Date: Fri, 17 Mar 95 11:40:10 PST
From: KUBLER AT MASS DOT dnet DOT hac DOT com
To: ARPA:@sun.soe.clarkson.edu, hac2arpa DOT hac DOT com AT sun DOT soe DOT clarkson DOT edu
PP-Warning: Parse error in original version of preceding line
Original-To: ARPA::"djgpp AT sun DOT soe DOT clarkson DOT edu"@hac2arpa.hac.com

>Mike Feldman said:
>
>Well, Dewar should know better than to project deadlines.:-)
>
>In any case, anyone working with a compiler like this really has to be
>willing to work with a "work in progress", reporting bugs, etc. I don;t
>know which version you had, but I'll bet it was _many_ months back, if
>Dewar was projecting March 95 for validation.
>> 
>> I deleted all the GNAT modules on my DOS machine but I kept the DJGPP
>> support because I found I could do real work with C and C++. 
>
>Well, that is because you are dealing there with mature compilers.
>There are folks who are willing to get in on the ground floor of
>a work-in-progress, and there are folks who are not.:-)
>
>I delete each old version of GNAT as I get the newest one, roughly
>once a month. How many months old was yours?

It was (1994) December ~15th release.
I give much credit to those working on GNAT, its just that my disappointment
was proportional to my expectations of performance from a compiler
just three months from validation.   At that stage compilers should
pass 95% of validation tests with the remaining issues just a mop-up.

>> 
>> >Also, GNAT is about 250k lines of (mostly) Ada 95, (a bit of C around 
>> >the edges) so its writers get a lot of opportunity to test the compiler 
>> >on itself! GNAT has been compiling itself since July 1993, and its own
>> >sources contain more and more of the Ada 95 extensions with each new 
>> >release (new releases about once a month now).
>> 
>> But the compiler isn't multitasking(?), so that feature will never
>> be self-tested.  Not to mention representation_clause, etc, maybe 10%
>> of Ada95 will be tested.
>
>Excuse me? You're right that GNAT does not use tasking, but neither does
>_any_ compiler for _any_ language. My impression is that a good deal of
>rep-clause stuff is now implemented. GNAT is now passing well over 50%
>of validation tests, probably somewhere in the 70% range.
>
Right, so a self-compiling compiler would not test tasking.  Also I
expect the compiler to be as free as possible from machine dependencies,
so the compiler would not _use_ rep-clauses and thus would not test
rep-clauses when compiling itself.  Self-compilation was my issue.  I
acknowledge that the compiler is being tested in other ways.

>I'm curious - why are people willing to deal with unvalidated C/C++
>compilers but get very cynical about Ada validation? How many C++
>compilers do you know that really handle templates properly (assuming
>anyone can define "properly" properly..:-))

The C/C++ compilers may be unvalidated, but they pass my test by producing
code that computes good numbers, which I double-check. And the code is
portable, which is of practical value.  I have no language prejudice,  I'd
use Basic if I could get results faster.

>Bottom line: GNAT is a lot of fun to work with, if you are willing to
>deal with encountering and reporting bugs. I lurk on NYU's internal
>mailing list and have gotten lots of insight into how responsive they
>are. _Of course_ it is immature - I'm surprised anyone expected
>otherwise...

Well, _He_ said it was ready :-).

I am used to dealing with "works in progress", and can scale my inputs
and deal with the outputs, especially when I'm told at the outset what
the situation is.  Then I will not push the envelope with a production
type of Ada program.

Yes, I am cynical about Ada validation.  Take the Vax compiler (and don't
bring it back.)

Happy computing,

Douglas Kubler