Conference vaxaxp::vmsnotes

537.0. "SDL incorrect for ESDF$T_NAME and ESDF$S_NAME?" by GIDDAY::GILLINGS (a crucible of informative mistakes) Mon Apr 28 1997 03:21

  A customer is writing PASCAL programs which produce Alpha object code.
  He has found that long symbol names cause array bounds errors when accessing
  the ESDF$T_NAME field. According to the OpenVMS Linker Utility Manual:

"B.3.2.1  GSD Subrecord for a Global Symbol Definition
               (EGSD$C_SYM with EGSY$V_DEF Set)
...

   SYMBOL NAME                             Name: ESDF$T_NAME
                                           Length: Variable, 1 to 64 bytes

   This field contains the symbol name in ASCII format."


  However, the Pascal, MACRO and Bliss definitions are for 31 characters
  only:
            ESDF$L_PSINDX : UNSIGNED;   (*Owning psect number               *)
            ESDF$B_NAMLNG : $UBYTE;     (*Length of name                    *)
            ESDF$T_NAME : PACKED ARRAY [1..31] OF CHAR; (*Symbol name       *)

$EQU    ESDF$B_NAMLNG   32
$EQU    ESDF$S_NAME     31
$EQU    ESDF$T_NAME     33

macro ESDF$B_NAMLNG = 32,0,8,0 %;       ! Length of name
macro ESDF$T_NAME = 33,0,0,0 %;
literal ESDF$S_NAME = 31;               ! Symbol name
  
  (I suspect other languages have a similar definition but haven't gone
  searching).

  Now, given that this subtype is overlayed with other subtypes which are
  plenty large enough to accommodate a 64 byte field, MACRO or Bliss programs
  might no be concerned with the apparently incorrect definition of 
  ESDF$S_NAME, but in Pascal, processing a symbol longer than 31 characters
  causes array bounds violations.

  I'm not sure where the customer is finding these long symbols, but they're
  causing trouble. Redefining the field as PACKED ARRAY [1..64] works as
  expected. 

  It seems likely that this is a bug in the SDL definition, but I can't
  figure out how to disassemble STARLETSD.TLB to prove it.

  Can anyone confirm that this should be QARed? OpenVMS/Alpha V6.1 and V7.1

						John Gillings, Sydney CSC

    re .0
    
    C and FORTRAN both define the symbol name as having length 31. (Alpha
    V6.2) I suggest you QAR it.
    
    (This could reflect a VAX limit that just hasn't been updated.?)