Re: CVS commit: pkgsrc/lang/gcc3-java

[John Marino <netbsd%marino.st@localhost>]

On 7/19/2012 06:55, David Holland wrote:
> On Sun, Jul 15, 2012 at 09:59:06AM +0200, John Marino wrote:
>   >  >Removed Files:
>   >  >    pkgsrc/lang/gcc3-java: DESCR Makefile PLIST buildlink3.mk
>   >  >
>   >  >Log Message:
>   >  >Remove gcc3-java as promised. Has not been buildable for years, if ever,
>   >  >and various attempts to fix it have failed. Plus, gcc3 is highly obsolete
>   >  >at this point and it's hardly worth spending effort on this. I believe
>   >  >gcc-java support is available in most or all of the gcc4 packages.
>   >
>   >  Might I suggest removing lang/gcc34-ada as well?
>
> I thought we'd already decided (back in like January) that you'd do
> that when you were ready to.

I didn't recall the conversation well, but I found a copy of it:
http://mail-index.netbsd.org/pkgsrc-bugs/2012/01/22/msg046220.html

Since that post, not only does lang/gnat-aux build 4 other language by 
default now (based on gcc 4.6), but lang/gcc-aux has been published as 
well.  It's based on on gcc 4.7, supports Ada, Fortran, C, C++, and ObjC 
with excellent results on the testsuite.  Since it was built with same 
directory structure as the other lang/gccXX packages, it was possible to 
incorporate gcc-aux into mk/compiler/gcc.mk and support USE_LANGUAGES+= Ada.

That final bit was enough to upgrade devel/florist.  As far as I know, 
all the Ada programs are built with gcc-aux.   My main hesitation back 
then was packages like florist would be orphaned, but there was only one 
package like that and it's been addressed.

I wouldn't remove a package myself, I was only opining on yes/no and when.


>   >  Main reason for keeping it:
>   >  The new Ada compilers won't build on NetBSD 4 or lower.
>   >  They only build on x86 platforms, but I can't imagine
>   >  lang/gcc34-ada ever built on anything else.  So the main reason to
>   >  keep it is for NetBSD 4 users, but they have no packages they can
>   >  build with it in the pkgsrc tree.
>
> Does it actually build and work on netbsd-4?
>
> I don't think very many people are still using netbsd-4 at all, let
> alone with Ada, but it is still theoretically supported so it might be
> better to wait until it's officially EOL.

No. The referenced posts mentions why.  Part of the problems is NetBSD's 
'clever' use of functions like socket getting renamed to __socket30. 
There are many major-release specific versions like that.  In c, the 
headers like <sys/stat.h> and <socket.h> sort out what the real function 
name is but Ada doesn't use c headers when it's  compiling.  As a 
result, you have to hardcode stuff like __socket30 instead of socket 
into the build of GNAT and you have to account for every permutation.  I 
had to create several netbsd-specific Ada files even to upgrade from 
release 5.0 to 6.0.  No other unix-like platform has to be treated that way.

It probably wouldn't be impossible to support NetBSD 4, but it would be 
a very big undertaking.  I certainty am not motivated to do it.

That said - I'd still recommend to kill the -ada packages for these reasons:
1) We already saw a person who wanted Ada gravitate to these packages 
before discovering gnat-aux and gcc-aux
2) The quality of the Ada compiler is completely unknown
3) I believe gcc34 also builds Ada with a provided Ada compiler.  So 
theoretically NetBSD 4 still has an option.

so I'd recommend for now (not NetBSD 4 EOL):
1) kill gcc3-ada
2) kill gcc34-ada
3) remove references of gcc34-ada from gcc34 options.mk and anywhere 
else it's mentioned as a bootstrap compiler candidate.

But I'm recommending it, not saying I'll remove the package.  I've taken 
too much flak this week already.

John