Re: Next Ada compiler for Debian: the votes so far

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

Dr. Adrian Wrigley  writes:
> Ludovic Brenta wrote:
>> But what is the platform that you are targetting today, and that
>> concerns you?  If you distribute proprietary software, you can always
>> make your own platform, or distribute GNAT 3.15p binaries with your
>> software.
>
> I am targeting x86 Linux.  I use 3.15p and GLADE (and other Ada
> libraries) There are bugs in both of these tools which bring me
> problems.  My next machine will be x86-64 to get increased
> addressing and speed.
>
> I should probably go to a dual Opteron, GNAT GPL 2005, PolyORB etc.
> but the time, cost and upheaval will be significant.  I want to be
> confident there will be real benefits and the tools will all work
> properly together.

With GNAT GPL 2005, there is no guarantee but if I were you I would
trust AdaCore for a distribution of excellent quality.  If you trust
the platform, a lot of the pain of migration goes away.  But you are
right: all transitions are painful.

> I am not distributing my software at present, so I don't have a
> direct interest in the choice of Debian compiler.  I run Debian and
> Red Hat at present, but I'm considering moving to all Debian.  In
> future, I may want to distribute Ada binaries or source, and don't
> want the opportunity to be obscured by changes in licensing of the
> Ada libraries.

So, you've been complaining loudly against the GPL for libgnat,
because of the remote possibility that you might someday decide to
distribute binary-only software?

Since you use both Red Hat and Debian, you know first-hand how much
better Debian is, especially as an Ada development platform.  But
using Debian is dangerous.  For one thing, it is addictive, and for
another, when comes the time to deploy your binaries on other
platforms, you'll find they lack half the Ada libraries you need, and
that the other half (which they do provide) are buggy.

You can solve this problem in a number of ways:

* Target Debian and only Debian, which runs on 12 hardware
  architectures anyway.  Who needs anything else than Debian?  Make
  sure your target is the latest stable release of Debian, not
  "testing" or "unstable", and you'll get the additional benefit that
  the platform will not move constantly.

* Assume that your targets provide *no* Ada support; assume they only
  conform to the Linux Standards Base (i.e. a certain glibc and a
  certain GCC compiler for C and C++, but no Ada).  Link your binaries
  statically against libgnat and any other Ada libraries you require,
  making sure that you use GMGPL sources of these libraries (either
  from GNAT 3.15p or from CVS).

* Assume your targets provide GCC 4.0 with Ada support, but no ASIS,
  no GLADE and no other libraries.  The rest is as above.

> What I'd like to see is a new version of GNAT and various libraries
> to become predominant.  A version suitable for all purposes.  A
> version with full Ada 2005 support, when possible.  A version for
> all the main Debian platforms. (I also run an Debian ARM system
> sometimes).  With robust Annex E and ASIS support.  I fear this this
> will not happen.  And I don't think ACT is minded to make it happen.
> But I may be mistaken.

If I were AdaCore, I would feel deeply offended by the mass rejection
of the GNAT GPL 2005 Edition that occured on this forum, and by the
pervasiveness of the "selfish argument".  So, I wouldn't count on
AdaCore going out of their way to please the people who complained.

If you don't want to use GNAT GPL 2005 Edition, then you have to fall
back on plan B, namely Debian Etch in your case.

Debian Etch will support amd64 and several other hardware platforms
beyond those supported in Sarge (namely i386, powerpc and sparc).

Debian Etch will have some Ada 200y features, but probably not all.
This depends on how GCC 4.1 turns out.

ASIS will probably work.  I'll take it from Martin Krishik's
SourceForge project because that's GMGPL.

GLADE will probably not work at first.  I'll take it from AdaCore's
CVS and try to compile it against GCC (3.4, 4.0 or 4.1, not decided
yet).  I will probably ask you, Adrian, for help in this area.

Debian Etch will not be as good as GNAT GPL 2005 Edition, and it will
arrive much later.  But it will support amd64.

> Perhaps I have been barking up the wrong tree, but it was commented
> earlied in this group that the 'exception' clauses for GtkAda and
> one or two other libraries had been removed.  I haven't seen this
> myself.  I had got the impression, however, that GLADE wasn't being
> maintained.  Checking the AdaCore CVS suggests I am wrong on this.
> But activity is moving to PolyORB, which doesn't seem ready for
> mission-critical Annex E use.

The special permission has been removed from the GNAT GPL 2005
Edition.  If you take the sources from CVS on gcc.gnu.org and
libre.adacore.com, you still get the GMGPL.  Therefore, if you want:

* the latest GNAT
* Ada 2005
* amd64 support
* the latest libraries
* support for all "x86 linux" distributions
* and the GMGPL,

then go get the sources and compile them for yourself.

I will do the same thing but only for Debian.  You may choose to wait
(several months) until this happens, or help me with it.  And please
understand that no matter what, it won't be as good or as
up-to-Ada-2006 as GNAT GPL 2005 Edition.

> Maybe as time progresses, we'll see that AdaCore will support a new
> "unversal" GNAT and libraries all under GMGPL type licensing.  Let's
> hope so.

I think it is a mistake to hope so.  It is much more productive get
down to work and build our own GMGPL platform.

-- 
Ludovic Brenta.