Re: GNAT for an unsupported Unix-system?

[Chris Morgan <mihalis@ix.netcom.com>]

"Tor-Einar Jarnbjo" <bjote@cs.tu-berlin.de> writes:

> I recently downloaded the source-codes for the latest GNAT Ada-compiler,
> but then I saw that I needed the last version of GNAT to compile the
> new version of GNAT. Well, this results in an interesting problem when
> trying to port GNAT to a Unix-system for which there are no binaries.

You can build a cross-compiler easily enough for almost every known
Unix variant (and all important ones). That's how the existing ports
get done. I think GNAT started off life on SunOS and OS/2 and all the
other versions started off life as cross-compiles.

  
> Does anyone have a suggestion on how to solve this? E.g. BinProlog is
> written in Prolog, but has a small part written in C, making it possible
> to compile enough of the compiler with gcc, to make it able to compile
> itself. Is there a similar do-magic-thingy for GNAT, or perhaps another
> free Ada-compiler written in C?

There was a discussion about this on this very newsgroup a while
back. I suggested a subset language Tina (Tina Is Not Ada) which
contained just the features that the GNAT compiler itself needs and
which could be translated to C for the bootstrap problem. I even
speculated that this translation could be aided by a hacked GNAT
front-end which could traverse the tree representation of the compiler
and emit the C. Robert Dewar replied (as I remember) that they
considered schemes like this originally but found cross-compiling to
be a far better solution.

Regards,

Chris


-- 
Chris Morgan <mihalis@ix.netcom.com>