A few bugs in the 19981206 snapshot...

[Craig Burley]

...which I hope to pursue more seriously when I can use my office
again (construction noises make it hard to stay here for long, see
my .plan for details):

  -  Since the 19981026 (maybe 19981019) snapshot, use of -fsyntax-only
     crashes egcs-g77 on many of my personal test cases, including:

      subroutine f(a)
      character a*44
      character b*20
      character c*24

      b = 'aap noot mies '
      c = 'wim zus jet '

      a = b // c
      print *,a

      end

     (An old test from Toon.)

     This problem persists in the 19981206 snapshot.

  -  While debugging this, I noticed the line numbering is off for some
     source files, including gcc/varasm.c and another gcc file or two,
     but not for any of the gcc/f/*.c files that happened to be on the
     backtrace at the time (I checked).  That is, at the crash site, the
     source display in Emacs (using a "stock" RH 5.2 Linux system FWIW)
     is off by some number of lines (verified to be 4 in one file at
     one point, 14 in varasm.c at another, IIRC) for some gcc/*.c files.
     It isn't immediately obvious why, or who's the culprit.  (Built
     via bootstrap, as usual for me.)

  -  Though it might be due to the bad line numbering, I'm having a heck
     of a time tracking down the source of the crash, since the fprintf
     of "\t%s\n", which appears to provide no argument for the %s, has
     proven hard to track down by my looking at the relevant gcc/config/
     files.  The line numbering (adjusting as best I can) seems to suggest
     the offending macro is ASM_OUTPUT_ALIGNED_LOCAL, but I can't find an
     effective definition of this offhand.

Clearly, I need to spend some quality time looking at preprocessor outputs,
maybe -g -S outputs, and so on.  But since I won't get to that until,
at least, tonight, I thought I'd mention these problems in case they
rang bells with anyone else!

        tq vm, (burley)