Re: [announcement] SYSAPI and SYSSVC for Windows

["Ekkehard Morgenstern" <ekkehard.morgenstern@onlinehome.de>]

"Hyman Rosen" <hyrosen@mail.com> wrote:
> One of the evolutionary aspects of GNAT is its ability to provide
> clear and coherent error messages that direct the programmer's
> attention to the most likely problem in an incorrect program. This
> has come about through years of feedback, where users of GNAT report
> messages which are less than what they would like to see, and GNAT
> is augmented to handle those cases.

This a very good aspect of GNAT that I already realized when using
it. Before I can make a decision whether to port GNAT and the GCC,
I will have to take a look at its source. I will do that soon.

As I understand the GCC bootstrapping phase, I will have to write
a minimal C compiler that conforms to some rules, and then I can
compile with it the first stage of GCC, and with that the second,
and so on. And then I just need to recompile the GNU libraries
and the GNAT compiler.

However, I haven't looked at the GCC source for some years, so I
have to update myself on that first.

My only concern is, that I won't need all the stuff that comes
with the GCC, the GNU libraries and GNAT. 

The whole concept also makes assumptions about storage of source
files. I want to store all code and sources into a database
and read them from there. Since my database supports versioning,
the scope of file names is inappropriate. My compiler shall work
without files, since I want to use it to write the base of an
operating system. Since many of the GNU projects are file-based,
this might present a problem.

> Without examining GNAT's code, and without access to its test suite,
> you will not have the benefit of those feedbacks, and your error
> messages will be of low quality.

This might indeed be a problem. Perhaps my Ada compiler will only
be for writing the base code that can boot and run the operating
system, and then I can use a POSIX layer to support the GNU stuff,
including GNAT. So on my OS, you would still be able to use GNAT,
if you want to.

I'm using my own compiler-generation tools, and I will write new
ones for Ada, but the automated translation approach that I've 
chosen leaves little possibility for good error messages. In the
horrible first version of my compiler-generator, the source text
would just be accepted or rejected, resulting in a single error
message! ;)  -- of course I cannot uphold that for compilers that
are used by the general public. I will think of a better concept
for my compiler-creation tools that will allow better checks. :)