note from Professor Knuth - Gpc

1 May 2000


      Dear GPC folks,
I just installed GPC 2.8.1 on my new Linux machine, which is replacing
the Sparc that I've been using since 1989. Naturally I want to be able
to achieve on my new machine what I could do on my old one; thus it is
important to me that I can still compile the absolutely official
Pascal-based version of TeX, even though everybody [including me]
uses the Web2C-translated version for production runs.
So I began today to look at those old programs and to remind myself
about Pascal syntax. When TeX was being developed in the late 70s
and early 80s, we chose Pascal for TeX82 because it was the most
widely understood language.  People told me that the Pascal versions
of TeX and METAFONT were ported to more than 100 different platforms.
I myself used Pascal-H on a TOPS10 machine, later Berkeley Pascal
on an early version of SunOS, later a much better Sun Pascal that
was developed I think in Russia.
Since these programs were much larger than typical student-oriented
Pascal source code, they tended to find at least one bug in every new
Pascal compiler they faced. "New users find new bugs."
Sure enough, it seems GNU Pascal is no exception... I believe your
otherwise-admirable program is actually otherwise-flaky, in the sense that
it fails to parse the "otherwise" keyword for case-statement defaults when
the case statement in question includes another (nested) case statement.
For example, gpc does not parse the following program:
---------- cut here for bug.p -------------------------------------
program bug(input,output);
var x : integer;
begin readln(x);
   case x mod 2 of
     0 : begin case x mod 4 of
           2 : writeln('oddly even');
           0 : writeln('evenly even');
         end; { case }
         end;
     otherwise writeln('odd');
   end; { case }
end.
-------------------------------------------------------------------
On the other hand it handles a similar program with ease:
---------- cut here for nobug.p -----------------------------------
program bug(input,output);
var x : integer;
begin readln(x);
   case x mod 2 of
     0 : begin if (x mod 4)=2 then writeln('oddly even')
           else writeln('evenly even');
         end;
     otherwise writeln('odd');
   end; { case }
end.
--------------------------------------------------------------------
To work around this, I had to change nested case statements
into programs that used only if-then-else.
Another problem I had was a segmentation fault on
  F_Flush(Output)
after defining
  procedure F_Flush(var aFile); asmname '_p_f_flush';
as recommended in your documentation. I didn't have time to pursue
this to figure out what was failing, because it was easier for me
just to say
  procedure Flushstdout; external;
and then load the C routine
  void Flushstdout(){fflush(stdout)}
with the compiled object code.
With these two changes, I successfully got the TANGLE program to
reproduce itself, so the first major hurdle towards generating
retro-TeX has been passed! If I run into any more bugs, I'll
let you know --- the TRIP and TRAP tests of TeX and METAFONT are
pretty good test suites.
But let's be optimistic and suppose that everything else works.
Then I won't have to write you again.
In any case when the new GNU-Pascal-based port of TeX is ready I will post
the sources to the CTAN archives; they will be at labrea.stanford.edu in
directory ~ftp/pub/tex/local.
Best wishes on your important project! -- Don Knuth