------------------------------------------------------------
Cyrille Comar,                                  E-mail: comar@cs.nyu.edu
Gnat Project                                    US phone: (212) 998-3489

Re: Unconstrained Objects

[Robert Dewar]

Dan is wrong on several points. Most notably, it is perfectly legal for
a compiler to use a represetnation for a subtype that is different from
the basee type, and in certain cases (constrained subtypes of unconstrained
variant records with default discriminants), it is a common implementation.
But some compilers will even use a different representation for subtypes
of integer, and this is perfectly legitimate.

Re: Unconstrained Objects

[Robert Dewar]

"Under what conditions does GNAT implicitly allocate from the heap"


Very few, we consider implicit heap allocations to be a bad thing. Here is
the complete list:

  1.  When a task is created, its dynamic storage is allocated from the heap

  2.  When the secondary stack for the main task has to be expanded, we
	get space from the heap (we will probably restrict this in a 
	future version).

  3.  At the start of execution, we use malloc to allocate space for
	dynamically sized global objects.

Re: Unconstrained Objects

[Robert Dewar]

Cyrille forgot the task creation case, but otherwise our two posts agreed!