Re: Why not gnat Ada in gcc?

[Laurent Guerby <guerby at acm dot org>]

RMS wrote:
> However, Guerby's message suggested that there are other issues
> involved, that this is not just a matter of making packages.

The only issue I mentionned (a while ago now) was some minor
performance issues on math functions where on x86 the GNAT NT version
produces faster code than the GNU/Linux version (due to asm inline on
NT, and calling C function on GNU/Linux which introduce a slight
overhead), and I believe this issue is resolved. 

ACT mentionned the other issues (shared lib) have been resolved.

The rest is just a matter of making package for popular distributions,
and testing them properly (we know we can't match ACT testing
here). Debian GNU/Linux packages are available.

RMS wrote:
> Anyway, would you please ask the people who make RPMs for the various
> versions of GNU/Linux to please call their work "GNAT for GNU/Linux"?

The only mention of the world Linux alone AFAIK is in the team name
"Ada for Linux Team" - ALT, all the web pages say GNU/Linux, you can
check at <http://www.gnuada.org/alt.html>.

Robert Dewar wrote:
> We have not yet got to the stage where we are sure that it is a good
> idea to use Linux Threads

The Linux Threads (based on the clone syscall) do not support some of
the optional real-time features (priority scheduling) and may be some
of the core feature (I'm not sure, ACATS testing will tell ;-) of Ada
95, but they offer multiprocessor support and syscall blocking only
one thread, whereas the FSU thread package is a user-space
implementation whith full support for all the real-time features, but
no multiprocessor and a blocking call blocks everything.

The ALT team chose to enable by default the Linux Threads
implementation which is the most used thread implementation and for
now the ACT package enables by default the FSU Threads which allows
for the full Ada 95 feature use. The two packagings allow to change
the thread implementation anyway, this is only a "default" issue.

It looks like priority scheduling use is pretty rare amongst the
GNU/Linux Ada community, I know of no freely available code for
GNU/Linux that depends on it (of course I guess it's pretty common for
the GNU RTOS RTEMS users ;-).

May be future version of Linux Threads will offer the needed features
though.

-- 
Laurent Guerby <guerby@acm.org>