3 testsuite failures under hpux 10.20

3 testsuite failures under hpux 10.20

[John David Anglin]

With yesterdays CVS source, the following failures occured in the testsuite:

Running /xxx/gnu/binutils-2.10.90/gas/testsuite/gas/hppa/parse/parse.exp ...
/xxx/gnu/binutils-2.10.90/gas/testsuite/gas/hppa/parse/linesepbug.s: Assembler messages:
/xxx/gnu/binutils-2.10.90/gas/testsuite/gas/hppa/parse/linesepbug.s:11: Fatal error: Unknown opcode: `$bar$,data'
FAIL: line separator bug
...
testcase /xxx/gnu/binutils-2.10.90/gas/testsuite/gas/hppa/parse/parse.exp completed in 20 seconds
Running /xxx/gnu/binutils-2.10.90/gas/testsuite/gas/hppa/reloc/reloc.exp ...
GAS LISTING /xxx/gnu/binutils-2.10.90/gas/testsuite/gas/hppa/reloc/blebug3.s 			page 1


   1              		.code
   2              		.align 4
   3              		.import yabba,code
   4              	
   5 0000 E4002000 	  	ble	R%yabba(%sr4,%r0)
   5      00000000 
PASS: blebug3.s: blebug3
PASS: Relocation reductions (part1)

a.out:     file format som

RELOCATION RECORDS FOR [$CODE$]:
OFFSET   TYPE              VALUE 
00000000 R_ENTRY           *ABS*+0x08000000
00000004 R_EXIT            *ABS*
00000004 R_DATA_PLABEL     foo
00000008 R_ENTRY           *ABS*+0x08000008
00000008 R_CODE_ONE_SYMBOL  $CODE$+0x00000004
0000000c R_CODE_ONE_SYMBOL  $CODE$+0x00000004
00000028 R_EXIT            *ABS*+0x00000010


FAIL: reduce.s: Test relocation reductions (part 2)
PASS: More relocation  reductions (part1)
...
../as-new   -o dump.o /xxx/gnu/binutils-2.10.90/gas/testsuite/gas/macros/strings.s
/xxx/gnu/binutils-2.10.90/gas/testsuite/../../binutils/objdump  -s -j .data dump.o > dump.out
extra regexps in /xxx/gnu/binutils-2.10.90/gas/testsuite/gas/macros/strings.d starting with "^Contents of section .data:$"
EOF from dump.out
FAIL: strings

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)

Re: 3 testsuite failures under hpux 10.20

[Alan Modra]

On Thu, 13 Jul 2000, John David Anglin wrote:

> FAIL: line separator bug
This one is a bug in the testsuite.  Directives need whitespace before
the `.' for LABELS_WITHOUT_COLONS targets.

> FAIL: reduce.s: Test relocation reductions (part 2)
This would appear to be a real gas bug.

> FAIL: strings
This is another testsuite bug.  For SOM, we need to look at $DATA$, not
.data

I'll have a go at fixing these later.

Regards, Alan Modra
-- 
Linuxcare.  Support for the Revolution.

Re: 3 testsuite failures under hpux 10.20

[John David Anglin]

Any luck in fixing the reduce.s bug?  I would guess the problem was
introduced in the last month or so since the version that I built
on June 17 passes this test.  However, I am puzzled why the other
two didn't show up before.  The strings test was fixed for a problem
similar to the line separator bug in June.

Regards,
Dave

> 
> On Thu, 13 Jul 2000, John David Anglin wrote:
> 
> > FAIL: line separator bug
> This one is a bug in the testsuite.  Directives need whitespace before
> the `.' for LABELS_WITHOUT_COLONS targets.
> 
> > FAIL: reduce.s: Test relocation reductions (part 2)
> This would appear to be a real gas bug.
> 
> > FAIL: strings
> This is another testsuite bug.  For SOM, we need to look at $DATA$, not
> .data
> 
> I'll have a go at fixing these later.
> 
> Regards, Alan Modra
> -- 
> Linuxcare.  Support for the Revolution.

-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)

Re: 3 testsuite failures under hpux 10.20

[Alan Modra]

On Tue, 18 Jul 2000, John David Anglin wrote:

> Any luck in fixing the reduce.s bug?

I haven't had time yet, David.  Too busy working on shared lib support for
elf32-hppa.

>  I would guess the problem was
> introduced in the last month or so since the version that I built
> on June 17 passes this test.

Hmm, that probably means it's something I broke, most likely when cleaning
up the code that handles arg relocs.  :-(  I'll have a look today, since
it seems likely I'm to blame.

>  However, I am puzzled why the other
> two didn't show up before.  The strings test was fixed for a problem
> similar to the line separator bug in June.

The strings test is relatively new, and the fix I checked in wasn't done
because I'd run the testsuite for SOM - It was more a case of noticing the
error by looking at the source.  It's unfortunate that hppa SOM support
in binutils needs HP-UX to compile...  Anyway, I now have access to a
HP-UX box, but I need to compile a few gnu packages, dejagnu at least,
before I can use it to painlessly run the testsuite.

-- 
Linuxcare.  Support for the Revolution.

Re: 3 testsuite failures under hpux 10.20

[John David Anglin]

> >  However, I am puzzled why the other
> > two didn't show up before.  The strings test was fixed for a problem
> > similar to the line separator bug in June.
> 
> The strings test is relatively new, and the fix I checked in wasn't done
> because I'd run the testsuite for SOM - It was more a case of noticing the
> error by looking at the source.  It's unfortunate that hppa SOM support
> in binutils needs HP-UX to compile...  Anyway, I now have access to a
> HP-UX box, but I need to compile a few gnu packages, dejagnu at least,
> before I can use it to painlessly run the testsuite.

I am 98% certain the June fix to strings seemed to work at the time.  I was
the one that reported the initial problem and I am fairly certain I tested
it since I never install binutils if there are any testsuite failures.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)

Re: 3 testsuite failures under hpux 10.20

[Alan Modra]

I'm checking this in to cure two of the failures dave@hiauly1.hia.nrc.ca
reported.  The easiest way to fix the SOM strings failure is just to
disable the test for SOM - properly fixing the test requires looking in
$DATA$ instead of .data

Alan Modra
-- 
Linuxcare.  Support for the Revolution.

gas/testuite/ChangeLog
	* gas/macros/macros.exp (test1): Run it for hppa*-*-linux*.
	(semi): Likewise.
	(strings): Don't run for hppa, except for hppa*-*-linux*.

	* gas/hppa/parse/linesepbug.s: Ensure we have whitespace before
	pseudos.

Index: gas/hppa/parse/linesepbug.s
===================================================================
RCS file: /cvs/src/src/gas/testsuite/gas/hppa/parse/linesepbug.s,v
retrieving revision 1.2
diff -u -p -r1.2 linesepbug.s
--- linesepbug.s	1999/08/30 20:09:00	1.2
+++ linesepbug.s	2000/07/19 04:37:27
@@ -8,5 +8,5 @@
 ; selectors to make sure they're handled correctly.
 
 foo:
-	.WORD 0 !.IMPORT $bar$,DATA
+	.WORD 0! .IMPORT $bar$,DATA
 
Index: gas/macros/macros.exp
===================================================================
RCS file: /cvs/src/src/gas/testsuite/gas/macros/macros.exp,v
retrieving revision 1.3
diff -u -p -r1.3 macros.exp
--- macros.exp	2000/06/20 13:52:30	1.3
+++ macros.exp	2000/07/19 04:37:29
@@ -1,6 +1,6 @@
 # Run some tests of gas macros.
 
-if ![istarget hppa*-*-*] {
+if { ![istarget hppa*-*-*] || [istarget *-*-linux*] } {
     run_dump_test test1
 }
 
@@ -19,11 +19,13 @@ if ![istarget *c54x*-*-*] {
 gas_test_error "err.s" "" "macro infinite recursion"
 
 case $target_triplet in {
-    { hppa*-*-* } { }
+    { hppa*-*-* } { if [istarget *-*-linux*] { run_dump_test semi } }
     { *c54x*-*-* } { }
     default {
 	run_dump_test semi
     }
 }
 
-run_dump_test strings
+if { ![istarget hppa*-*-*] || [istarget *-*-linux*] } {
+    run_dump_test strings
+}

Re: 3 testsuite failures under hpux 10.20

[Alan Modra]

Yep, it was a silly thinko, SOM instead of OBJ_SOM.  Committing the
following to fix it.

Regards, Alan Modra
-- 
Linuxcare.  Support for the Revolution.

gas/ChangeLog
	* config/tc-hppa.c (hppa_fix_adjustable): Correct LR%/RR% comment.
	(arg_reloc_stub_needed): #ifdef OBJ_SOM, not #ifdef SOM.
	(pa_type_args): Same here.

Index: gas/config/tc-hppa.c
===================================================================
RCS file: /cvs/src/src/gas/config/tc-hppa.c,v
retrieving revision 1.60
diff -u -p -r1.60 tc-hppa.c
--- tc-hppa.c	2000/07/15 12:57:14	1.60
+++ tc-hppa.c	2000/07/19 06:14:05
@@ -4390,7 +4390,7 @@ md_undefined_symbol (name)
   return 0;
 }
 
-#if defined (SOM) || defined (ELF_ARG_RELOC)
+#if defined (OBJ_SOM) || defined (ELF_ARG_RELOC)
 #define arg_reloc_stub_needed(CALLER, CALLEE) \
   ((CALLEE) && (CALLER) && ((CALLEE) != (CALLER)))
 #else
@@ -6715,7 +6715,7 @@ pa_type_args (symbolP, is_export)
 	  name = input_line_pointer;
 	  c = get_symbol_end ();
 	  arg_reloc = pa_align_arg_reloc (temp, pa_build_arg_reloc (name));
-#if defined (SOM) || defined (ELF_ARG_RELOC)
+#if defined (OBJ_SOM) || defined (ELF_ARG_RELOC)
 	  symbol_arg_reloc_info (symbolP) |= arg_reloc;
 #endif
 	  *input_line_pointer = c;
@@ -6729,7 +6729,7 @@ pa_type_args (symbolP, is_export)
 	  name = input_line_pointer;
 	  c = get_symbol_end ();
 	  arg_reloc = pa_build_arg_reloc (name);
-#if defined (SOM) || defined (ELF_ARG_RELOC)
+#if defined (OBJ_SOM) || defined (ELF_ARG_RELOC)
 	  symbol_arg_reloc_info (symbolP) |= arg_reloc;
 #endif
 	  *input_line_pointer = c;
@@ -8373,19 +8373,20 @@ hppa_fix_adjustable (fixp)
      eg. Suppose we have
      .		ldil	LR%foo+0,%r21
      .		ldw	RR%foo+0(%r21),%r26
-     .		ldw	RR%foo+10(%r21),%r25
+     .		ldw	RR%foo+4(%r21),%r25
 
-     If foo is at address 4090 (decimal) in section `sect', then after
-     reducing to the section symbol, we get
-     .			LR%sect+4090 == L%sect+0
-     .			RR%sect+4090 == R%sect+4090
-     .			RR%sect+4100 == R%sect-4092 (4100 - 8192)
-     and the last address loses.
-
-     Obviously, in cases where the LR% expression is identical to the
-     RR% one we will never have a problem, but is so happens that gcc
-     rounds addends involved in LR% field selectors to work around a
-     HP linker bug.  ie. We often have addresses like the last case
+     If foo is at address 4092 (decimal) in section `sect', then after
+     reducing to the section symbol we get
+     .			LR%sect+4092 == (L%sect)+0
+     .			RR%sect+4092 == (R%sect)+4092
+     .			RR%sect+4096 == (R%sect)-4096
+     and the last address loses because rounding the addend to 8k
+     mutiples takes us up to 8192 with an offset of -4096.
+
+     In cases where the LR% expression is identical to the RR% one we
+     will never have a problem, but is so happens that gcc rounds
+     addends involved in LR% field selectors to work around a HP
+     linker bug.  ie. We often have addresses like the last case
      above where the LR% expression is offset from the RR% one.  */
 
   if (hppa_fix->fx_r_field == e_lrsel

Re: 3 testsuite failures under hpux 10.20

[John David Anglin]

> Yep, it was a silly thinko, SOM instead of OBJ_SOM.  Committing the
> following to fix it.

The unexpected failures are now all fixed (or bypassed).

Thanks,
Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)