test failures building perl on Cygwin - continued

test failures building perl on Cygwin - continued

[Gerrit P. Haase]

On Tue, 16 Oct 2001 11:36:33 -0400, you wrote:

John,

>I have not had time to investigate; I'm still trying to work on the 
>CygWin core's with PERL_DESTRUCT_LEVEL=2.  

I'm getting the same problems here with perl-5.6.1 if built with
debugging, so it is probably a change that was made to both, bleadperl
and maintperl or a change in cygwin.

Gerrit

-- 
=^..^=
bIg BrOtHeR iS wAtChInG YoU!

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on Cygwin - continued

[John Peacock]

"Gerrit P. Haase" wrote:
> 
> On Tue, 16 Oct 2001 11:36:33 -0400, you wrote:
> 
> John,
> 
> >I have not had time to investigate; I'm still trying to work on the
> >CygWin core's with PERL_DESTRUCT_LEVEL=2.
> 
> I'm getting the same problems here with perl-5.6.1 if built with
> debugging, so it is probably a change that was made to both, bleadperl
> and maintperl or a change in cygwin.
> 
> Gerrit

I am positive (no really) that I was getting clean tests in July, so I 
would have to think that it was a change in CygWin as opposed to Perl.  
I'll try and go back to 5.6.0 and build for debugging and see if it 
happens there too.  That would pretty much rule out Perl itself as 
the source.

John

-- 
John Peacock
Director of Information Research and Technology
Rowman & Littlefield Publishing Group
4720 Boston Way
Lanham, MD 20706
301-459-3366 x.5010
fax 301-429-5747

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on Cygwin - continued

[Gerrit P. Haase]

On Wed, 17 Oct 2001 11:10:21 -0400, you wrote:

>"Gerrit P. Haase" wrote:
>> On Tue, 16 Oct 2001 11:36:33 -0400, you wrote:
>> 
>> >I have not had time to investigate; I'm still trying to work on the
>> >CygWin core's with PERL_DESTRUCT_LEVEL=2.
>> 
>> I'm getting the same problems here with perl-5.6.1 if built with
>> debugging, so it is probably a change that was made to both, bleadperl
>> and maintperl or a change in cygwin.
>
>I am positive (no really) that I was getting clean tests in July, so I 
>would have to think that it was a change in CygWin as opposed to Perl.  
>I'll try and go back to 5.6.0 and build for debugging and see if it 
>happens there too.  That would pretty much rule out Perl itself as 
>the source.

Well, maybe there were two little changes, one here, one there, one
opens a door, another one stumbles in...

Gerrit

-- 
=^..^=
bIg BrOtHeR iS wAtChInG YoU!

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on Cygwin - continued

[John Peacock]

"Gerrit P. Haase" wrote:
> 
> On Wed, 17 Oct 2001 11:10:21 -0400, you wrote:
> 
> >"Gerrit P. Haase" wrote:
> >> On Tue, 16 Oct 2001 11:36:33 -0400, you wrote:
> >>
> >> >I have not had time to investigate; I'm still trying to work on the
> >> >CygWin core's with PERL_DESTRUCT_LEVEL=2.
> >>
> >> I'm getting the same problems here with perl-5.6.1 if built with
> >> debugging, so it is probably a change that was made to both, bleadperl
> >> and maintperl or a change in cygwin.
> >
> >I am positive (no really) that I was getting clean tests in July, so I
> >would have to think that it was a change in CygWin as opposed to Perl.
> >I'll try and go back to 5.6.0 and build for debugging and see if it
> >happens there too.  That would pretty much rule out Perl itself as
> >the source.
> 
> Well, maybe there were two little changes, one here, one there, one
> opens a door, another one stumbles in...

OK, 5.6.0 builds and tests ok (5 failures but these are not suprising)
under CygWin 1.3.2.  Nothing like 5.6.1/bleadperl with lots of test 
successes but coredumps on completion.

So it appears that both bleadperl and 5.6.1 have something that is not
clean under CygWin.  The next task is to build cygwin1.dll with the 
additional memory debugging and try and track it down.

John

-- 
John Peacock
Director of Information Research and Technology
Rowman & Littlefield Publishing Group
4720 Boston Way
Lanham, MD 20706
301-459-3366 x.5010
fax 301-429-5747

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on cygwin - continued

[Christopher Faylor]

On Wed, Oct 17, 2001 at 12:22:40PM -0400, John Peacock wrote:
>So it appears that both bleadperl and 5.6.1 have something that is not
>clean under CygWin.  The next task is to build cygwin1.dll with the 
>additional memory debugging and try and track it down.

IIRC, this was the "next step" that was suggested the last time this
issue was raised.

We seem to have inexplicably skipped back to the "I dunno I think it's
1.3.3" stage of the discussion, for some reason.  This could easily be
the case but the fact that perl is dying in a call to free() does not
pinpoint the issue enough to draw that conclusion.

cgf

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on Cygwin - continued

[Jarkko Hietaniemi]

On Wed, Oct 17, 2001 at 05:37:43PM +0200, Gerrit P. Haase wrote:
> On Wed, 17 Oct 2001 11:10:21 -0400, you wrote:
> 
> >"Gerrit P. Haase" wrote:
> >> On Tue, 16 Oct 2001 11:36:33 -0400, you wrote:
> >> 
> >> >I have not had time to investigate; I'm still trying to work on the
> >> >CygWin core's with PERL_DESTRUCT_LEVEL=2.
> >> 
> >> I'm getting the same problems here with perl-5.6.1 if built with
> >> debugging, so it is probably a change that was made to both, bleadperl
> >> and maintperl or a change in cygwin.
> >
> >I am positive (no really) that I was getting clean tests in July, so I 
> >would have to think that it was a change in CygWin as opposed to Perl.  
> >I'll try and go back to 5.6.0 and build for debugging and see if it 
> >happens there too.  That would pretty much rule out Perl itself as 
> >the source.
> 
> Well, maybe there were two little changes, one here, one there, one
> opens a door, another one stumbles in...

Maybe.  But also I got 100% clean tests in July with the then Cygwin
and the then bleadperl, so John need not suspect his memory.

> Gerrit
> 
> -- 
> =^..^=
> bIg BrOtHeR iS wAtChInG YoU!

-- 
$jhi++; # http://www.iki.fi/jhi/
        # There is this special biologist word we use for 'stable'.
        # It is 'dead'. -- Jack Cohen

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on Cygwin - continued

[Gerrit P. Haase]

Jarkko Hietaniemi schrieb am 2001-10-18, 4:53:

>On Wed, Oct 17, 2001 at 05:37:43PM +0200, Gerrit P. Haase wrote:
>> On Wed, 17 Oct 2001 11:10:21 -0400, you wrote:
>> 
>> >"Gerrit P. Haase" wrote:
>> >> On Tue, 16 Oct 2001 11:36:33 -0400, you wrote:
>> >> 
>> >> >I have not had time to investigate; I'm still trying to work on the
>> >> >CygWin core's with PERL_DESTRUCT_LEVEL=2.
>> >> 
>> >> I'm getting the same problems here with perl-5.6.1 if built with
>> >> debugging, so it is probably a change that was made to both, bleadperl
>> >> and maintperl or a change in cygwin.
>> >
>> >I am positive (no really) that I was getting clean tests in July, so I 
>> >would have to think that it was a change in CygWin as opposed to Perl.  
>> >I'll try and go back to 5.6.0 and build for debugging and see if it 
>> >happens there too.  That would pretty much rule out Perl itself as 
>> >the source.
>> 
>> Well, maybe there were two little changes, one here, one there, one
>> opens a door, another one stumbles in...
>
>Maybe.  But also I got 100% clean tests in July with the then Cygwin
>and the then bleadperl, so John need not suspect his memory.

I get also the cores. Also with perl-5.6.1-12350, I will start a build 
now with the 5.6.1 release version.

Gerrit


-- 
=^..^=

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on cygwin - continued

[John Peacock]

Christopher Faylor wrote:
> 
> On Wed, Oct 17, 2001 at 12:22:40PM -0400, John Peacock wrote:
> >So it appears that both bleadperl and 5.6.1 have something that is not
> >clean under CygWin.  The next task is to build cygwin1.dll with the
> >additional memory debugging and try and track it down.
> 
> IIRC, this was the "next step" that was suggested the last time this
> issue was raised.

I have taken this step with 1.3.3, but I am still hampered by the now
identified gdb bug, which is what is really holding me back now.  Now 
that I know that 1.3.3 is not going to play nice with gdb under a Win2K 
workstation in an NT domain (is that qualified enough for you  ;~), I 
need to go back and get a copy of the last 1.3.2 source and build the 
augmented dll there.

If you can get the gdb bug fixed, even as a slipstream patch that I 
will have to apply myself, I can work vs 1.3.3.  Otherwise, the only 
thing I can do go back to a known configuration and try and work 
forward in time.

> 
> We seem to have inexplicably skipped back to the "I dunno I think it's
> 1.3.3" stage of the discussion, for some reason.  This could easily be
> the case but the fact that perl is dying in a call to free() does not
> pinpoint the issue enough to draw that conclusion.
> 

I'm not running 1.3.3, I am running 1.3.2.  I do not lightly accuse
CygWin of being at fault.  But I have tried Perl distributions known to
have tested clean before, and they now core during the cleanup phase.  
So if the application behavior has changed, but the application has 
not, that is very strong evidence that the underlying code (being 
CygWin) has to be the primary suspect.  It might not be cygwin1.dll 
itself; it may be gcc that is messing up an optimization, or it may be 
some other library that was updated in the last 3 months.

I may have to go back in time and install only those modules of CygWin
available in early July and then, assuming I can get clean builds, 
slowly work forward until everything goes south.

John

-- 
John Peacock
Director of Information Research and Technology
Rowman & Littlefield Publishing Group
4720 Boston Way
Lanham, MD 20706
301-459-3366 x.5010
fax 301-429-5747

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on cygwin - continued

[Christopher Faylor]

On Thu, Oct 18, 2001 at 09:17:27AM -0400, John Peacock wrote:
>Christopher Faylor wrote:
>> 
>> On Wed, Oct 17, 2001 at 12:22:40PM -0400, John Peacock wrote:
>> >So it appears that both bleadperl and 5.6.1 have something that is not
>> >clean under CygWin.  The next task is to build cygwin1.dll with the
>> >additional memory debugging and try and track it down.
>> 
>> IIRC, this was the "next step" that was suggested the last time this
>> issue was raised.
>
>I have taken this step with 1.3.3, but I am still hampered by the now
>identified gdb bug, which is what is really holding me back now.  Now 
>that I know that 1.3.3 is not going to play nice with gdb under a Win2K 
>workstation in an NT domain (is that qualified enough for you  ;~), I 
>need to go back and get a copy of the last 1.3.2 source and build the 
>augmented dll there.

What does building cygwin with malloc debugging have to do with gdb?
You don't need gdb to use this.

cgf

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on cygwin - continued

[John Peacock]

Christopher Faylor wrote:
> 
> On Thu, Oct 18, 2001 at 09:17:27AM -0400, John Peacock wrote:
> >Christopher Faylor wrote:
> >>
> >> On Wed, Oct 17, 2001 at 12:22:40PM -0400, John Peacock wrote:
> >> >So it appears that both bleadperl and 5.6.1 have something that is not
> >> >clean under CygWin.  The next task is to build cygwin1.dll with the
> >> >additional memory debugging and try and track it down.
> >>
> >> IIRC, this was the "next step" that was suggested the last time this
> >> issue was raised.
> >
> >I have taken this step with 1.3.3, but I am still hampered by the now
> >identified gdb bug, which is what is really holding me back now.  Now
> >that I know that 1.3.3 is not going to play nice with gdb under a Win2K
> >workstation in an NT domain (is that qualified enough for you  ;~), I
> >need to go back and get a copy of the last 1.3.2 source and build the
> >augmented dll there.
> 
> What does building cygwin with malloc debugging have to do with gdb?
> You don't need gdb to use this.

No, but if I want to step through the code to see exactly where the
malloc goes awry, I need gdb, don't I.  I'll take a look at this next
week (I am leaving for a long weekend).  Point me at any other docs you
might think useful (like do I need to install a new cygwin root or can
I just copy over the cygwin1.dll?).

Thanks

John

-- 
John Peacock
Director of Information Research and Technology
Rowman & Littlefield Publishing Group
4720 Boston Way
Lanham, MD 20706
301-459-3366 x.5010
fax 301-429-5747

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on cygwin - continued

[Gerrit P. Haase]

Christopher Faylor schrieb am 2001-10-18, 14:25:
> On Thu, Oct 18, 2001 at 09:17:27AM -0400, John Peacock wrote:
>> Christopher Faylor wrote:
>>> On Wed, Oct 17, 2001 at 12:22:40PM -0400, John Peacock wrote:
>>>>So it appears that both bleadperl and 5.6.1 have something that is not
>>>>clean under CygWin.  The next task is to build cygwin1.dll with the
>>>>additional memory debugging and try and track it down.
>>>
>>>IIRC, this was the "next step" that was suggested the last time this
>>>issue was raised.
>>
>>I have taken this step with 1.3.3, but I am still hampered by the now
>>identified gdb bug, which is what is really holding me back now.  Now 
>>that I know that 1.3.3 is not going to play nice with gdb under a Win2K 
>>workstation in an NT domain (is that qualified enough for you  ;~), I 
>>need to go back and get a copy of the last 1.3.2 source and build the 
>>augmented dll there.
>
>What does building cygwin with malloc debugging have to do with gdb?
>You don't need gdb to use this.

How do I do that?  What is the difference to a 'normal' debug version?

Thanks,

Gerrit

-- 
=^..^=

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on cygwin - continued

[Gerrit P. Haase]

John Peacock schrieb am 2001-10-18, 9:17:

>Christopher Faylor wrote:
[...heavy cut...]
>> We seem to have inexplicably skipped back to the "I dunno I think it's
>> 1.3.3" stage of the discussion, for some reason.  This could easily be
>> the case but the fact that perl is dying in a call to free() does not
>> pinpoint the issue enough to draw that conclusion.

What about this change?
'w32api' current dates from 2001-09-14 and previous is from 2001-05-20.

CVSROOT:	/cvs/src
Module name:	src
Changes by:	earnie@sources.redhat.com	2001-06-11 13:06:55

Modified files:
	winsup/w32api  : ChangeLog 
	winsup/w32api/include: objidl.h 

Log message:
	* objidl.h (IMalloc::ReAlloc()): Correct declaration.
	(IMalloc::Free()): Ditto.

Patches:
http://sources.redhat.com/cgi-bin/cvsweb.cgi/src/winsup/w32api/ChangeLog.diff?cvsroot=src&r1=1.96&r2=1.97
http://sources.redhat.com/cgi-bin/cvsweb.cgi/src/winsup/w32api/include/objidl.h.diff?cvsroot=src&r1=1.4&r2=1.5

looks like this:

===================================================================
RCS file: /cvs/src/src/winsup/w32api/include/objidl.h,v
retrieving revision 1.4
retrieving revision 1.5
diff -u -r1.4 -r1.5
--- src/winsup/w32api/include/objidl.h	2001/04/17 12:58:06	1.4
+++ src/winsup/w32api/include/objidl.h	2001/06/11 20:06:54	1.5
@@ -482,8 +482,8 @@
 	STDMETHOD_(ULONG,AddRef)(THIS) PURE;
 	STDMETHOD_(ULONG,Release)(THIS) PURE;
 	STDMETHOD_(void*,Alloc)(THIS_ ULONG) PURE;
-	STDMETHOD_(void,ReAlloc)(THIS_ void*) PURE;
-	STDMETHOD_(void*,Free)(THIS_ void*) PURE;
+	STDMETHOD_(void*,ReAlloc)(THIS_ void*,ULONG) PURE;
+	STDMETHOD_(void,Free)(THIS_ void*) PURE;
 	STDMETHOD_(ULONG,GetSize)(THIS_ void*) PURE;
 	STDMETHOD_(int,DidAlloc)(THIS_ void*) PURE;
 	STDMETHOD_(void,HeapMinimize)(THIS) PURE;


[...more cutted...]

Comments?

Gerrit


-- 
=^..^=

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on cygwin - continued

[Christopher Faylor]

On Thu, Oct 18, 2001 at 05:00:21PM -0400, John Peacock wrote:
>No, but if I want to step through the code to see exactly where the
>malloc goes awry, I need gdb, don't I.

Early in this discussion, Egor Duda suggested that you take a look at
a document that he wrote called "how-to-debug-cygwin.txt".  This tells
you how to configure cygwin with malloc debugging.  This causes cygwin
to go through very extensive tests on each malloc/free and report on
block overruns or double frees with information about where the block
was first allocated.

This is what I mean by malloc debugging.

cgf

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on cygwin - continued

[Gerrit P. Haase]

Gerrit P. Haase schrieb am 2001-10-19, 0:38:

>John Peacock schrieb am 2001-10-18, 9:17:
>
>>Christopher Faylor wrote:
>[...heavy cut...]
>>> We seem to have inexplicably skipped back to the "I dunno I think it's
>>> 1.3.3" stage of the discussion, for some reason.  This could easily be
>>> the case but the fact that perl is dying in a call to free() does not
>>> pinpoint the issue enough to draw that conclusion.
>
>What about this change?
>'w32api' current dates from 2001-09-14 and previous is from 2001-05-20.

No, with cygwin-1.3.2 it is the same problem.

Gerrit



-- 
=^..^=

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/

Re: test failures building perl on cygwin - continued

[Gerrit P. Haase]

Gerrit P. Haase schrieb am 2001-10-19, 8:55:

>>What about this change?
>>'w32api' current dates from 2001-09-14 and previous is from 2001-05-20.
>
>No, with cygwin-1.3.2 it is the same problem.

And it is also the same problem with cygwin-1.3.1

Gerrit

-- 
=^..^=

--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Bug reporting:         http://cygwin.com/bugs.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/