Help with Meridian compiler, please

Help with Meridian compiler, please

[Angel Alvarez]

The following simple program fails to compile with the PC Meridian compiler
(both 286 and 386):

procedure w is
  a: NATURAL range 0..1:=0;
begin
  null;
end w;

However, with the addition of one blank space, it compiles OK:

procedure w is
  a: NATURAL range 0..1 :=0;
-----------------------^
begin
  null;
end w;

Am I facing a (super) weak scanner? ... Am I missing something too obvious?
(anybody from Meridian out there?)

Thanks,
Angel Alvarez

DIT (B-312)
E.T.S.I. Telecomunicacion
Technical University of Madrid
E-28040 Madrid SPAIN

Re: Help with Meridian compiler, please

[Robert I. Eachus]

In article <1993Feb4.144224.7460@goya.uu.es> aalvarez@dit.upm.es (Angel Alvarez
) writes:

   >  a: NATURAL range 0..1:=0;

    Ooooh Ouch!  Tricky syntax analyzer test!  To make it harder use 

    A: Natural range 0..2:=0;

    The reason Meridan trips up (and yes it is a compiler bug) is
probably that 2:1101: is a legal numeric literal (with a decimal value
of 13).  I suspect that the compiler starts reading with the syntax of
an integer literal and when that fails backs up incorrectly, reading
the value 1 correctly but losing the colon.


					Robert I. Eachus

with Standard_Disclaimer;
use  Standard_Disclaimer;
function Message (Text: in Clever_Ideas) return Better_Ideas is...
--

					Robert I. Eachus

with Standard_Disclaimer;
use  Standard_Disclaimer;
function Message (Text: in Clever_Ideas) return Better_Ideas is...

Re: Help with Meridian compiler, please

[cis.ohio-state.edu!news.sei.cmu.edu!firth]

In article <EACHUS.93Feb5103840@dr_no.mitre.org> eachus@dr_no.mitre.org (Robert
 I. Eachus) writes:

>    A: Natural range 0..2:=0;
>
>    The reason Meridan trips up (and yes it is a compiler bug) is
>probably that 2:1101: is a legal numeric literal (with a decimal value
>of 13).

I think you're right.  This situation requires a two-character lookahead,
which is a problem with some scanners, especially machine-generated ones.
Also, unlike the almost-ambiguous 'X (attribute or character literal?),
the left syntactic context is no help.

The solution, of course, is to do the two-character lookahead.  My own
disgusting hack, on finding "2:=", is to put back just one character, a
synthetic lexical token that means ":=", and then we're back in sync.

Odd, though, that it should fail on the colon, having already solved
the identical problem of distinguishing between a range starting "0."
and a real constant starting "0.".

Re: Help with Meridian compiler, please

[darwin.sura.net!sgiblab!news.cs.indiana.edu!arizona.edu!east.pima.edu!rha]

In article <1993Feb4.144224.7460@goya.uu.es>, aalvarez@dit.upm.es 
(Angel Alvarez) writes:
> The following simple program fails to compile with the PC Meridian compiler
> (both 286 and 386):
> 
> procedure w is
>   a: NATURAL range 0..1:=0;
> begin
>   null;
> end w;

Never quite saw this one before!  The hint is in the actual error message
generated (which often is helpful to include when posting to such a wide
audience):  "illegal base in numeric literal".  Understanding the source of the
error is not trivial!  First, you must remember from LRM 2.10/3 that "sharp
characters (#) of a based literal can be replaced by colons (:)...".  The
compiler "thinks" you are about to supply a based literal as the endpoint of
the range, perhaps something like 16#FFF#, only you've "chosen" to substitute
colons like 16:FFF: .  The lexer probably doesn't look FIRST for the entire
syntactic construct for a based literal before deciding that it really IS what
you intended.

Next time I'm on the phone with them, I'll pass this along!
-----
Ray Harwood           |Data Basix           |Adjunct Faculty, East Campus,
Voice: (602)721-1988  |PO Box 18324         |   Pima Community College
FAX:   (602)721-7240  |Tucson, AZ 85731     |Instructor in Ada and Pascal
CompuServe: 76645,1370|AppleLink: DATA.BASIX|Internet: rharwood@east.pima.edu