"Do not use" feature list for GNAT

"Do not use" feature list for GNAT

[Florian Weimer]

Would you like to share a list of language areas which you should
avoid when using GNAT because AdaCore customers do not stress-test
them and they are riddled with strange bugs?

My current list looks roughly like this:

  - access discriminants (several ICE-on-legal bugs)

  - subunits (ICE, maybe on invalid input, I'm not sure)

  - the 'Valid attribute (wrong-code bug with GCC 4)

  - conditional entry calls (run-time bug in GNAT 3.15p)

Anything I've missed so far?

Re: "Do not use" feature list for GNAT

[Ludovic Brenta]

Florian Weimer <fw@deneb.enyo.de> writes:
> Would you like to share a list of language areas which you should
> avoid when using GNAT because AdaCore customers do not stress-test
> them and they are riddled with strange bugs?
>
> My current list looks roughly like this:
>
>   - access discriminants (several ICE-on-legal bugs)
>
>   - subunits (ICE, maybe on invalid input, I'm not sure)
>
>   - the 'Valid attribute (wrong-code bug with GCC 4)
>
>   - conditional entry calls (run-time bug in GNAT 3.15p)
>
> Anything I've missed so far?

Sometimes -gnatVa causes ICEs; I normally compile all Debian packages
with -gnatafno -gnatVa -O2, but I disable -gnatVa for certain files.

I hane not had any bad experiences with subunits and I haven't used
the other features enough to judge.

This is with 3.15p.

-- 
Ludovic Brenta.

Re: "Do not use" feature list for GNAT

[Preben Randhol]

Ludovic Brenta wrote:

> Sometimes -gnatVa causes ICEs; I normally compile all Debian packages
> with -gnatafno -gnatVa -O2, but I disable -gnatVa for certain files.

Stupid question: What does ICE mean?

Preben

Re: "Do not use" feature list for GNAT

[Ludovic Brenta]

Preben Randhol <randhol+njus2227@pvv.org> writes:
> Ludovic Brenta wrote:
>
>> Sometimes -gnatVa causes ICEs; I normally compile all Debian packages
>> with -gnatafno -gnatVa -O2, but I disable -gnatVa for certain files.
>
> Stupid question: What does ICE mean?

Internal Compiler Error.  aka "bug box".

-- 
Ludovic Brenta.

Re: "Do not use" feature list for GNAT

[Niklas Holsti]

Florian Weimer wrote:
> Would you like to share a list of language areas which you should
> avoid when using GNAT because AdaCore customers do not stress-test
> them and they are riddled with strange bugs?
> 
> My current list looks roughly like this:
> 
>   - access discriminants (several ICE-on-legal bugs)
> 
>   - subunits (ICE, maybe on invalid input, I'm not sure)
> 
>   - the 'Valid attribute (wrong-code bug with GCC 4)
> 
>   - conditional entry calls (run-time bug in GNAT 3.15p)
> 
> Anything I've missed so far?

I have seen wrong results from the predefined "=" operator for 
packed Boolean arrays (logically equal values were reported as not 
equal). This was GNAT 3.15p on Debian Sarge. I worked around the 
problem by defining my own, ugly, element-by-element version of 
the "=" operator.

-- 
Niklas Holsti
Tidorum Ltd
niklas holsti tidorum fi
       .      @       .

Re: "Do not use" feature list for GNAT

[Dmitry A. Kazakov]

On Tue, 23 Aug 2005 18:41:53 +0200, Florian Weimer wrote:

What I saw:

- corrupted visibility in generic subunits;

- requeue w/o abort is abortable;

- missing error messages when calling an abstract primitive subprogram from
a non-primitive one. [Not always, I cannot figure out when. Fortunately, it
leads to a link error.]

P.S. What if users of free GNAT grounded a non-profit organization with
some small subscription. If it collected enough money, it could become a
paying customer of ACT, solely to report bugs.

[Just a guess, I cannot tell if that would be legally possible.]

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de

Re: "Do not use" feature list for GNAT

[David Trudgett]

"Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de> writes:

> P.S. What if users of free GNAT grounded a non-profit organization with
> some small subscription. If it collected enough money, it could become a
> paying customer of ACT, solely to report bugs.
>
> [Just a guess, I cannot tell if that would be legally possible.]

Accepting bug reports only from subscribers seems like a very strange
concept...

David



-- 

David Trudgett
http://www.zeta.org.au/~wpower/

You have never been invaded and never been under dictatorship.  So
like the frog in the boiling water you may not realize until it's too
late that you've lost your capacity to think for yourselves and accept
personal responsibility. That is the quiet danger facing America.

    -- Lech Walesa
       (Leader of Poland's Solidarity union &
       President of Poland 1990-1995) 

Re: "Do not use" feature list for GNAT

[Adrien Plisson]

Dmitry A. Kazakov wrote:
> P.S. What if users of free GNAT grounded a non-profit organization with
> some small subscription. If it collected enough money, it could become a
> paying customer of ACT, solely to report bugs.
> 
> [Just a guess, I cannot tell if that would be legally possible.]

isn't it possible to find a paying customer of ACT that will kindly 
report bugs we are discussing here ?

(i don't know the details of ACT subscription and what it implies in 
term of bug reporting and fixing)

-- 
rien

Re: "Do not use" feature list for GNAT

[Ludovic Brenta]

Anyone can report bugs, but paying customers can get a fix quicker.

If you use Debian, use reportbug or M-x debian-bug from within emacs;
bug reports will come to me and anyone who subscribes to the relevant
package.  You can also send mail to report@bugs.debian.org, but be
sure to include detailed information (the tools I mentioned do that
automatically for you).  You can browse the bug database on
http://bugs.debian.org/gnat

If you use another distribution, follow your distribution's
instructions for reporting bugs.

If you compiled GCC or GNAT yourself, report bugs to GCC's Bugzilla
database: http://gcc.gnu.org/buzilla

If you use AdaCore's binary distribution of GNAT 3.15p, report bugs by
sending mail to report@gnat.com.  In this case, your bug report is not
published; you may want to help the community by making your report
public.  For this, you can use the Debian bug tracking system, or
GCC's bugzilla.

So, finding a kind paying customer or setting up a nonprofit org
just for reporting bugs is unnecessary; a nonprofit might help getting
bugs fixed faster, but the rates are prohibitive for casual or
amateur programmers; currently, the best they have is this forum.

-- 
Ludovic Brenta.

Re: "Do not use" feature list for GNAT

[Simon Wright]

"Ludovic Brenta" <ludovic.brenta@insalien.org> writes:

> If you use AdaCore's binary distribution of GNAT 3.15p, report bugs by
> sending mail to report@gnat.com.

And include GNAT in the subject line.

Re: "Do not use" feature list for GNAT

[Simon Wright]

Adrien Plisson <aplisson-news@stochastique.net> writes:

> isn't it possible to find a paying customer of ACT that will kindly
> report bugs we are discussing here ?

Not unless those bugs directly affect our project, using a supported
compiler or tool release; it would be quite dishonest to do so.

Re: "Do not use" feature list for GNAT

[Florian Weimer]

* Dmitry A. Kazakov:

> P.S. What if users of free GNAT grounded a non-profit organization with
> some small subscription. If it collected enough money, it could become a
> paying customer of ACT, solely to report bugs.

In my dealings with AdaCore, they were very explicit letting me know
that they don't want such support-by-proxy schemes.

Such an organisation should rather fund development on the FSF GCC
tree, IMHO, even though this is somewhat unrealistic because of the
scale of the task.