current binutils trunk failures on cygwin

current binutils trunk failures on cygwin

[Christian Joensson]

well, just a ping here...

Windows XP Pro/SP2 cygwin Pentium M processor 2.13GHz system with packages:

binutils             20050610-1     2.16.91 20050610
bison                2.1-1          2.1
cygwin               1.5.19-4
dejagnu              20021217-2     1.4.2.x
expect               20030128-1     5.26
gcc                  3.4.4-1
gmp                  4.1.4-2
make                 3.80-1         OK

tcltk                20030901-1     8.4

LAST_UPDATED: Fri Apr 28 18:16:17 UTC 2006

Native configuration is i686-pc-cygwin

		=== binutils tests ===


Running target unix
ERROR: /usr/local/src/trunk/src/binutils/testsuite/binutils-all/copy-1.s:
assembly failed
UNRESOLVED: copy with seting section flags 1
FAIL: copy with seting section flags 2

		=== binutils Summary ===

# of expected passes		73
# of unexpected failures	1
# of expected failures		1
# of unresolved testcases	1
		=== gas tests ===


Running target unix

		=== gas Summary ===

# of expected passes		108
# of expected failures		1
		=== ld tests ===


Running target unix
FAIL: cdtest with -Ur
XPASS: align1
FAIL: weak symbols

		=== ld Summary ===

# of expected passes		39
# of unexpected failures	2
# of unexpected successes	1
# of expected failures		3
# of unsupported tests		1


Any comments on the background of these failures? Would you like me to
provide the .log file information too of the testsuite runs?

--
Cheers,

/ChJ

Re: current binutils trunk failures on cygwin

[Christian Joensson]

On 4/28/06, Christian Joensson <christian.joensson@gmail.com> wrote:

>                 === binutils tests ===
>
>
> Running target unix
> ERROR: /usr/local/src/trunk/src/binutils/testsuite/binutils-all/copy-1.s:
> assembly failed
> UNRESOLVED: copy with seting section flags 1
> FAIL: copy with seting section flags 2

still there....

--
Cheers,

/ChJ

Re: current binutils trunk failures on cygwin

[H. J. Lu]

On Mon, May 01, 2006 at 02:43:28PM +0200, Christian Joensson wrote:
> On 4/28/06, Christian Joensson <christian.joensson@gmail.com> wrote:
> 
> >                === binutils tests ===
> >
> >
> >Running target unix
> >ERROR: /usr/local/src/trunk/src/binutils/testsuite/binutils-all/copy-1.s:
> >assembly failed
> >UNRESOLVED: copy with seting section flags 1

This test should be ELF only. I will check in this patch.

> >FAIL: copy with seting section flags 2
> 

objcopy --set-section-flags .text=alloc,data

doesn't work on COFF. Someone needs to fix it or I can make it ELF
only.


H.J.
---
2006-05-01  H.J. Lu  <hongjiu.lu@intel.com>

	* binutils-all/objcopy.exp: Run "copy-1" for ELF only.

--- binutils-all/objcopy.exp.elf	2006-04-26 07:57:54.000000000 -0700
+++ binutils-all/objcopy.exp	2006-05-01 07:02:24.000000000 -0700
@@ -723,7 +723,7 @@ if { ([istarget "ia64-*-elf*"]
 if [is_elf_format] {
     objcopy_test "ELF unknown section type" unknown.s
     objcopy_test_readelf "ELF group" group.s
+    run_dump_test "copy-1"
 }
 
-run_dump_test "copy-1"
 run_dump_test "copy-2"

Re: current binutils trunk failures on cygwin

[Christian Joensson]

On 5/1/06, H. J. Lu <hjl@lucon.org> wrote:

> This test should be ELF only. I will check in this patch.
>
> > >FAIL: copy with seting section flags 2
> >
>
> objcopy --set-section-flags .text=alloc,data
>
> doesn't work on COFF. Someone needs to fix it or I can make it ELF
> only.

Right, what we have now is this:

LAST_UPDATED: Mon May  1 14:33:15 UTC 2006

Native configuration is i686-pc-cygwin

                === binutils tests ===


Running target unix
FAIL: copy with seting section flags 2

                === binutils Summary ===

# of expected passes            73
# of unexpected failures        1
# of expected failures          1

--
Cheers,

/ChJ

Re: current binutils trunk failures on cygwin

[Nick Clifton]

[-- Attachment #1: Type: text/plain, Size: 1163 bytes --]

Hi H.J.

>>> FAIL: copy with seting section flags 2
> 
> objcopy --set-section-flags .text=alloc,data
> 
> doesn't work on COFF. Someone needs to fix it or I can make it ELF
> only.

It does work with COFF targets (eg alpha-linuxecoff).  The problem is PE 
based targets where there is code to ensure that certain named sections 
(eg .text) are given specific section flags.  This is in order to comply 
with the PE spec, so really this is not a bug, but a feature.

So I am going to apply the attached patch to resolve this problem by 
changing the name of the section involved from .text to foo.  Actually 
it does not resolve the problem for AOUT based targets since they do not 
support arbitrarily named sections, so the test is just marked as 
unsupported for them.  But I think that skipping AOUT targets is better 
than skipping PE targets.

Cheers
   Nick

binutils/testsuite/ChangeLog
2006-05-02  Nick Clifton  <nickc@redhat.com>

	* binutils-all/copy-2.d: Change the name of the section whose
	flags are changed to "foo" so that the test will work with PE
	based targets.  Skip this test for AOUT based targeted.
	* binutils-all/copytest.s: New file.


[-- Attachment #2: copy-2.d.patch --]
[-- Type: text/x-patch, Size: 1482 bytes --]

Index: binutils/testsuite/binutils-all/copy-2.d
===================================================================
RCS file: /cvs/src/src/binutils/testsuite/binutils-all/copy-2.d,v
retrieving revision 1.1
diff -c -3 -p -r1.1 copy-2.d
*** binutils/testsuite/binutils-all/copy-2.d	26 Apr 2006 13:37:05 -0000	1.1
--- binutils/testsuite/binutils-all/copy-2.d	2 May 2006 12:12:55 -0000
***************
*** 1,14 ****
  #PROG: objcopy
  #objdump: -h
! #objcopy: --set-section-flags .text=alloc,data
  #name: copy with seting section flags 2
! #source: bintest.s
  
  .*: +file format .*
  
  Sections:
  Idx.*
  #...
!   [0-9]* .text.*
!                   CONTENTS, ALLOC, LOAD, RELOC, DATA
  #...
--- 1,18 ----
  #PROG: objcopy
  #objdump: -h
! #objcopy: --set-section-flags foo=contents,alloc,load,code
  #name: copy with seting section flags 2
! #source: copytest.s
! #not-target: *-*-aout
! # Note - we use copytest.s and a section named "foo" rather
! # than .text because for some file formats (eg PE) the .text
! # section has a fixed set of flags and these cannot be changed.
  
  .*: +file format .*
  
  Sections:
  Idx.*
  #...
!   [0-9]* foo.*
!                   CONTENTS, ALLOC, LOAD, CODE
  #...
*** /dev/null	2006-05-02 09:00:29.883801480 +0100
--- binutils/testsuite/binutils-all/copytest.s	2006-05-02 12:47:30.000000000 +0100
***************
*** 0 ****
--- 1,7 ----
+ 	.globl foo_symbol
+ 	.section foo
+ foo_symbol:	
+ 	.long	1
+ 	.section bar
+ bar_symbol:
+ 	.long	2

Re: current binutils trunk failures on cygwin

[H. J. Lu]

On Tue, May 02, 2006 at 01:16:21PM +0100, Nick Clifton wrote:
> Hi H.J.
> 
> >>>FAIL: copy with seting section flags 2
> >
> >objcopy --set-section-flags .text=alloc,data
> >
> >doesn't work on COFF. Someone needs to fix it or I can make it ELF
> >only.
> 
> It does work with COFF targets (eg alpha-linuxecoff).  The problem is PE 
> based targets where there is code to ensure that certain named sections 
> (eg .text) are given specific section flags.  This is in order to comply 
> with the PE spec, so really this is not a bug, but a feature.
> 
> So I am going to apply the attached patch to resolve this problem by 
> changing the name of the section involved from .text to foo.  Actually 
> it does not resolve the problem for AOUT based targets since they do not 
> support arbitrarily named sections, so the test is just marked as 
> unsupported for them.  But I think that skipping AOUT targets is better 
> than skipping PE targets.
> 

I chose .text on purpose since we didn't say --set-section-flags
doesn't work on certain sections. copy-1.d already has a non-standard
section. I would prefer to add a new test and make copy-2.d ELF only.

H.J.

Re: current binutils trunk failures on cygwin

[Nick Clifton]

Hi H. J.

> I chose .text on purpose since we didn't say --set-section-flags
> doesn't work on certain sections. copy-1.d already has a non-standard
> section. I would prefer to add a new test and make copy-2.d ELF only.

I am OK with adding the new test, but copy-2.d should be non-PE targets 
only, not just ELF only.  It is only PE targets which have the problem.

Cheers
   Nick

Re: current binutils trunk failures on cygwin

[H. J. Lu]

On Tue, May 02, 2006 at 04:31:40PM +0100, Nick Clifton wrote:
> Hi H. J.
> 
> >I chose .text on purpose since we didn't say --set-section-flags
> >doesn't work on certain sections. copy-1.d already has a non-standard
> >section. I would prefer to add a new test and make copy-2.d ELF only.
> 
> I am OK with adding the new test, but copy-2.d should be non-PE targets 
> only, not just ELF only.  It is only PE targets which have the problem.
> 

Does COFF have the same issue? Here is a patch.


H.J.
---
2006-05-02  H.J. Lu  <hongjiu.lu@intel.com>

	* binutils-all/copy-3.d: New.

	* objcopy.exp: Run copy-3.

--- binutils/testsuite/binutils-all/copy-3.d.text	2006-05-02 09:36:37.000000000 -0700
+++ binutils/testsuite/binutils-all/copy-3.d	2006-05-02 09:48:33.000000000 -0700
@@ -0,0 +1,17 @@
+#PROG: objcopy
+#objdump: -h
+#objcopy: --set-section-flags .text=alloc,data
+#name: copy with setting section flags 3
+#source: bintest.s
+#not-target: *-*-*pe* *-*-*coff i*86-*-cygwin* i*86-*-mingw32*
+# The .text # section in PE/COFF has a fixed set of flags and these
+# cannot be changed.  We skip it for them.
+
+.*: +file format .*
+
+Sections:
+Idx.*
+#...
+  [0-9]* .text.*
+                  CONTENTS, ALLOC, LOAD, RELOC, DATA
+#...
--- binutils/testsuite/binutils-all/objcopy.exp.text	2006-05-01 11:04:15.000000000 -0700
+++ binutils/testsuite/binutils-all/objcopy.exp	2006-05-02 09:44:32.000000000 -0700
@@ -727,3 +727,4 @@ if [is_elf_format] {
 }
 
 run_dump_test "copy-2"
+run_dump_test "copy-3"

Re: current binutils trunk failures on cygwin

[H. J. Lu]

On Tue, May 02, 2006 at 09:53:21AM -0700, H. J. Lu wrote:
> On Tue, May 02, 2006 at 04:31:40PM +0100, Nick Clifton wrote:
> > Hi H. J.
> > 
> > >I chose .text on purpose since we didn't say --set-section-flags
> > >doesn't work on certain sections. copy-1.d already has a non-standard
> > >section. I would prefer to add a new test and make copy-2.d ELF only.
> > 
> > I am OK with adding the new test, but copy-2.d should be non-PE targets 
> > only, not just ELF only.  It is only PE targets which have the problem.
> > 
> 
> Does COFF have the same issue? Here is a patch.
> 
> 

I am going to check in this patch.


H.J.
---
2006-05-02  H.J. Lu  <hongjiu.lu@intel.com>

	* binutils-all/copy-3.d: New.

	* objcopy.exp: Run copy-3.

--- binutils/testsuite/binutils-all/copy-3.d.text	2006-05-02 09:36:37.000000000 -0700
+++ binutils/testsuite/binutils-all/copy-3.d	2006-05-03 07:08:50.000000000 -0700
@@ -0,0 +1,17 @@
+#PROG: objcopy
+#objdump: -h
+#objcopy: --set-section-flags .text=alloc,data
+#name: copy with setting section flags 3
+#source: bintest.s
+#not-target: *-*-aout *-*-*pe* *-*-*coff i*86-*-cygwin* i*86-*-mingw32*
+# The .text # section in PE/COFF has a fixed set of flags and these
+# cannot be changed.  We skip it for them.
+
+.*: +file format .*
+
+Sections:
+Idx.*
+#...
+  [0-9]* .text.*
+                  CONTENTS, ALLOC, LOAD, RELOC, DATA
+#...
--- binutils/testsuite/binutils-all/objcopy.exp.text	2006-05-01 11:04:15.000000000 -0700
+++ binutils/testsuite/binutils-all/objcopy.exp	2006-05-02 09:44:32.000000000 -0700
@@ -727,3 +727,4 @@ if [is_elf_format] {
 }
 
 run_dump_test "copy-2"
+run_dump_test "copy-3"

Re: current binutils trunk failures on cygwin

[Nick Clifton]

Hi H. J.

>>> I chose .text on purpose since we didn't say --set-section-flags
>>> doesn't work on certain sections. copy-1.d already has a non-standard
>>> section. I would prefer to add a new test and make copy-2.d ELF only.
>> I am OK with adding the new test, but copy-2.d should be non-PE targets 
>> only, not just ELF only.  It is only PE targets which have the problem.
> 
> Does COFF have the same issue?

No - it is only PE that has the issue.

> Here is a patch.

> +#not-target: *-*-*pe* *-*-*coff i*86-*-cygwin* i*86-*-mingw32*

You do not need the *-*-coff here.  In fact it should not be there.

Cheers
   Nick

Re: current binutils trunk failures on cygwin

[H. J. Lu]

On Wed, May 03, 2006 at 04:01:33PM +0100, Nick Clifton wrote:
> Hi H. J.
> 
> >>>I chose .text on purpose since we didn't say --set-section-flags
> >>>doesn't work on certain sections. copy-1.d already has a non-standard
> >>>section. I would prefer to add a new test and make copy-2.d ELF only.
> >>I am OK with adding the new test, but copy-2.d should be non-PE targets 
> >>only, not just ELF only.  It is only PE targets which have the problem.
> >
> >Does COFF have the same issue?
> 
> No - it is only PE that has the issue.
> 
> >Here is a patch.
> 
> >+#not-target: *-*-*pe* *-*-*coff i*86-*-cygwin* i*86-*-mingw32*
> 
> You do not need the *-*-coff here.  In fact it should not be there.

But it failed on i386-coff the same way as i386-cygwin:

Executing on host:
/export/build/gnu/binutils-import-cross/build-i386-coff/gas/as-new
/export/gnu/src/binutils-import/binutils/binutils/testsuite/binutils-all/bintest.s
-o tmpdir/bintest.o    (timeout = 300)
/export/build/gnu/binutils-import-cross/build-i386-coff/binutils/objcopy
--set-section-flags .text=alloc,data tmpdir/bintest.o tmpdir/copy.o
Executing on host:
/export/build/gnu/binutils-import-cross/build-i386-coff/binutils/objcopy
--set-section-flags .text=alloc,data tmpdir/bintest.o tmpdir/copy.o
(timeout = 300)
/export/build/gnu/binutils-import-cross/build-i386-coff/binutils/objdump
-h tmpdir/copy.o > tmpdir/dump.out
regexp_diff match failure
regexp "^                  CONTENTS, ALLOC, LOAD, RELOC, DATA$"
line   "                  CONTENTS, ALLOC, LOAD, RELOC, CODE"
FAIL: copy with setting section flags 3 


H.J.

RE: current binutils trunk failures on cygwin

[Dave Korn]

On 03 May 2006 16:02, Nick Clifton wrote:


>> Does COFF have the same issue?
> 
> No - it is only PE that has the issue.
> 
>> Here is a patch.
> 
>> +#not-target: *-*-*pe* *-*-*coff i*86-*-cygwin* i*86-*-mingw32*
> 
> You do not need the *-*-coff here.  In fact it should not be there.

  Say, isn't that first pattern going to inadvertently match *-*-openbsd,
*-*-mpeix, *-*-openedition,  *-*-openstep, *-*-superux?  (This based on a
quick grepping through top-level config rather than detailed knowledge of
which platforms are supported by binutils).

  Should it perhaps be just "*-*-pe*", which matches the test in config.sub?
I can't see any PE-coff targets that have anything prefixing 'pe' in the third
part of the triplet.


    cheers,
      DaveK
-- 
Can't think of a witty .sigline today....

Re: current binutils trunk failures on cygwin

[H. J. Lu]

On Wed, May 03, 2006 at 05:02:16PM +0100, Dave Korn wrote:
> On 03 May 2006 16:02, Nick Clifton wrote:
> 
> 
> >> Does COFF have the same issue?
> > 
> > No - it is only PE that has the issue.
> > 
> >> Here is a patch.
> > 
> >> +#not-target: *-*-*pe* *-*-*coff i*86-*-cygwin* i*86-*-mingw32*
> > 
> > You do not need the *-*-coff here.  In fact it should not be there.
> 
>   Say, isn't that first pattern going to inadvertently match *-*-openbsd,
> *-*-mpeix, *-*-openedition,  *-*-openstep, *-*-superux?  (This based on a
> quick grepping through top-level config rather than detailed knowledge of
> which platforms are supported by binutils).
> 
>   Should it perhaps be just "*-*-pe*", which matches the test in config.sub?
> I can't see any PE-coff targets that have anything prefixing 'pe' in the third
> part of the triplet.
> 

It should be "*-*-pe".


H.J.