Re: GNAT packages in Linux distributions

[Ludovic Brenta <ludovic@ludovic-brenta.org>]

Dmitry A. Kazakov writes:
>> http://people.debian.org/~lbrenta/debian-ada-policy.html
>
> Thanks. I was aware of the document, but didn't read it thoroughly.
>
>> See section 2.3.4 FSF releases and (now outdated) Appendix B. You will
>> see that the GCC 4.3 you are using is roughly equivalent to GNAT GPL
>> 2007; Debian testing has GCC 4.4 which is roughly equivalent to GNAT GPL
>> 2008.
>
> I.e.
>
> 2010 - 2008 = 2 (Fedora - gprbuild, GPS, GNADE ODBC, APQ)
> 2010 - 2007 = 3 (Debian + gprbuild, GPS, GNADE ODBC, APQ)
>
> Right?

Not entirely; you should qualify which version of Debian or Fedora you
are talking about:

2010 - 2008 = 2 (Fedora ??  - gprbuild, GPS, GNADE ODBC, APQ)
2010 - 2007 = 3 (Debian 5.0 - gprbuild, + GPS, GNADE ODBC, - APQ)
2010 - 2008 = 2 (Debian 6.0 + gprbuild, GPS, GNADE ODBC, APQ)

>> Note that I backport some bug fixes into the Debian version of
>> GCC.
>
> Do you have access to AdaCore wavefronts?

Not in my capacity as an unpaid, unsupported volunteer.

> To me. Many I have already forgotten, it is two years since then...
>
>> The bug database at AdaCore is private (and with good reason).
>
> Whatever reasons might be, they aren't good. (:-()

Oh no?  What about top-secret Ada code from the military that triggers a
bug in GNAT?

>> If you use the FSF version of GNAT, you should use the corresponding
>> bug database: Debian, Fedora or GCC upstream.
>
> Hmm, that would make sense only for bugs related to the compiler
> environment, packaging etc. If the bug is a compiler problem, there is
> no chance it can be fixed unless somebody is working on the compiler
> itself.  For example, there is a bug in assignment of controlled
> objects. It crashes Debian, leaks in Fedora, fixed in GPL 2009. Should
> it be reported? Where to?

To those that you use actively.  If you report it to Debian and I
determine that the bug was not introduced by Debian, then I will forward
it to the FSF.  If you have a workaround, it is better to document it
with the bug than let it remain unknown.  The workaround may help other
people avoid the bug and may also help people working on the compiler to
fix the bug (if not already fixed).

>> There are a few contributors external to AdaCore that provide bug
>> fixes into the FSF GCC, sometimes ahead of AdaCore; I normally
>> backport such fixes into the Debian version.  Eventually, all bug
>> fixes end up in both GNAT Pro/GPL and in FSF GCC.  The keyword is
>> "eventually"; the time lag can be months or years as you noted.
>
> So, when (if) GNAT GPL 2010 come, nothing will sufficiently change for
> FSF?  That is very disappointing. I hoped to jump off the GPL track
> (not because of the license issue).

FSF GCC 4.5.0 has just been released; this is roughly equivalent to GNAT
GPL 2009.  If you want the bleeding edge from the FSF, that's what you
want.  Unfortunately this is not packaged yet for Debian (and probably
not for Fedora either, but I don't know).  OTOH, if you can identify a
patch in GCC 4.5.0 that fixes a bug (and does only that), then I'll be
happy to backport it into Debian's GCC 4.4.4 for you.  Browsing the
changelog is, however, time-consuming.

-- 
Ludovic Brenta.