Improving P2Ada

Improving P2Ada

[Gautier]

Hi everybody. I'm about to improve
the P2Ada tool in order to translate Pascal
sources from the Borland constellation.
That should be feasible...

The point is that the yacc/aflex compiler
(a gnat/win32 version found on the net)
produces a code that hangs or does nothing
- it seems at initialisation.
The grammar is inchanged from M. Carlisle's
sources (v 2.2).

If anyone knows about the "dangerous curves" of
yacc/aflex, let me know...

-- 
Gautier

--------
http://members.xoom.com/gdemont/

Re: Improving P2Ada

[Gautier]

Thanks for replies...
The P2Ada way doesn't seem the only one.
Though, here is what I would/will add to current P2Ada (2.2) to
let it translate happily Borland-style sources:

  * no forced order in CONST, TYPE, VAR, PROCEDURE/FUNCTION clauses
    (like Ada 95) - most TP programs aren't "Pascal orthodox" on this
    point (fortunately), so P2Ada fails there...
  * CASE statements can have an ELSE part ( when others=> in Ada )
  * there are ASM..END blocks like BEGIN..END (just comment out the
    assembler opcodes...)
  * there are SHL & SHR binary operators ( shift_left(.,.) in Ada )
  * strings can have explicit ASCII characters: 'Blabla'#13#10're-blabla'
    that would become "Blabla" & character'val(13) & ...

A later but more "banana-skinny" step is the OO part; to begin with...

  TYPE xyz = OBJECT         ->  type xyz is tagged record
  TYPE xyz = OBJECT(abc)    ->  type xyz is new abc with record
  PROCEDURE xyz.toto;       ->  procedure toto(prefixed_obj: xyz) is

G.

Improving P2Ada (2)

[Gautier]

Hi. Here is a new (very provisory) version that
allows translation of many (not all) Turbo/Borland
Pascal sources (no OO yet but it will come...).

  ftp://ftp.unine.ch/incoming/gautier/newp2ada.zip

The archive contains all you need to (re-)build P2Ada
and customize further...

Gautier

---

Enhancement list from 1997 version
==================================

26.VIII.1999 [GdM]

a) Borland-ish add-ons:

- Str translated ( bug: should be put(s,123) instead of put(123,s) )
- ' character, written '' ( bug: should trash the 2nd ' )
- arbitrary order of declaration kinds, instead of label-const-type-var-proc
- typeless var parameters
- DIV becomes /
- SHL & SHR operators ( bug: translated by "a Shift_left( ; ) b" )
- programs not beginning by 'Program' !
- global initialized VARiables - in the CONSTant parts - ; aggregates
- strings with contraints
- typeless files
- else part in case..of
- asm...end blocks ( bug: not complete)
- special directives in proc/func decl. (like near/far/assembler)
- characters with explicit ascii (#223)
- borland string constants ('aaa'#123'bbb')

b) Improvements or bugfixes (any Pascal)
- begin/end simplified when safe (then/else/do/case)
- outputs "Null;" for empty statement
- outputs "WHEN others=> Null;" when no else part

Re: Improving P2Ada (2)

[Gautier]

> >- outputs "WHEN others=> Null;" when no else part

> I know this is the correct translation, but would it be
> possible to have a flag that disables this feature?

It is - by enclosing the Text_IO.Put(...) in pascal.y
with an if flag then...end if;
But is it not complicating uselessly the tool ?

My arguments are:

- it's easier to remove a "WHEN others=> Null;" than
  to add it
- the cases when "WHEN others=> Null;" is needed are
  much more frequent than when it isn't

-- 
Gautier

--------
http://members.xoom.com/gdemont/

Re: Improving P2Ada (2)

[Gautier]

> It appears that I have managed to argue that the flag isn't
> necessary after all. ;-(

Nice he he he. Hum, I prefer to wait a while to see
others' opinions.
Provisorily for your tests you could replace at line 1016 of pascal.y

    | {pascalhelp.put_line("WHEN OTHERS=> Null; ");}   -- EMPTY

with:

    |  -- EMPTY

BTW the "null;" for empty statements is not avoided for the case
of an empty statement appearing in a sequence, like

if abc then begin
  x;
  y; { empty statement after y ! }
end;

(NB: the ';' is a separator in Pascal) so there will
be protests, too... A way reprogram that can be the rules

- a single statement is either empty (crac: issue "null;") or not;
- a statement-in-sequence is either empty (don't issue "null;" !) or not;
- a sequence is either empty (crac: issue "null;") or not;

... with the risk to confuse ayacc a bit more.

-- 
Gautier

--------
http://members.xoom.com/gdemont/