Return to henrybug CVS log

Up to [Research Unix] / researchv10no / cmd / PDP11 / 11as

Annotation of researchv10no/cmd/PDP11/11as/henrybug, revision 1.1.1.1

1.1       root        1: From zoo.toronto.edu!henry Thu Aug 16 13:39:15 EDT 1990
                      2: Received: by dutoit; Thu Aug 16 13:39:39 EDT 1990
                      3: Received: by inet.att.com; Thu Aug 16 13:39 EDT 1990
                      4: From: [email protected]
                      5: Date: Thu, 16 Aug 90 13:39:15 EDT
                      6: To: [email protected]
                      7: Subject: bug fix to 11as
                      8: 
                      9: Looks like I found a bug in the pdp11 cross-assembler.  When the second
                     10: pass sees a change of segment, it forces dot to be even.  Unfortunately,
                     11: the first pass is completely ignorant of this, and confusion and madness
                     12: result.  For example, this program:
                     13: 
                     14:        .data
                     15:        <a>
                     16:        .data
                     17: b:     <c>
                     18: 
                     19: puts the "c" at address 1, but makes "b" equal 2.  I ran into this with
                     20: a test program that flipped back and forth between data and text while
                     21: allocating strings.
                     22: 
                     23: I conjecture that the old native version of the assembler may have
                     24: done the forcing in both passes.  When I moved our major pdp11 stuff to
                     25: the cross-compiler, one of the things I had to do to get identical bytes
                     26: out the end was to add a ".even" in one place (after putting an odd-
                     27: length string into the text segment).  Evidently there was a difference
                     28: in behavior between native and cross versions, but I didn't investigate
                     29: it closely at the time.  I assumed that the native version had forced
                     30: alignment and the (newer) cross version had stopped doing so for some
                     31: reason, but it may well have been this bug manifesting itself.
                     32: 
                     33: Simply taking out the force-even code in the second pass appears to be
                     34: a workable fix (see enclosed patch).  It cures the problem in my test
                     35: program, and running all our major cross-compiled code through the new
                     36: version produces byte-for-byte-identical output, so it appears not to
                     37: be fouling up anything else.
                     38: 
                     39:                                          Henry Spencer at U of Toronto Zoology
                     40:                                           [email protected]   utzoo!henry
                     41: 
                     42: *** as26.c.old Thu Aug 16 13:27:38 1990
                     43: --- as26.c     Thu Aug 16 13:27:18 1990
                     44: ***************
                     45: *** 124,130 ****
                     46:        case T__TEXT:
                     47:        case T__DATA:
                     48:        case T__BSS:
                     49: !              *dot=(*dot+1)&~1;
                     50:                savdot[*dotrel-T_TEXT] = *dot;
                     51:                if (passno!=0 && t!=T__BSS) {
                     52:                        fseek(txtf,tseek[t-T__TEXT],0);
                     53: --- 124,130 ----
                     54:        case T__TEXT:
                     55:        case T__DATA:
                     56:        case T__BSS:
                     57: ! /*           *dot=(*dot+1)&~1;               */
                     58:                savdot[*dotrel-T_TEXT] = *dot;
                     59:                if (passno!=0 && t!=T__BSS) {
                     60:                        fseek(txtf,tseek[t-T__TEXT],0);
                     61: 

This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.