Ada/GNAT runtime library?

Ada/GNAT runtime library?

[Bernd Trog]

Hi,
does anyone knows for which OSs the GNAT runtime
library works?
Or even better, where it is documented?

After digging in the gcc/ada/ tree for some time I
found 
some 'OS-strings':

DOS, Linux, OS2, OpenVMS, win32

Is this list complete?

Thanks,
Bernd


__________________________________________________
Do you Yahoo!?
Faith Hill - Exclusive Performances, Videos & More
http://faith.yahoo.com

Re: Ada/GNAT runtime library?

[Geert Bosch]

On Thursday, Oct 17, 2002, at 16:01 America/New_York, Bernd Trog wrote:
> does anyone knows for which OSs the GNAT runtime
> library works?
> Or even better, where it is documented?

See LIBGNAT_TARGET_PAIRS in ada/Makefile.in.
However, for the GCC 3.x back end, many of these will not work
yet. Targets with highest chance of success are x86/GNU Linux,
SPARC/Solaris and PA-RISC/HP-UX.

Re: Ada/GNAT runtime library?

[Bernd Trog]

Geert Bosch <bosch@gnat.com> wrote:
> yet. Targets with highest chance of success are
> x86/GNU Linux,
> SPARC/Solaris and PA-RISC/HP-UX.

Is it possible to build a Ada cross compiler for the
other CPUs that are supported by the backend, when I
disable the GNARL?

Bernd




__________________________________________________
Do you Yahoo!?
Faith Hill - Exclusive Performances, Videos & More
http://faith.yahoo.com

Re: Ada/GNAT runtime library?

[Geert Bosch]

On Thursday, Oct 17, 2002, at 17:22 America/New_York, Bernd Trog wrote:
> Is it possible to build a Ada cross compiler for the
> other CPUs that are supported by the backend, when I
> disable the GNARL?

Yes, you should be able to build "any" cross-compiler.
GNAT is designed to only need a subset of the run-time
library which doesn't require a target-specific implementation.
For example, the compiler does not use tasking and similar
Ada features.

   -Geert

Re: Ada/GNAT runtime library?

[Bernd Trog]

Geert Bosch <bosch@gnat.com> wrote:
> Yes, you should be able to build "any"
> cross-compiler.
> GNAT is designed to only need a subset of the
> run-time
> library which doesn't require a target-specific
> implementation.
> For example, the compiler does not use tasking and
> similar Ada features.
OK, thats makes sense.

Are there some Ada-specfic commands for building a
cross-compiler w/o the GNAR?

Or is it as simple as that:

configure --target=$cpu --prefix=/src/$cpu
--enable-languages=ada,c

(assuming the binutils for the target have been build
before)

Bernd,
who only build the GCC-CVS/Ada compiler on linux for
linux successfully :-/ 





__________________________________________________
Do you Yahoo!?
Faith Hill - Exclusive Performances, Videos & More
http://faith.yahoo.com

Re: Ada/GNAT runtime library?

[Nix]

On Fri, 18 Oct 2002, Florian Weimer said:
> Ah, you can't get this information on Linux.  It's shielded by the
> kernel.

On 2.4.19, I see an AT_CLKTCK ELF note (numbered 17) which may
suit. It doesn't exist in 2.2.x and below kernels though.

-- 
`It's hard to properly dramatize, say, the domestic effects of Dad's
 bank overdraft when a giant writhing kraken is levelling the city.'
          --- Shiner & Sterling, _Turkey City Lexicon_

Re: Ada/GNAT runtime library?

[Bernd Trog]

Robert Dewar <dewar@gnat.com> wrote:
> > What about Ada on GNU/Linux ARM? (its a 32 bit,
> too)

> We are by the way in the process of doing a GNAT
> port for the ARM

Thats _REALY_ greate news for me  -  and the
embedded-systems-world, I guess :)

Its good to see that there is now sufficient
commercial
interest :)


Bernd




__________________________________________________
Do you Yahoo!?
Faith Hill - Exclusive Performances, Videos & More
http://faith.yahoo.com

Re: Ada/GNAT runtime library?

[Bernd Trog]

Robert Dewar <dewar@gnat.com> wrote:
> > OK, I'll take a look on the 705 .ads files
> .........
> 
> Well you should look at only the ones relating to
> the particular function you are interested in.

Thanks,
I'll take a look.

Bernd


 

__________________________________________________
Do you Yahoo!?
Faith Hill - Exclusive Performances, Videos & More
http://faith.yahoo.com

Re: Ada/GNAT runtime library?

[Bernd Trog]

Joel Sherrill wrote:
> There are multiple pieces required
> to support an arbitrary language/platform 
> combination.

[ .. ]
Thanks for all details, Joel!

It look like a major project to me.
Maybe its better to abandon this idea, and keep the
developers fixing all the bugs :-/

> For a language/platform combination to work, these
> general categories must be supported in gcc assuming
> that the language and platform are supported at all:
> 
>    + port backend must support all features required
>      by language frontend
>    + run-time must be ported to CPU family   
>    + run-time must be ported to platform
 

> I would think that the existence of this list would
> help people choose a project to work on.

But maybe not worth the effort - it has to be
maintained, too.


> > What about Ada on GNU/Linux ARM? 
> In this case, the first requirement I listed above 
> is not satisfied.  

and Geert Bosch <bosch@gnat.com> wrote in
<41069ECB-E21A-11D6-A8D0-00039344BF4A@gnat.com>

> GNAT is designed to only need a subset of the
> run-time library which doesn't require a
> target-specific implementation.

Well, I'm sorry, but I don't get it. 

What are these features?(pointers are welcome)
Which backends don't support the 'Ada-features'?
Are these just m68k, sh, (ARM; work in progess)?

What about the AVR backend, does it support the
'Ada-features'?


Bernd






__________________________________________________
Do you Yahoo!?
Faith Hill - Exclusive Performances, Videos & More
http://faith.yahoo.com

Re: Ada/GNAT runtime library?

[Florian Weimer]

dewar@gnat.com (Robert Dewar) writes:

>> These are GNU libc bugs (or missing features) and should be fixed
>> there.

> No, that's quite wrong, on most targets the thread package is supplied
> by the manufacturer it has nothing to do with GNU libc.

I was talking about the GNU platforms, of course.  autoconf-based
portability to other GNU-supported architectures should be possible
(if POSIX threads are provided by GNU libc), otherwise there is some
portability problem which might not affect only GNAT, but a wider
range of software.  That's why I think these issues should be
addressed in GNU libc, so that all applications can benefit.

>> Timer frequencies?
>
> Just an example, the issue of how you know the timer frequency, whether
> the timer has to be initialized etc.

Ah, you can't get this information on Linux.  It's shielded by the
kernel.  There are some traditional values for HZ on most
architectures, but these values can change (and recently did on x86,
AFAIK).

Re: Ada/GNAT runtime library?

[Robert Dewar]

> These are GNU libc bugs (or missing features) and should be fixed
> there.


No, that's quite wrong, on most targets the thread package is supplied
by the manufacturer it has nothing to do with GNU libc.

> Timer frequencies?

Just an example, the issue of how you know the timer frequency, whether
the timer has to be initialized etc.

Re: Ada/GNAT runtime library?

[Arnaud Charlet]

> These are GNU libc bugs (or missing features) and should be fixed
> there.

Not every OS uses GNU libc. And what one would consider as a bug is called
a "feature" by others.

Arno

Re: Ada/GNAT runtime library?

[Florian Weimer]

dewar@gnat.com (Robert Dewar) writes:

> I think that's right, but this would only take care of the easy things.
> The difficult things (subtle semantic differences in thread packages,

These are GNU libc bugs (or missing features) and should be fixed
there.

> timer frequencies etc) are a harder nut to crack.

Timer frequencies?

Re: Ada/GNAT runtime library?

[Robert Dewar]

> autoconf could probably be used to generate these files.

I think that's right, but this would only take care of the easy things.
The difficult things (subtle semantic differences in thread packages,
timer frequencies etc) are a harder nut to crack.

Re: Ada/GNAT runtime library?

[Joel Sherrill]

Robert Dewar wrote:
> 
> > In this case, the first requirement I listed above is not
> > satisfied.  The ARM port backend and Ada do not get along
> > well enough to complete a build of Ada.  Similarly for the
> > m68k and sh.
> 
> That's what we are correcting. We now have a version internally that is
> starting to show signs of life, and will contribute changes ASAP.

Fabulous!!! I would be happy to do some testing on ARM/RTEMS and
ARM/Linux when it becomes available.

-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel@OARcorp.com                 On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
Support Available                (256) 722-9985

Re: Ada/GNAT runtime library?

[Florian Weimer]

Bernd Trog <berndtrog@yahoo.com> writes:

> What puzzels me, is the close dependancy between the
> OS and the CPU for the GNARL.
> e.g.
> GNU/Linux x86 works
> GNU/Linux Alpha don't work

Some Ada files duplicate constants from the libc header files, and
these constants differ on different Linux platforms.

autoconf could probably be used to generate these files.

Re: Ada/GNAT runtime library?

[Robert Dewar]

> What about Ada on GNU/Linux ARM? (its a 32 bit, too)

Who knows? Depends whether the linux threads interface is identical or not.
I don't know if anyone has tried this. 

We are by the way in the process of doing a GNAT port for the ARM
(although not on GNU/Linux) so we will be shaking out problems in that
port (and a lot of the problems will of course apply to any GNU/Linux
port on ARM as well).

Re: Ada/GNAT runtime library?

[Robert Dewar]

> In this case, the first requirement I listed above is not
> satisfied.  The ARM port backend and Ada do not get along
> well enough to complete a build of Ada.  Similarly for the
> m68k and sh.

That's what we are correcting. We now have a version internally that is
starting to show signs of life, and will contribute changes ASAP.

Re: Ada/GNAT runtime library?

[Robert Dewar]

> > It's easy enough to build a GNU/Linux Alpha version
> > of GNAT, others have done it for sure.
> 
> So, we can add GNU/Linux Alpha to the list?

No, because the necessary tasking files are not there. Someone has to
put them there. We don't have them because we have not done this port.
When I say it is easy to build, I mean that it is not that hard to 
do the work for the port, not that you can press a button and build
this version.

Re: Ada/GNAT runtime library?

[Robert Dewar]

> OK, I'll take a look on the 705 .ads files .........

Well you should look at only the ones relating to the particular
function you are interested in. If you are interested in the
interface to tasking from the compiler for example, look at the
specs that relate to this

(rtsfind.ads is a useful centralized catalog of runtime interfaces,
and you will see that only a small fraction of the files are used
by the compiler. The great majority of the ads files you mention
have nothing to do with compiler/runtime interface).

Re: Ada/GNAT runtime library?

[Robert Dewar]

> What about Ada on GNU/Linux ARM? (its a 32 bit, too)

Who knows? Depends whether the linux threads interface is identical or not.
I don't know if anyone has tried this. 

We are by the way in the process of doing a GNAT port for the ARM
(although not on GNU/Linux) so we will be shaking out problems in that
port (and a lot of the problems will of course apply to any GNU/Linux
port on ARM as well).

Re: Ada/GNAT runtime library?

[Robert Dewar]

> > It's easy enough to build a GNU/Linux Alpha version
> > of GNAT, others have done it for sure.
> 
> So, we can add GNU/Linux Alpha to the list?

No, because the necessary tasking files are not there. Someone has to
put them there. We don't have them because we have not done this port.
When I say it is easy to build, I mean that it is not that hard to 
do the work for the port, not that you can press a button and build
this version.

Re: Ada/GNAT runtime library?

[Robert Dewar]

> I have tested). I'm almost sure there are DOS caused limitations
> (eg. tasking is unlikely to work, etc).

Right, an Ada implementation without tasking is like a car without
any headlights. OK for limited use, but really not qualifying for
being regarded as an implementation of Ada. That's why I said that
I thought the DOS port had never been completed. There was a full
featured DOS port early on (done by Doug Rupp who now works for
ACT, but not on the DOS port, which never had any commercial
customers).

Note incidentally that the name of the language is Ada, not ADA. Us
Ada folks are picky about that. It's named after a person (Ada Augusta
Byron, Countess of Lovelace (*), given credit as the worlds first programmer
because of her work for Babbage). It is not an acronym, ADA may be the
American Dental Associationl, or the Americans with Disabilities Act,
(**) but it is not a programming language :-)


(*) This connection explains the name Linda that Gerlernter chose for
his parallel programming language. it was named after the other
(in)famous Lovelace.

(**) There is a wonderful poster with Clinton signing a bill, with a
caption something like "ADA IS IMPORTANT TO YOU" :-)

Re: Ada/GNAT runtime library?

[Joel Sherrill]

Bernd Trog wrote:
> 
> Robert Dewar <dewar@gnat.com> wrote:
> > > Why not put it somewhere on gcc.gnu.org?
> > It's clear in the makefile, and publishing the list
> > might make people think that GNAT can be built for
> > all these targets which is not the case (at least
> > we have not tried some of them).
> 
> Well, right now I don't see ANY restritions mentioned
> on gcc.gnu.org regarding language/platform
> compatibility :-(

There are multiple pieces required
to support an arbitrary language/platform combination.
Each can either not support or be broken.  Unfortunately,
I don't think that is a master list of what piece does
not support what. :(  I don't think a single person can 
generate it given the wide variety of knowledge required.  
It could be merged by a single person though.  

For a language/platform combination to work, these 
general categories must be supported in gcc assuming 
that the language and platform are supported at all:

   + port backend must support all features required
     by language frontend
   + run-time must be ported to CPU family
   + run-time must be ported to platform

So I would think that the list you are looking for would 
have to include a top level that was CPU family vs language
and run-time.  A yes would indicate that any platform within
that family is supported.  

The next level down would have to list platforms within
a CPU family.  

For example, say Ada works on powerpc-linux but not 
powerpc-winnt.  It would get a yes on the highest level
for powerpc, but a mix of yes and no on the powerpc 
platform specific section.

I would think that the existence of this list would help
people choose a project to work on.  

> > > What puzzels me, is the close dependancy between
> > the
> > > OS and the CPU for the GNARL.
> > > e.g.
> > > GNU/Linux x86 works
> > > GNU/Linux Alpha don't work
> >
> > Why should this puzzle you. The interface to the
> > threads library is different in these two versions
> > of GNU/Linux, in particular, one is 32 bit and one
> > is 64-bit.
> OK, that makes sense.
> 
> What about Ada on GNU/Linux ARM? (its a 32 bit, too)

In this case, the first requirement I listed above is not
satisfied.  The ARM port backend and Ada do not get along
well enough to complete a build of Ada.  Similarly for the
m68k and sh.

> > It's easy enough to build a GNU/Linux Alpha version
> > of GNAT, others have done it for sure.
> 
> So, we can add GNU/Linux Alpha to the list?

Sounds like it based upon Dr. Dewar's comments.

> Bernd
> 
> __________________________________________________
> Do you Yahoo!?
> Faith Hill - Exclusive Performances, Videos & More
> http://faith.yahoo.com

-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel@OARcorp.com                 On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
Support Available                (256) 722-9985

Re: Ada/GNAT runtime library?

[Bernd Trog]

Robert Dewar <dewar@gnat.com> wrote:
> package specs (Ada documents all such interfaces
more > or less automatically :-)

OK, I'll take a look on the 705 .ads files .........

Bernd





__________________________________________________
Do you Yahoo!?
Faith Hill - Exclusive Performances, Videos & More
http://faith.yahoo.com

Re: Ada/GNAT runtime library?

[Bernd Trog]

Robert Dewar <dewar@gnat.com> wrote:
> > Why not put it somewhere on gcc.gnu.org?
> It's clear in the makefile, and publishing the list
> might make people think that GNAT can be built for 
> all these targets which is not the case (at least
> we have not tried some of them).

Well, right now I don't see ANY restritions mentioned
on gcc.gnu.org regarding language/platform
compatibility :-(

> > What puzzels me, is the close dependancy between
> the
> > OS and the CPU for the GNARL.
> > e.g.
> > GNU/Linux x86 works
> > GNU/Linux Alpha don't work
> 
> Why should this puzzle you. The interface to the
> threads library is different in these two versions 
> of GNU/Linux, in particular, one is 32 bit and one
> is 64-bit.
OK, that makes sense.

What about Ada on GNU/Linux ARM? (its a 32 bit, too)


> It's easy enough to build a GNU/Linux Alpha version
> of GNAT, others have done it for sure.

So, we can add GNU/Linux Alpha to the list?


Bernd






__________________________________________________
Do you Yahoo!?
Faith Hill - Exclusive Performances, Videos & More
http://faith.yahoo.com

Re: Ada/GNAT runtime library?

[Andris Pavenis]

On Friday 18 October 2002 01:00, Robert Dewar wrote:
> <<does anyone knows for which OSs the GNAT runtime
> library works?
> Or even better, where it is documented?
>
> As far as I know, the DOS port was never completed.

I built also GNAT when building GCC-3.2 for DJGPP distribution.
(I used --target=i586-pc-msdosdjgpp, but built on Pentium III).
Only 2 small patches were needed, to get it build. Files are uploaded
for DJGPP distribution, but I haven't get any feedback about them,
so I don't know even how complete thay work, as I'm not using
ADA myself (some trivial test samples compiles and run, that's all
I have tested). I'm almost sure there are DOS caused limitations
(eg. tasking is unlikely to work, etc).

Off topic here, but I had to modify libstdc++-v3 headers for DJGPP in 
ports of gcc-3.0.X, and I made a terrible bug there (std::ofstream class 
was near completely broken), but no one else reported it for almost half a 
year. It was really reported by somebody else when I had already fixed that. 
I guess that ADA is less used than C++ with DJGPP, so it's even less likely 
to expect much feedback (unfortunatelly). 

So I'm afraid that DOS support may be slightly bitrotted. But without real 
feedback and users participation this bitrotting is perhaps going to continue.

Andris

Re: Ada/GNAT runtime library?

[Robert Dewar]

> Are there any documents about the 'abstraction level'?

I have no idea what this means, if you are asking about the levels of
interfaces, this is all self evident in the various package specs (Ada
documents all such interfaces more or less automatically :-)

Re: Ada/GNAT runtime library?

[Robert Dewar]

> Was DOS the reason for the 8.3 file names? 

Well DOS and all the formats (e.g. floppy and CD formats) that borrowed
this restriction, plus keeping file names short is convenient anways.
Actually gcc has largely maintained the 8+3 rule itself.

> Thanks for the list, Robert!
> Why not put it somewhere on gcc.gnu.org?

It's clear in the makefile, and publishing the list might make people think
that GNAT can be built for all these targets which is not the case (at least
we have not tried some of them).

> What puzzels me, is the close dependancy between the
> OS and the CPU for the GNARL.
> e.g.
> GNU/Linux x86 works
> GNU/Linux Alpha don't work

Why should this puzzle you. The interface to the threads library is
different in these two versions of GNU/Linux, in particular, one is
32 bit and one is 64-bit.

It's easy enough to build a GNU/Linux Alpha version of GNAT, others have
done it for sure.

Re: Ada/GNAT runtime library?

[Bernd Trog]

Robert Dewar <dewar@gnat.com> wrote:
> As far as I know, the DOS port was never completed.

Was DOS the reason for the 8.3 file names? 

> A partial list of targets for which runtimes exist
> are
[list]

Thanks for the list, Robert!
Why not put it somewhere on gcc.gnu.org?

What puzzels me, is the close dependancy between the
OS and the CPU for the GNARL.
e.g.
GNU/Linux x86 works
GNU/Linux Alpha don't work

Are there any documents about the 'abstraction level'?


Bernd






__________________________________________________
Do you Yahoo!?
Faith Hill - Exclusive Performances, Videos & More
http://faith.yahoo.com

Re: Ada/GNAT runtime library?

[Robert Dewar]

<See LIBGNAT_TARGET_PAIRS in ada/Makefile.in.
However, for the GCC 3.x back end, many of these will not work
yet. Targets with highest chance of success are x86/GNU Linux,
SPARC/Solaris and PA-RISC/HP-UX.
>>


NT/XP is also in good shape, but this is harder to build for all
sorts of reasons.

Re: Ada/GNAT runtime library?

[Robert Dewar]

<<does anyone knows for which OSs the GNAT runtime
library works?
Or even better, where it is documented?
>>

This is a confused question. There is no "the GNAT runtime", instead
the GNAT runtime is different for every system. Have a look at the
makefile to see what targets are supported.

<<DOS, Linux, OS2, OpenVMS, win32

Is this list complete?
>>

As far as I know, the DOS port was never completed.

The OS/2 port is by now rotting, and a lot of teh technology is
unfortunately way behind.

A partial list of targets for which runtimes exist are

IRIX
HPUX
Tru64
Solaris SPARC
Solaris x86 (but ACT has abandoned this port, so it will start to rot :-(
NT/XP (also works on WIN98/ME more or less)
OpenVMS Alpha
GNU/Linux x86
Interix (has a new name, I forget what)
AIX