dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Jason Tishler]

Michael,

On Thu, Dec 06, 2001 at 10:15:42AM -0500, Michael Hudson wrote:
> know anything about this?
> 
> "no" is a perfectly acceptable answer :)

Unfortunately, yes.  See below...

> building under Cygwin ends with the following error:
> 
> building 'gdbm' extension
> gcc -g -O3 -Wall -Wstrict-prototypes -DUSE_DL_IMPORT -I. -I/cygdrive/e/Python-2.2b2/./Include -I/usr/local/include -IInclude/ -c /cygdrive/e/Python-2.2b2/Modules/gdbmmodule.c -o build/temp.cygwin-1.3.5-i686-2.2/gdbmmodule.o
> e:\buildpy\python.exe: *** unable to remap 
> C:\cygnus\bin\cygssl.dll to same address as parent -- 0x1A7C0000
>       0 [main] python 1292 sync_with_child: child 1108 (0xF0) died before initialization with status code 0x1
>   25901 [main] python 1292 sync_with_child: *** child state child loading dlls
> error: Resource temporarily unavailable
> make: *** [sharedmods] Error 1

The above occurs during Cygwin's fork() when the Cygwin DLL cannot
load a DLL to the same address in the child that it had in the parent.
I have seen this during Python 2.1.1 regression tests with threads
enabled.

My WAG is that during a regression test with threads enabled more
DLLs (i.e., extension modules) have been loaded (i.e., imported) into
the parent's address space.  Hence, the probability of a DLL address
clash is higher than without threads enabled.  Otherwise, just to be
clear, I feel that this problem has nothing to do with threads.

I have just reproduced this build problem with Python 2.2b2 too.  However,
I don't believe that it is related to the particular Python version one
attempts to build.  The reason that we are observing this problem now is
that very recently some packages (e.g., OpenSSL) were just updated with
their static libraries replaced by their shared (i.e., DLL) counterparts.
Hence, the probability of an address clash just increased some more.

Previously, I tried suggestion (a) from:

    http://sources.redhat.com/ml/cygwin/2001-09/msg01165.html

Unfortunately, even the non-threads Python 2.1.1 failed its regression
test with this error.

Also, I searched the MSDN and found the rebase utility:

    http://msdn.microsoft.com/library/default.asp?url=/library/en-us/tools/perfutil_2z39.asp
    
which could possibly solve this problem.  I haven't tried it yet, but
even if it does solve the problem, then people without the MS SDK will
not be able to build Cygwin Python. :,(

Solving this problem has been on my list, but I thought that I could
procrastinate until after the threads issue was resolved.  Apparently,
I was wrong.  And as typical with my luck, major Cygwin Python problems
are synchronized with Python releases.  Sigh...

A few months ago, I did some investigation on this problem.  What I
learned from adding some printfs to dll_list::load_after_fork() was
that the DLL in question was *already* remapped from the its preferred
image-base in the parent while this address was still available in
the child.  This empirical evidence seems to contradict the following
source code comment:

    /* It loaded in the wrong place.  Dunno why this happens but it
       always seems to happen when there are multiple DLLs attempting
       to load into the same address space.  In the "forked" process,
       the second DLL always loads into a different location. */

In this case, it seems like the DLL was loaded in a "different" location
in the parent not the child.

Since this code *and* my understanding of the issues seemed tenuous at
best, I placed this one very gingerly back onto my "round tuit" list.

Any help with this problem would be greatly appreciated.

> Comment By: Michael Hudson (mwh)
> Date: 2001-12-06 05:53
> 
> PS: can we add jason as a developer?  then we can assign
> bugs to him <wink>.

Hmm... :,)

Thanks,
Jason

--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Charles Wilson]

Jason Tishler wrote:


>>
>>building 'gdbm' extension
>>gcc -g -O3 -Wall -Wstrict-prototypes -DUSE_DL_IMPORT -I. -I/cygdrive/e/Python-2.2b2/./Include -I/usr/local/include -IInclude/ -c /cygdrive/e/Python-2.2b2/Modules/gdbmmodule.c -o build/temp.cygwin-1.3.5-i686-2.2/gdbmmodule.o
>>e:\buildpy\python.exe: *** unable to remap 
>>C:\cygnus\bin\cygssl.dll to same address as parent -- 0x1A7C0000
>>      0 [main] python 1292 sync_with_child: child 1108 (0xF0) died before initialization with status code 0x1
>>  25901 [main] python 1292 sync_with_child: *** child state child loading dlls
>>error: Resource temporarily unavailable
>>make: *** [sharedmods] Error 1
>>
> 
> The above occurs during Cygwin's fork() when the Cygwin DLL cannot
> load a DLL to the same address in the child that it had in the parent.
> I have seen this during Python 2.1.1 regression tests with threads
> enabled.


Part of the problem may be that cyggdbm.dll was built with 
--auto-image-base.  It was later demonstrated that this can cause 
problems with fork; you're better off just letting ld assign the default 
dllbase, which means that EVERY process will remap the dll at runtime. 
Thus, no hardcoded conflicts.  Downside: *very* slightly delay in 
loading DLLs -- probably unnoticeable.

(Did I get that right, robert?)

Anyway, I plan to redo cyggdbm "eventually" without the 
--auto-image-base.  Doing so *may* fix this problem, but I'm not sure...

--Chuck






--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Michael Hudson]

On Thu, 6 Dec 2001, Charles Wilson wrote:

> Jason Tishler wrote:
> 
> 
> >>
> >>building 'gdbm' extension
> >>gcc -g -O3 -Wall -Wstrict-prototypes -DUSE_DL_IMPORT -I. -I/cygdrive/e/Python-2.2b2/./Include -I/usr/local/include -IInclude/ -c /cygdrive/e/Python-2.2b2/Modules/gdbmmodule.c -o build/temp.cygwin-1.3.5-i686-2.2/gdbmmodule.o
> >>e:\buildpy\python.exe: *** unable to remap 
> >>C:\cygnus\bin\cygssl.dll to same address as parent -- 0x1A7C0000
> >>      0 [main] python 1292 sync_with_child: child 1108 (0xF0) died before initialization with status code 0x1
> >>  25901 [main] python 1292 sync_with_child: *** child state child loading dlls
> >>error: Resource temporarily unavailable
> >>make: *** [sharedmods] Error 1
> >>
> > 
> > The above occurs during Cygwin's fork() when the Cygwin DLL cannot
> > load a DLL to the same address in the child that it had in the parent.
> > I have seen this during Python 2.1.1 regression tests with threads
> > enabled.
> 
> 
> Part of the problem may be that cyggdbm.dll was built with 
> --auto-image-base.  It was later demonstrated that this can cause 
> problems with fork; you're better off just letting ld assign the default 
> dllbase, which means that EVERY process will remap the dll at runtime. 
> Thus, no hardcoded conflicts.  Downside: *very* slightly delay in 
> loading DLLs -- probably unnoticeable.
> 
> (Did I get that right, robert?)
> 
> Anyway, I plan to redo cyggdbm "eventually" without the 
> --auto-image-base.  Doing so *may* fix this problem, but I'm not sure...

I didn't get Jason's email.  Odd that.

Anyway, it's more than just gdbm; if I stop that building, then termios 
fails in a similar manner; disable that and it's resource.  And so on.

This is really ******* annoying (tm); I'm trying to fix generic Python 
bugs and have to scp my changes onto another machine to test them...

Cheers,
M.


--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[David Abrahams]

FWIW, Jason's patch worked for me, except for curses of course which seems
to be broken in cygwin.

-Dave

----- Original Message -----
From: "Michael Hudson" <mwh@python.net>
To: "Charles Wilson" <cwilson@ece.gatech.edu>
Cc: "Jason Tishler" <jason@tishler.net>;
<david_abrahams@users.sourceforge.net>; "Cygwin" <cygwin@sources.redhat.com>
Sent: Friday, December 07, 2001 9:50 AM
Subject: Re: dll_list::load_after_fork() blues (was Re:
[ python-Bugs-489709 ] Building Fails ...)


> On Thu, 6 Dec 2001, Charles Wilson wrote:
>
> > Jason Tishler wrote:
> >
> >
> > >>
> > >>building 'gdbm' extension
> >
>>gcc -g -O3 -Wall -Wstrict-prototypes -DUSE_DL_IMPORT -I. -I/cygdrive/e/Pyt
hon-2.2b2/./Include -I/usr/local/include -IInclude/ -c
/cygdrive/e/Python-2.2b2/Modules/gdbmmodule.c -o
build/temp.cygwin-1.3.5-i686-2.2/gdbmmodule.o
> > >>e:\buildpy\python.exe: *** unable to remap
> > >>C:\cygnus\bin\cygssl.dll to same address as parent -- 0x1A7C0000
> > >>      0 [main] python 1292 sync_with_child: child 1108 (0xF0) died
before initialization with status code 0x1
> > >>  25901 [main] python 1292 sync_with_child: *** child state child
loading dlls
> > >>error: Resource temporarily unavailable
> > >>make: *** [sharedmods] Error 1
> > >>
> > >
> > > The above occurs during Cygwin's fork() when the Cygwin DLL cannot
> > > load a DLL to the same address in the child that it had in the parent.
> > > I have seen this during Python 2.1.1 regression tests with threads
> > > enabled.
> >
> >
> > Part of the problem may be that cyggdbm.dll was built with
> > --auto-image-base.  It was later demonstrated that this can cause
> > problems with fork; you're better off just letting ld assign the default
> > dllbase, which means that EVERY process will remap the dll at runtime.
> > Thus, no hardcoded conflicts.  Downside: *very* slightly delay in
> > loading DLLs -- probably unnoticeable.
> >
> > (Did I get that right, robert?)
> >
> > Anyway, I plan to redo cyggdbm "eventually" without the
> > --auto-image-base.  Doing so *may* fix this problem, but I'm not sure...
>
> I didn't get Jason's email.  Odd that.
>
> Anyway, it's more than just gdbm; if I stop that building, then termios
> fails in a similar manner; disable that and it's resource.  And so on.
>
> This is really ******* annoying (tm); I'm trying to fix generic Python
> bugs and have to scp my changes onto another machine to test them...
>
> Cheers,
> M.
>


--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Jason Tishler]

Dave,

On Fri, Dec 07, 2001 at 10:03:24AM -0500, David Abrahams wrote:
> FWIW, Jason's patch worked for me,

Thanks for the feedback.

> except for curses of course which seems to be broken in cygwin.

The above is a known problem, see the following for details:

    http://sources.redhat.com/ml/cygwin-apps/2001-11/msg00260.html

Jason

--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Robert Collins]

----- Original Message -----
From: "Charles Wilson" <cwilson@ece.gatech.edu>
> > The above occurs during Cygwin's fork() when the Cygwin DLL cannot
> > load a DLL to the same address in the child that it had in the
parent.
> > I have seen this during Python 2.1.1 regression tests with threads
> > enabled.
>
>
> Part of the problem may be that cyggdbm.dll was built with
> --auto-image-base.  It was later demonstrated that this can cause
> problems with fork; you're better off just letting ld assign the
default
> dllbase, which means that EVERY process will remap the dll at runtime.
> Thus, no hardcoded conflicts.  Downside: *very* slightly delay in
> loading DLLs -- probably unnoticeable.
>
> (Did I get that right, robert?)

Yes. There is actually a longer term solution... which is to 'rebase'
every cygwin linked .dll on a particular system to not conflict with
each other - which has to be done by setup.exe.

Rob

> Anyway, I plan to redo cyggdbm "eventually" without the
> --auto-image-base.  Doing so *may* fix this problem, but I'm not
sure...
>
> --Chuck
>
>
>
>
>
>
> --
> 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/
>
>


--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Charles Wilson]

Robert Collins wrote:

> > Part of the problem may be that cyggdbm.dll was built with
> > --auto-image-base.  It was later demonstrated that this can cause
> > problems with fork; you're better off just letting ld assign the
> default
> > dllbase, which means that EVERY process will remap the dll at runtime.
> > Thus, no hardcoded conflicts.  Downside: *very* slightly delay in
> > loading DLLs -- probably unnoticeable.
> >
> > (Did I get that right, robert?)
> 
> Yes. There is actually a longer term solution... which is to 'rebase'
> every cygwin linked .dll on a particular system to not conflict with
> each other - which has to be done by setup.exe.

Yes, but only with the apps and dll's that setup knows about.  Recall
the discussion on this list some months ago concerning sybase DLL's (I
think).   Somebody had a custom cygwin app that linked to
vendor-supplied database DLLs as well as cygwin stuff (which is fine as
long as the resulting app is not distributed...GPL conflicts
notwithstanding).

Anyway, they had a problem after upgrading to a new cygwinish dll
(cygncurses?? I think) w.r.t. load-on-fork.  There's no way setup/rebase
can be used to avoid that problem a_priori...is there?  (As I recall,
the person did a 'hand rebase' on his own system to work around
it...aware that he would have to repeat the process every time he
updated that problem DLL from the cygwin dist)

--Chuck

--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Christopher Faylor]

On Fri, Dec 07, 2001 at 07:54:32PM -0500, Charles Wilson wrote:
>Anyway, they had a problem after upgrading to a new cygwinish dll
>(cygncurses?? I think) w.r.t. load-on-fork.  There's no way setup/rebase
>can be used to avoid that problem a_priori...is there?  (As I recall,
>the person did a 'hand rebase' on his own system to work around
>it...aware that he would have to repeat the process every time he
>updated that problem DLL from the cygwin dist)

A rebase utility for cygwin would be pretty nice, I think.

We could even include a .bat file which could iterate over all of the
dlls in /bin and make sure there were no conflicts.

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/

rebase utility (was Re: dll_list::load_after_fork() blues ...)

[Jason Tishler]

[-- Attachment #1: Type: text/plain, Size: 578 bytes --]

Chris,

On Fri, Dec 07, 2001 at 08:02:22PM -0500, Christopher Faylor wrote:
> A rebase utility for cygwin would be pretty nice, I think.

Attached is the beginning of a rebase utility.  It is modeled after the
MS one:

    http://msdn.microsoft.com/library/default.asp?url=/library/en-us/tools/perfutil_2z39.asp

So far I have only implemented the -b and -d options.

If a stand-alone utility is still deemed useful (as opposed to integrating
the functionality directly into setup.exe), then I will "GNU-ize", clean
up, and submit it to cygwin-patches for consideration.

Jason

[-- Attachment #2: rebase.cc --]
[-- Type: text/plain, Size: 3236 bytes --]

/*
 * Copyright (c) 2001 Jason Tishler
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; either version 2
 * of the License, or (at your option) any later version.
 * 
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
 *
 * $Id: rebase.cc,v 1.4 2001/12/12 12:36:57 jtishler Exp $
 */

#include <iostream>
#include <sstream>
#include <string>
#include <time.h>
#include <stdlib.h>
#include <getopt.h>
#include <sys/cygwin.h>
#include <windows.h>
#include <imagehlp.h>

using namespace std;

string PosixToWin32(const string& aPosixPath);
void ParseArgs(int argc, char* argv[]);
unsigned long StringToUlong(const string& aString);
void Usage();

ULONG theImageBase = 0;
BOOL theDownFlag = FALSE;
int theArgsIndex = 0;

void
main(int argc, char* argv[])
{
	ParseArgs(argc, argv);
	ULONG aNewImageBase = theImageBase;

	for (int i = theArgsIndex; i < argc; i++)
	{
		string aFile = PosixToWin32(argv[i]);
		ULONG anOldImageSize, anOldImageBase, aNewImageSize;
		ULONG aPrevNewImageBase = aNewImageBase;
		BOOL aStatus = ReBaseImage(
			const_cast<char*>(aFile.c_str()), // CurrentImageName
			0, // SymbolPath
			TRUE, // fReBase
			FALSE, // fRebaseSysfileOk
			theDownFlag, // fGoingDown
			0, // CheckImageSize
			&anOldImageSize, // OldImageSize
			&anOldImageBase, // OldImageBase
			&aNewImageSize, // NewImageSize
			&aNewImageBase, // NewImageBase
			time(0)); // TimeStamp

		// ReBaseImage seems to never returns false!
		DWORD aStatus2 = GetLastError();
		if (aStatus2 != 0)
		{
			cerr << "ReBaseImage() failed with last error = " <<
				GetLastError() << endl;
			exit(2);
		}
		cout << aFile << hex << ": new base = " <<
			((theDownFlag) ? aNewImageBase : aPrevNewImageBase) <<
			", new size = " << aNewImageSize << endl;
	}

	exit(0);
}

string
PosixToWin32(const string& aPosixPath)
{
	char aWin32Path[MAX_PATH];
	cygwin_conv_to_win32_path(aPosixPath.c_str(), aWin32Path);
	return aWin32Path;
}

void
ParseArgs(int argc, char* argv[])
{
	const char* anOptions = "b:d";
	for (int anOption; (anOption = getopt(argc, argv, anOptions)) != -1;)
	{
		switch (anOption)
		{
			case 'b':
				theImageBase = StringToUlong(optarg);
				break;
			case 'd':
				theDownFlag = TRUE;
				break;
			default:
				Usage();
				exit(1);
				break;
		}
	}

	if (theImageBase == 0)
	{
		Usage();
		exit(1);
	}

	theArgsIndex = optind;
}

unsigned long
StringToUlong(const string& aString)
{
	stringstream ss;
	unsigned long aUlong;
	string::size_type anIndex = aString.find("0x");
	if (anIndex == 0)
		ss << hex << string(aString, 2, aString.size() - 2);
	else
		ss << aString;
	ss >> aUlong;
	return aUlong;
}

void
Usage()
{
	cerr << "usage: rebase -b BaseAddress [-d] ImageFileName ..." << endl;
}


[-- Attachment #3: Type: text/plain, Size: 214 bytes --]

--
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: rebase utility (was Re: dll_list::load_after_fork() blues ...)

[Robert Collins]

----- Original Message -----
From: "Jason Tishler" <jason@tishler.net>
To: <cygwin@cygwin.com>
Sent: Thursday, December 13, 2001 12:07 AM
Subject: rebase utility (was Re: dll_list::load_after_fork() blues ...)


> Chris,
>
> On Fri, Dec 07, 2001 at 08:02:22PM -0500, Christopher Faylor wrote:
> > A rebase utility for cygwin would be pretty nice, I think.
>
> Attached is the beginning of a rebase utility.  It is modeled after
the
> MS one:
>
>
http://msdn.microsoft.com/library/default.asp?url=/library/en-us/tools/p
erfutil_2z39.asp
>
> So far I have only implemented the -b and -d options.
>
> If a stand-alone utility is still deemed useful (as opposed to
integrating
> the functionality directly into setup.exe), then I will "GNU-ize",
clean
> up, and submit it to cygwin-patches for consideration.

If I may add two cents here.

Binutils could use this.
Setup could use this.
A standalone tool may be useful.
libtool could use a binutils or standalone util.
install.sh could "

So I'd suggest you make your utilitie a verrry thin wrapper around a
library, and put everything functional in the library.

Rob


--
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: rebase utility (was Re: dll_list::load_after_fork() blues ...)

[Jason Tishler]

On Wed, Dec 12, 2001 at 08:07:59AM -0500, Jason Tishler wrote:
> On Fri, Dec 07, 2001 at 08:02:22PM -0500, Christopher Faylor wrote:
> > A rebase utility for cygwin would be pretty nice, I think.
> 
> Attached is the beginning of a rebase utility.  It is modeled after the
> MS one:
> 
> [snip]

Oops!  I forgot to mention how to build rebase:

    $ g++ -O2 -o rebase rebase.cc -limagehlp

Jason

--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Robert Collins]

----- Original Message -----
From: "Charles Wilson" <cwilson@ece.gatech.edu>
> > Yes. There is actually a longer term solution... which is to
'rebase'
> > every cygwin linked .dll on a particular system to not conflict with
> > each other - which has to be done by setup.exe.
>
> Yes, but only with the apps and dll's that setup knows about.  Recall
> the discussion on this list some months ago concerning sybase DLL's (I
> think).   Somebody had a custom cygwin app that linked to
> vendor-supplied database DLLs as well as cygwin stuff (which is fine
as
> long as the resulting app is not distributed...GPL conflicts
> notwithstanding).

Actually what I have in mind is
* changes to install
(pseudo code):
if (installing a .dll or .exe)
  rebase and store info in /etc/setup
* changes to setup
if (installing a .dll or .exe)
  rebase and store info in /etc/setup

rebase:
find object size - sz
  lookup a gap of sz in the address table in /etc/setup
find object dependencies
  foreach
    if (a cygwin .dll)
      rebase this
    if (a non-cygwin .dll)
      store the base and size info in /etc/setup


> Anyway, they had a problem after upgrading to a new cygwinish dll
> (cygncurses?? I think) w.r.t. load-on-fork.  There's no way
setup/rebase
> can be used to avoid that problem a_priori...is there?  (As I recall,

Yes - following the pseudo code above should be ok (because its install
system dependent, not build system dependent).

MS actually have a tool for developers to do this with - as part of
their programs setup.exe. We may even be able to use that tool - which
would use the MS local machine database, not one of our own.

Rob


--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Norman Vine]

Robert Collins writes:
>
>----- Original Message -----
>From: "Charles Wilson" <cwilson@ece.gatech.edu>
>> > The above occurs during Cygwin's fork() when the Cygwin DLL cannot
>> > load a DLL to the same address in the child that it had in the parent.
>> > I have seen this during Python 2.1.1 regression tests with threads
>> > enabled.
>>
>>
>> Part of the problem may be that cyggdbm.dll was built with
>> --auto-image-base.  It was later demonstrated that this can cause
>> problems with fork; you're better off just letting ld assign the default
>> dllbase, which means that EVERY process will remap the dll at runtime.
>> Thus, no hardcoded conflicts.  Downside: *very* slightly delay in
>> loading DLLs -- probably unnoticeable.
>>
>> (Did I get that right, robert?)
>
>Yes. There is actually a longer term solution... which is to 'rebase'
>every cygwin linked .dll on a particular system to not conflict with
>each other - which has to be done by setup.exe.

FWIW
I just tried patching Cygwin's dlopen() function to always rebase dll's
on loading with a 'slightly hacked to force rebasing' version of
niftyload.cpp
< available from http://thecodeproject/dll/niftyload.asp >
and got the same errors as Jason reported in his original message.

  C:\home\jtishler\src\PythonCvs\nothreads\python.exe: *** unable to remap
C:\cygwin\bin\cygssl.dll to same address as parent -- 0x1A2F0000
          0 [main] python 852 sync_with_child: child 2772(0x158) died before
initialization with status code 0x1
        996 [main] python 852 sync_with_child: *** child state child loading
dlls
    Traceback (most recent call last):
      File "test2.py", line 4, in ?
        pid = os.fork()
    OSError: [Errno 11] Resource temporarily unavailable

The author of the code states that the base address is chosen the wrong way.
"Rebased dll's should start from the top of address space and work down"

BTW
If it isn't obvious I know next to nothing about dll rebaseing issues

Cheers

Norman


--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Robert Collins]

----- Original Message -----
From: "Norman Vine" <nhv@cape.com>

> FWIW
> I just tried patching Cygwin's dlopen() function to always rebase
dll's
> on loading with a 'slightly hacked to force rebasing' version of

Interesting idea. Unfortunately it only covers the dlopened dll's - and
we need to ensure the imported .dll's relocate (or not) consistently.

It's the consistency thats the problem, not the actual relocation.

Rob


--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Jason Tishler]

[-- Attachment #1: Type: text/plain, Size: 2264 bytes --]

On Sat, Dec 08, 2001 at 10:08:14AM +1100, Robert Collins wrote:
> Yes. There is actually a longer term solution... which is to 'rebase'
> every cygwin linked .dll on a particular system to not conflict with
> each other - which has to be done by setup.exe.

I just tried a hand rebase of my system using the MS supplied rebase
tool to see if this will fix the problem at least for the Python case.
Specifically, I rebased the following DLLs:

    o Python DLL (e.g., libpython2.2.dll)
    o all Python standard shared extension modules (e.g., _socket.dll)
    o all Cygwin DLLs in /usr/bin that match cyg*.dll except for the
      following:

          - cygwin1.dll: since I believe that it relies on being based
            at 0x61000000
          - cygcurl-2.dll: because it gets "whacked" by rebase and
            AFAICT is not used by Python anyway
          - cygtclpip80.dll: because it appears not to be relocatable

Additionally, following the suggestions from the MSDN, I rebased from
0x68000000 down.  So, all of the above DLLs were rebased into the range
of 0x672e0000 - 0x68000000.

After rebasing, the minimal test case that previously exhibited the
problem:

    http://cygwin.com/ml/cygwin/2001-12/msg00419.html

now works fine.

Unfortunately, when I run the complete Python regression test, I still
get the same three test failures as reported by Michael without rebasing:

    test_popen2
    test_pty
    test_socket

When I run these tests individually (i.e., not part of the complete test
suite), then they pass.  Hence, the rebasing appears not to completely
solve this problem.

See the attached snippet of output from a regression test run (and
search for 0x1A).  It shows that although there should not be DLL base
address conflicts, some DLLs are being rebased in the parent anyway.
A few examples are:

    _socket.dll: rebased to 0x67f50000 loaded at 0x1A260000
    cygz.dll:    rebased to 0x678b0000 loaded at 0x1A310000

Would other Python users (with access to MS's rebase tool) be willing
to try to reproduce my findings to eliminate the possibility of cockpit
error on my part?

Does anyone understand why the DLLs are being rebased even though there
theoretically is no chance of a base address conflict anymore?

Thanks,
Jason

[-- Attachment #2: rebase.txt --]
[-- Type: text/plain, Size: 7627 bytes --]

test_popen2
      0 [main] python 648 dll_list::load_after_fork: remap failed for C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\strop.dll with old handle = 0x1A230000, new handle = 0x420000
  97411 [main] python 648 dll_list::load_after_fork: remap failed for C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\readline.dll with old handle = 0x1A340000, new handle = 0x1A790000
 115881 [main] python 648 dll_list::load_after_fork: remap failed twice for C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\readline.dll with old handle = 0x1A340000, new handle = 0x1A790000
 123685 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\libpython2.2.dll handle = 0x675E0000
 123997 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\math.dll handle = 0x675C0000
 124122 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\time.dll handle = 0x67370000
 124742 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\errno.dll handle = 0x67890000
 124875 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\struct.dll handle = 0x673D0000
 125032 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\_codecs.dll handle = 0x67FE0000
 125155 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\strop.dll handle = 0x1A230000
 125473 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\_testcapi.dll handle = 0x67F30000
 125600 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\cStringIO.dll handle = 0x67E70000
 126484 [main] python 648 dll_list::load_after_fork: name = C:\cygwin\bin\cygcrypto.dll handle = 0x67D40000
 126616 [main] python 648 dll_list::load_after_fork: name = C:\cygwin\bin\cygssl.dll handle = 0x67A60000
 126924 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\_socket.dll handle = 0x1A260000
 127044 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\cPickle.dll handle = 0x1A2A0000
 127650 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\binascii.dll handle = 0x67EC0000
 127777 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\termios.dll handle = 0x1A2E0000
 127924 [main] python 648 dll_list::load_after_fork: name = C:\cygwin\bin\cygz.dll handle = 0x1A310000
 128040 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\zlib.dll handle = 0x672E0000
 128337 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\_locale.dll handle = 0x67F80000
 128459 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\pcre.dll handle = 0x67500000
 129262 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\select.dll handle = 0x67430000
 129388 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\regex.dll handle = 0x674B0000
 129694 [main] python 648 dll_list::load_after_fork: name = C:\cygwin\bin\cygncurses5.dll handle = 0x67BB0000
 129814 [main] python 648 dll_list::load_after_fork: name = C:\cygwin\bin\cygreadline5.dll handle = 0x67AB0000
 133807 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\readline.dll handle = 0x1A340000
 133958 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\fcntl.dll handle = 0x67870000
 143197 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\_weakref.dll handle = 0x67F20000
 143353 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\array.dll handle = 0x67F00000
 143780 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\audioop.dll handle = 0x410000
 143907 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\cmath.dll handle = 0x67E50000
 144735 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\crypt.dll handle = 0x67E30000
 144864 [main] python 648 dll_list::load_after_fork: name = C:\cygwin\bin\cyggdbm.dll handle = 0x67D10000
 145173 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\gdbm.dll handle = 0x67840000
 145295 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\grp.dll handle = 0x67830000
 145425 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\md5.dll handle = 0x675A0000
 145541 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\sha.dll handle = 0x67410000
 146303 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\_hotshot.dll handle = 0x640000
 146453 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\parser.dll handle = 0x670000
 146761 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\imageop.dll handle = 0x6B0000
 146880 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\rgbimg.dll handle = 0x67470000
 147536 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\xreadlines.dll handle = 0x6E0000
 147664 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\operator.dll handle = 0x67560000
 147969 [main] python 648 dll_list::load_after_fork: name = C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\mmap.dll handle = 0x700000
C:\home\jtishler\src\PythonCvs\nothreads\python.exe: *** unable to remap C:\home\jtishler\src\PythonCvs\nothreads\build\lib.cygwin-1.3.5-i686-2.2\mmap.dll to same address as parent -- 0x1A790000
      0 [main] python 1628 sync_with_child: child 648(0x1D8) died before initialization with status code 0x1
    225 [main] python 1628 sync_with_child: *** child state child loading dlls
test test_popen2 crashed -- exceptions.OSError: [Errno 11] Resource temporarily unavailable


[-- Attachment #3: Type: text/plain, Size: 214 bytes --]

--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Robert Collins]

Possibly because the cygwin heap is getting allocated across where those
.dll's would go.

Rob
===
----- Original Message -----
From: "Jason Tishler" <jason@tishler.net>
To: "Cygwin" <cygwin@sources.redhat.com>
Cc: "Michael Hudson" <mwh@python.net>;
<david_abrahams@users.sourceforge.net>
Sent: Monday, December 10, 2001 11:46 PM
Subject: Re: dll_list::load_after_fork() blues (was Re: [
python-Bugs-489709 ] Building Fails ...)


> On Sat, Dec 08, 2001 at 10:08:14AM +1100, Robert Collins wrote:
> > Yes. There is actually a longer term solution... which is to
'rebase'
> > every cygwin linked .dll on a particular system to not conflict with
> > each other - which has to be done by setup.exe.
>
> I just tried a hand rebase of my system using the MS supplied rebase
> tool to see if this will fix the problem at least for the Python case.
> Specifically, I rebased the following DLLs:
>
>     o Python DLL (e.g., libpython2.2.dll)
>     o all Python standard shared extension modules (e.g., _socket.dll)
>     o all Cygwin DLLs in /usr/bin that match cyg*.dll except for the
>       following:
>
>           - cygwin1.dll: since I believe that it relies on being based
>             at 0x61000000
>           - cygcurl-2.dll: because it gets "whacked" by rebase and
>             AFAICT is not used by Python anyway
>           - cygtclpip80.dll: because it appears not to be relocatable
>
> Additionally, following the suggestions from the MSDN, I rebased from
> 0x68000000 down.  So, all of the above DLLs were rebased into the
range
> of 0x672e0000 - 0x68000000.
>
> After rebasing, the minimal test case that previously exhibited the
> problem:
>
>     http://cygwin.com/ml/cygwin/2001-12/msg00419.html
>
> now works fine.
>
> Unfortunately, when I run the complete Python regression test, I still
> get the same three test failures as reported by Michael without
rebasing:
>
>     test_popen2
>     test_pty
>     test_socket
>
> When I run these tests individually (i.e., not part of the complete
test
> suite), then they pass.  Hence, the rebasing appears not to completely
> solve this problem.
>
> See the attached snippet of output from a regression test run (and
> search for 0x1A).  It shows that although there should not be DLL base
> address conflicts, some DLLs are being rebased in the parent anyway.
> A few examples are:
>
>     _socket.dll: rebased to 0x67f50000 loaded at 0x1A260000
>     cygz.dll:    rebased to 0x678b0000 loaded at 0x1A310000
>
> Would other Python users (with access to MS's rebase tool) be willing
> to try to reproduce my findings to eliminate the possibility of
cockpit
> error on my part?
>
> Does anyone understand why the DLLs are being rebased even though
there
> theoretically is no chance of a base address conflict anymore?
>
> Thanks,
> Jason
>


------------------------------------------------------------------------
--------


> --
> 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/


--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Jason Tishler]

Rob,

On Mon, Dec 10, 2001 at 11:42:01PM +1100, Robert Collins wrote:
> Possibly because the cygwin heap is getting allocated across where those
> .dll's would go.

That is what I figured.  Any suggestions on a good address to use during
the rebasing?

Thanks,
Jason

--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Robert Collins]

----- Original Message -----
From: "Jason Tishler" <jason@tishler.net>


> Rob,
>
> On Mon, Dec 10, 2001 at 11:42:01PM +1100, Robert Collins wrote:
> > Possibly because the cygwin heap is getting allocated across where
those
> > .dll's would go.
>
> That is what I figured.  Any suggestions on a good address to use
during
> the rebasing?

Leave ~ the size of the cygwin heap between each .dll needed to load
bash + (say) vim. Then use process explorer to see where the heap is
occuring, and rebase everything second time avoiding that base + the
size of the maximum heap.

Rob


--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Michael Hudson]

On Mon, 10 Dec 2001, Jason Tishler wrote:

> Unfortunately, when I run the complete Python regression test, I still
> get the same three test failures as reported by Michael without rebasing:
> 
>     test_popen2
>     test_pty
>     test_socket
> 
> When I run these tests individually (i.e., not part of the complete test
> suite), then they pass.  Hence, the rebasing appears not to completely
> solve this problem.

FWIW, and I don't know how much that is, all tests pass if I link _socket 
statically.  Oh, and this is building without threads, it seems.  I'll do 
a new build with threads and see if anything changes, but I doubt it.

So what is it that dynamically loading _socket does that screws fork()?

Cheers,
M.


--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Norman Vine]

Michael Hudson writes:
>
>On Mon, 10 Dec 2001, Jason Tishler wrote:
>
>> Unfortunately, when I run the complete Python regression test, I still
>> get the same three test failures as reported by Michael without rebasing:
>>
>>     test_popen2
>>     test_pty
>>     test_socket
>>
>> When I run these tests individually (i.e., not part of the complete test
>> suite), then they pass.  Hence, the rebasing appears not to completely
>> solve this problem.
>
>FWIW, and I don't know how much that is, all tests pass if I link _socket
>statically.  Oh, and this is building without threads, it seems.  I'll do
>a new build with threads and see if anything changes, but I doubt it.

GREAT IDEA !

I just rebuilt python 2.1.1 with threads and linking _socket statically
and all seems to work :-)

=== / usr / src / python-2.1.1 / Modules / Setup.local
# Edit this file for local setup changes
# socket line above, and possibly edit the SSL variable:
SSL=/usr
_socket socketmodule.c \
	-DUSE_SSL -I$(SSL)/include -I$(SSL)/include/openssl \
	-L$(SSL)/lib -lssl -lcrypto


Thanks Michael

Norman


--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Jason Tishler]

On Mon, Dec 10, 2001 at 11:52:08AM -0500, Norman Vine wrote:
> Michael Hudson writes:
> >FWIW, and I don't know how much that is, all tests pass if I link _socket
> >statically.  Oh, and this is building without threads, it seems.  I'll do
> >a new build with threads and see if anything changes, but I doubt it.
> 
> GREAT IDEA !
> 
> I just rebuilt python 2.1.1 with threads and linking _socket statically
> and all seems to work :-)
> 
> === / usr / src / python-2.1.1 / Modules / Setup.local
> # Edit this file for local setup changes
> # socket line above, and possibly edit the SSL variable:
> SSL=/usr
> _socket socketmodule.c \
> 	-DUSE_SSL -I$(SSL)/include -I$(SSL)/include/openssl \
> 	-L$(SSL)/lib -lssl -lcrypto

I just tried the above and it works.  However, it only works if one has
not fiddled around with rebasing their DLLs.

Although this is a good short-term workaround (and the one that I will
probably use when I release Python 2.2), I think that we should focus
our efforts on trying to solve this problem at its root cause -- Cygwin
fork() and DLL base address conflicts.  Otherwise, when something else
changes we will be back in the same situation.

I encourage the interested parties to try the rebase utility that I just
posted:

    http://cygwin.com/ml/cygwin/2001-12/msg00635.html

I would be interested if others can reproduce my findings and/or determine
how to rebase to fix this Python problem without having to build _socket
statically.

Thanks,
Jason

--
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/

Cygwin fork() rebase solution (was Re: dll_list::load_after_fork() blues ...)

[Jason Tishler]

[-- Attachment #1: Type: text/plain, Size: 3496 bytes --]

On Wed, Dec 12, 2001 at 08:57:44AM -0500, Jason Tishler wrote:
> On Mon, Dec 10, 2001 at 11:52:08AM -0500, Norman Vine wrote:
> > Michael Hudson writes:
> > >FWIW, and I don't know how much that is, all tests pass if I link _socket
> > >statically.  Oh, and this is building without threads, it seems.  I'll do
> > >a new build with threads and see if anything changes, but I doubt it.
> > 
> > GREAT IDEA !
> > 
> > I just rebuilt python 2.1.1 with threads and linking _socket statically
> > and all seems to work :-)
> > 
> > [snip]
> 
> I just tried the above and it works.  However, it only works if one has
> not fiddled around with rebasing their DLLs.
> 
> Although this is a good short-term workaround (and the one that I will
> probably use when I release Python 2.2), I think that we should focus
> our efforts on trying to solve this problem at its root cause -- Cygwin
> fork() and DLL base address conflicts.  Otherwise, when something else
> changes we will be back in the same situation.

I believe that I have found a rebase solution to the Cygwin fork()
problem that has been causing Cygwin Python some grief lately.  I added
an offset option to the attached rebase tool.  If I spread the DLLs out
by an extra 0x10000, then the fork() problem seems to be mitigate.

The following is the command line necessary to build rebase:

    $ g++ -O2 -o rebase rebase.cc -limagehlp

After rebasing the necessary Cygwin DLLs:

    $ cd /usr/bin
    $ rebase -d -b 0x68000000 -o 0x10000 cygXpm-X4.dll cygXpm-noX4.dll \
    cygbz21.0.dll cygcrypto.dll cygform5.dll cygform6.dll \
    cyggdbm.dll cyghistory4.dll cyghistory5.dll cygintl.dll \
    cygitcl30.dll cygitk30.dll cygjbig1.dll cygjpeg6b.dll cygmenu5.dll \
    cygmenu6.dll cygncurses++5.dll cygncurses++6.dll cygncurses5.dll \
    cygncurses6.dll cygpanel5.dll cygpanel6.dll cygpcre.dll \
    cygpcreposix.dll cygpng2.dll cygreadline4.dll cygreadline5.dll \
    cygregex.dll cygssl.dll cygtcl80.dll cygtclreg80.dll cygtiff3.dll \
    cygtk80.dll cygz.dll

I am able to build Python *without* the following patch:

    http://sourceforge.net/tracker/index.php?func=detail&aid=491107&group_id=5470&atid=305470

After rebasing the necessary Cygwin and Python DLLs:

    $ cd /usr/bin
    $ rebase -d -b 0x68000000 -o 0x10000 cygXpm-X4.dll cygXpm-noX4.dll \
    cygbz21.0.dll cygcrypto.dll cygform5.dll cygform6.dll \
    cyggdbm.dll cyghistory4.dll cyghistory5.dll cygintl.dll \
    cygitcl30.dll cygitk30.dll cygjbig1.dll cygjpeg6b.dll cygmenu5.dll \
    cygmenu6.dll cygncurses++5.dll cygncurses++6.dll cygncurses5.dll \
    cygncurses6.dll cygpanel5.dll cygpanel6.dll cygpcre.dll \
    cygpcreposix.dll cygpng2.dll cygreadline4.dll cygreadline5.dll \
    cygregex.dll cygssl.dll cygtcl80.dll cygtclreg80.dll cygtiff3.dll \
    cygtk80.dll cygz.dll ~/src/PythonCvs/nothreads/libpython2.2.dll \
    ~/src/PythonCvs/nothreads/build/lib.cygwin-1.3.6-i686-2.2/*.dll

I can run the full regression test without any failures *even though*
_socket is a shared module.

I encourage those interested in Cygwin Python to determine whether or not
the above rebase solution works for them too.  I'm particular interested
in 9x/Me reports since I do not have access to those platforms.  Please
post your results to the Cygwin and Python mailing lists.

If my findings are corroborated, then I will work with the Cygwin team
to integrate rebase into setup.exe so that rebasing automatically occurs
every time that setup.exe is run.

Thanks,
Jason

[-- Attachment #2: rebase.cc --]
[-- Type: text/plain, Size: 3447 bytes --]

/*
 * Copyright (c) 2001 Jason Tishler
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; either version 2
 * of the License, or (at your option) any later version.
 * 
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
 *
 * $Id: rebase.cc,v 1.5 2001/12/18 20:21:19 jtishler Exp $
 */

#include <iostream>
#include <sstream>
#include <string>
#include <time.h>
#include <stdlib.h>
#include <getopt.h>
#include <sys/cygwin.h>
#include <windows.h>
#include <imagehlp.h>

using namespace std;

string PosixToWin32(const string& aPosixPath);
void ParseArgs(int argc, char* argv[]);
unsigned long StringToUlong(const string& aString);
void Usage();

ULONG theImageBase = 0;
BOOL theDownFlag = FALSE;
ULONG theOffset = 0;
int theArgsIndex = 0;

void
main(int argc, char* argv[])
{
	ParseArgs(argc, argv);
	ULONG aNewImageBase = theImageBase;

	for (int i = theArgsIndex; i < argc; i++)
	{
		if (theDownFlag)
			aNewImageBase -= theOffset;

		string aFile = PosixToWin32(argv[i]);
		ULONG anOldImageSize, anOldImageBase, aNewImageSize;
		ULONG aPrevNewImageBase = aNewImageBase;
		BOOL aStatus = ReBaseImage(
			const_cast<char*>(aFile.c_str()), // CurrentImageName
			0, // SymbolPath
			TRUE, // fReBase
			FALSE, // fRebaseSysfileOk
			theDownFlag, // fGoingDown
			0, // CheckImageSize
			&anOldImageSize, // OldImageSize
			&anOldImageBase, // OldImageBase
			&aNewImageSize, // NewImageSize
			&aNewImageBase, // NewImageBase
			time(0)); // TimeStamp

		// ReBaseImage seems to never returns false!
		DWORD aStatus2 = GetLastError();
		if (aStatus2 != 0)
		{
			cerr << "ReBaseImage() failed with last error = " <<
				GetLastError() << endl;
			exit(2);
		}
		cout << aFile << hex << ": new base = " <<
			((theDownFlag) ? aNewImageBase : aPrevNewImageBase) <<
			", new size = " << aNewImageSize + theOffset << endl;

		if (!theDownFlag)
			aNewImageBase += theOffset;
	}

	exit(0);
}

string
PosixToWin32(const string& aPosixPath)
{
	char aWin32Path[MAX_PATH];
	cygwin_conv_to_win32_path(aPosixPath.c_str(), aWin32Path);
	return aWin32Path;
}

void
ParseArgs(int argc, char* argv[])
{
	const char* anOptions = "b:do:";
	for (int anOption; (anOption = getopt(argc, argv, anOptions)) != -1;)
	{
		switch (anOption)
		{
			case 'b':
				theImageBase = StringToUlong(optarg);
				break;
			case 'd':
				theDownFlag = TRUE;
				break;
			case 'o':
				theOffset = StringToUlong(optarg);
				break;
			default:
				Usage();
				exit(1);
				break;
		}
	}

	if (theImageBase == 0)
	{
		Usage();
		exit(1);
	}

	theArgsIndex = optind;
}

unsigned long
StringToUlong(const string& aString)
{
	stringstream ss;
	unsigned long aUlong;
	string::size_type anIndex = aString.find("0x");
	if (anIndex == 0)
		ss << hex << string(aString, 2, aString.size() - 2);
	else
		ss << aString;
	ss >> aUlong;
	return aUlong;
}

void
Usage()
{
	cerr << "usage: rebase -b BaseAddress [-d] -o Offset ImageFileName ..." << endl;
}


[-- Attachment #3: Type: text/plain, Size: 214 bytes --]

--
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: Cygwin fork() rebase solution (was Re: dll_list::load_after_fork() blues ...)

[Robert Collins]

Your results are enough for me Jason - I'm happy to include this in
setup in some form.

Rob


--
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: cygwin fork() rebase solution (was Re: dll_list::load_after_fork() blues ...)

[Christopher Faylor]

On Wed, Dec 19, 2001 at 08:33:50AM +1100, Robert Collins wrote:
>Your results are enough for me Jason - I'm happy to include this in
>setup in some form.

Nice job, Jason.  Thanks for following this through.

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: cygwin fork() rebase solution (was Re: dll_list::load_after_fork() blues ...)

[Jason Tishler]

Chris,

On Tue, Dec 18, 2001 at 10:28:13PM -0500, Christopher Faylor wrote:
> On Wed, Dec 19, 2001 at 08:33:50AM +1100, Robert Collins wrote:
> >Your results are enough for me Jason - I'm happy to include this in
> >setup in some form.
> 
> Nice job, Jason.  Thanks for following this through.

You are very welcome.  I still have more following through to do:

1. formal contribution of the tool (most likely to binutils)
2. integration with setup.exe

Jason

--
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: cygwin fork() rebase solution (was Re: dll_list::load_after_fork() blues ...)

[Chris McDonough]

Hi Jason et. al,

FYI, Zope was broken under out-of-the-box Cygwin 1.3.6 because of the fork
issue.  The rebase solution allows Zope to work again.

Many thanks,

- C

----- Original Message -----
From: "Jason Tishler" <jason@tishler.net>
To: <cygwin@cygwin.com>
Sent: Friday, December 21, 2001 7:51 AM
Subject: Re: cygwin fork() rebase solution (was Re:
dll_list::load_after_fork() blues ...)


> Chris,
>
> On Tue, Dec 18, 2001 at 10:28:13PM -0500, Christopher Faylor wrote:
> > On Wed, Dec 19, 2001 at 08:33:50AM +1100, Robert Collins wrote:
> > >Your results are enough for me Jason - I'm happy to include this in
> > >setup in some form.
> >
> > Nice job, Jason.  Thanks for following this through.
>
> You are very welcome.  I still have more following through to do:
>
> 1. formal contribution of the tool (most likely to binutils)
> 2. integration with setup.exe
>
> Jason
>
> --
> 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/
>
>


--
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: cygwin fork() rebase solution (was Re: dll_list::load_after_fork() blues ...)

[Jason Tishler]

Chris,

On Sat, Dec 22, 2001 at 10:22:37AM -0500, Chris McDonough wrote:
> FYI, Zope was broken under out-of-the-box Cygwin 1.3.6 because of the fork
> issue.  The rebase solution allows Zope to work again.
> 
> Many thanks,

You are very welcome!  Thanks for taking the time to see if this solution
solved your problem too.

Jason

--
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: Cygwin fork() rebase solution (was Re: dll_list::load_after_fork() blues ...)

[Norman Vine]

Jason Tishler writes:
>
>I believe that I have found a rebase solution to the Cygwin fork()
>problem that has been causing Cygwin Python some grief lately.  I added
>an offset option to the attached rebase tool.  If I spread the DLLs out
>by an extra 0x10000, then the fork() problem seems to be mitigate.
>

>After rebasing the necessary Cygwin and Python DLLs:

>I can run the full regression test without any failures *even though*
>_socket is a shared module.

> If my findings are corroborated, then I will work with the Cygwin team
> to integrate rebase into setup.exe so that rebasing automatically occurs
> every time that setup.exe is run.

HI Jason

'Rebase'ing appears to work :-)

Following report on a Win98 box with a 'fresh' Cygwin install

Let me know if you need more testing done

Cheers

Norman

==========
bash-2.05a$ uname -a
CYGWIN_98-4.10 NHV 1.3.6(0.47/3/2) 2001-12-08 17:02 i686 unknown
==========
bash-2.05a$ ./rebase.sh
cygcrypto.dll: new base = 67f40000, new size = c0000
cygform5.dll: new base = 67f20000, new size = 20000
cygform6.dll: new base = 67f00000, new size = 20000
cyggdbm.dll: new base = 67ee0000, new size = 20000
cyghistory4.dll: new base = 67ec0000, new size = 20000
cyghistory5.dll: new base = 67ea0000, new size = 20000
cygmenu5.dll: new base = 67e80000, new size = 20000
cygmenu6.dll: new base = 67e60000, new size = 20000
cygncurses++5.dll: new base = 67e20000, new size = 40000
cygncurses++6.dll: new base = 67de0000, new size = 40000
cygncurses5.dll: new base = 67d90000, new size = 50000
cygncurses6.dll: new base = 67d40000, new size = 50000
cygpanel5.dll: new base = 67d20000, new size = 20000
cygpanel6.dll: new base = 67d00000, new size = 20000
cygreadline4.dll: new base = 67cd0000, new size = 30000
cygreadline5.dll: new base = 67c90000, new size = 40000
cygregex.dll: new base = 67c60000, new size = 30000
cygz.dll: new base = 67c30000, new size = 30000
cygssl.dll: new base = 67bf0000, new size = 40000
========
bash-2.05a$ python regrtest.py
Python 2.1.1 (#34, Dec 19 2001, 20:24:16)
[GCC 2.95.3-5 (cygwin special)] on cygwin

.......

15 tests skipped:
test_al test_bsddb test_cd test_cl test_dbm test_dl test_gl test_imgfile
test_largefile
test_linuxaudiodev test_locale test_nis test_socketserver test_sunaudiodev
test_sundry
=========
        Cygwin DLL version info:
        DLL version: 1.3.6
        DLL epoch: 19
        DLL bad signal mask: 19005
        DLL old termios: 5
        DLL malloc env: 28
        API major: 0
        API minor: 47
        Shared data: 3
        DLL identifier: cygwin1
        Mount registry: 2
        Cygnus registry name: Cygnus Solutions
        Cygwin registry name: Cygwin
        Program options name: Program Options
        Cygwin mount registry name: mounts v2
        Cygdrive flags: cygdrive flags
        Cygdrive prefix: cygdrive prefix
        Cygdrive default prefix:
        Build date: Sat Dec 8 17:02:30 EST 2001
        CVS tag: cygwin-1-3-6-6
        Shared id: cygwin1S3


--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Robert Collins]

----- Original Message -----
From: "Jason Tishler" <jason@tishler.net>
>           - cygwin1.dll: since I believe that it relies on being based
>             at 0x61000000

It does not rely on any base address per se. It *does* reply on the base
address being consistent between forked/spawned and exec'd parent and
children. That means that the base address must be one of the following:
a) Whatever is in the .dll, in all cygwin processs (for a given process
hierarchy). (ie it can be different for different invocations of
cygwin.bat).
b) Relocated consistently on the fly in all cygwin processs (for a given
process hierarchy). (ie it can be different for different invocations of
cygwin.bat).

>           - cygcurl-2.dll: because it gets "whacked" by rebase and
>             AFAICT is not used by Python anyway

Can you detail 'whack' a little more?

>           - cygtclpip80.dll: because it appears not to be relocatable

What error do you get?

> Unfortunately, when I run the complete Python regression test, I still
> get the same three test failures as reported by Michael without
rebasing:
>
>     test_popen2
>     test_pty
>     test_socket
>
> When I run these tests individually (i.e., not part of the complete
test
> suite), then they pass.  Hence, the rebasing appears not to completely
> solve this problem.

Can you get process explorer to show their address space - get them to
fail, and stop rather than terminating (usually there is a dialog you
could leave open or you can use the CYGWIN debug flags).

Rob


--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Jason Tishler]

Rob,

On Sun, Dec 16, 2001 at 07:26:37PM +1100, Robert Collins wrote:
> From: "Jason Tishler" <jason@tishler.net>
> >           - cygwin1.dll: since I believe that it relies on being based
> >             at 0x61000000
> 
> It does not rely on any base address per se. It *does* reply on the base
> address being consistent between forked/spawned and exec'd parent and
> children.
> [snip]

I knew that I was going to put my foot in my mouth with the above feeble
explanation.  Thanks for setting the record straight.

> >           - cygcurl-2.dll: because it gets "whacked" by rebase and
> >             AFAICT is not used by Python anyway
> 
> Can you detail 'whack' a little more?

    $ cp /usr/bin/cygcurl-2.dll .
    $ file cygcurl-2.dll
    cygcurl-2.dll: MS Windows PE Intel 80386 console DLL
    $ rebase -b 0x68000000 -d cygcurl-2.dll 
    cygcurl-2.dll: new base = 68000000, new size = 20000
    $ file cygcurl-2.dll 
    cygcurl-2.dll: MS-DOS executable (EXE), OS/2 or MS Windows
    $ objdump -p cygcurl-2.dll
    objdump: cygcurl-2.dll: File format not recognized

The above seems to indicated that cygcurl-2.dll did not "survive" a rebase
which is why I used the term "whacked."

> >           - cygtclpip80.dll: because it appears not to be relocatable
> 
> What error do you get?

None.  The rebase is just ineffective which should not be surprising
since:

    $ file cygtclpip80.dll 
    cygtclpip80.dll: MS Windows PE Intel 80386 ... not relocatable
                                                       ^^^^^^^^^^^
> > Hence, the rebasing appears not to completely solve this problem.
> 
> Can you get process explorer to show their address space - get them to
> fail, and stop rather than terminating (usually there is a dialog you
> could leave open or you can use the CYGWIN debug flags).

By "process explorer," do you mean Process Explorer (aka HandleEx) from
Sysinternals?  Or, Process Viewer that is part of Visual Studio?  Or,
something else?

Thanks,
Jason

--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Robert Collins]

> By "process explorer," do you mean Process Explorer (aka HandleEx)
from
> Sysinternals?

Yes.


--
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: dll_list::load_after_fork() blues (was Re: [ python-Bugs-489709 ] Building Fails ...)

[Jason Tishler]

[-- Attachment #1: Type: text/plain, Size: 917 bytes --]

The ASSume principle applies once again...

On Thu, Dec 06, 2001 at 12:44:26PM -0500, Jason Tishler wrote:
> I have just reproduced this build problem with Python 2.2b2 too.  However,
> I don't believe that it is related to the particular Python version one
> attempts to build.

Actually, setup.py has been enhanced to check the validity of each
module by importing it during the build problem.  Hence, the probability
of a DLL address clash during fork has increased significantly.  The
attached patch workarounds this problem.

I intend to submit this patch (redone against CVS) to the Python patch
collector ASAP so that Python 2.2 does not go out the door on 12/19 busted
for Cygwin.  If you have any objections or better solutions, please speak
up sooner rather than later.

Note that I still intend to fix the Cygwin Python fork problem, if
possible.  Hopefully, this workaround will be only temporary.

Jason

[-- Attachment #2: setup.py.diff --]
[-- Type: text/plain, Size: 601 bytes --]

--- setup.py.orig	Fri Dec  7 08:21:56 2001
+++ setup.py	Fri Dec  7 08:31:21 2001
@@ -159,6 +159,11 @@
         if 'Carbon' in ext.extra_link_args:
         	self.announce('WARNING: skipping import check for Carbon-based "%s"' % ext.name)
         	return
+        # Workaround for Cygwin: Cygwin currently has fork issues when many
+        # modules have been imported
+        if self.get_platform() == 'cygwin':
+        	self.announce('WARNING: skipping import check for Cygwin-based "%s"' % ext.name)
+        	return
         try:
             __import__(ext.name)
         except ImportError:


[-- Attachment #3: Type: text/plain, Size: 214 bytes --]

--
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/