Re: heads up: wine 20020605 broken on FreeBSD, workaround included

[Gerald Pfeifer]

[ I'm now also Cc:ing this to the Wine developers, thus the full-quote.
  In fact, all non-GLIBC platforms seem to be affected by this bug! ]

Based on your pointer, I believe I found the problem. In aclocal.m4 we
have:

AC_DEFUN([WINE_CHECK_ERRNO],
  [AC_CACHE_CHECK([for reentrant libc: $1],[wine_cv_libc_r_$1],
    [AC_TRY_RUN([int myerrno = 0;
  char buf[256];
  int *$1(){return &myerrno;}
  main(){connect(0,buf,255); exit(!myerrno);}],
    wine_cv_libc_r_$1=yes, wine_cv_libc_r_$1=no,
    wine_cv_libc_r_$1=yes)])
  AS_IF([test "$wine_cv_libc_r_$1" = "yes"],[$2],[$3])])

which, in configure, becomes:

int myerrno = 0;
  char buf256;
  int *__error(){return &myerrno;}
  main(){connect(0,buf,255); exit(!myerrno);}

That is, instead of an array of 256 characters, we have one character,
and the invocation of connect() is definitely incorrect.

Could someone more familiar with autoconf please have a look at this?

Gerald

On Mon, 1 Jul 2002, Scott Bolte wrote:
> Over the weekend I tried to get wine 20020605 to run on a
>    new installation of FreeBSD 4.6. It took a while, I haven't
>    tried wine since 1997 after all, but I determined that my
>    problems were due to configure's reentrant libc test failing
>    when building the port.
>
> The wine_cv_libc_reentrant test in configure.ac first checks
>    for Linux (__errno_location) and then tests for FreeBSD
>    (__error).  I don't have more time to figure out why, but
>    nesting the FreeBSD test that way results in a false
>    negative. The configuration will continue, as will the
>    build, but the resulting non-reentrant code goes into an
>    endless loop when trying to spawn threads.
>
> A crude workaround, for FreeBSD users at least, is to
>    replace the Linux test with the FreeBSD one. This needs to
>    be done before configure is run. I did it by hand, but I
>    expect the following code would do the trick.
>
> % cd /usr/ports/emulators/wine
>   % make patch
>   % perl -i -pe 's/__errno_location/__error/' work/wine-20020605/configure.ac
>   % make
>   % make install
>
> Below is the autoconf test script that demonstrates the
>    problem. It needs to be run in
>    /usr/ports/emulators/wine/work/wine-20020605/configure.ac
> Scott
>
> P.S. I'm not on the freebsd-emulators mailing list, but I
>         see from freebsd-questions that at least one other
>         person got wine to compile on 4.6. I can't explain
>         why it worked for him and not for me. However, I
>         installed a brand new copy of 4.6 from the standard
>         ISO images just to run wine.
>
> ---- testconf.ac
> dnl Process this file with autoconf to produce a configure script.
> dnl Original author: Michael Patra
> dnl See ChangeLog file for detailed change history.
>
> m4_define(WINE_VERSION,regexp(m4_include(VERSION),[version ([-.0-9A-Za-z]+)],[\1]))
>
> AC_PREREQ(2.53)
> AC_INIT([Wine],WINE_VERSION)
> AC_CONFIG_SRCDIR(server/atom.c)
> AC_CONFIG_AUX_DIR(tools)
>
> dnl **** Check for some programs ****
>
> AC_PROG_CC
>
> dnl **** Check for reentrant libc ****
>
> wine_cv_libc_reentrant=no
>
> dnl	#####
> dnl	##### There is a problem if the FreeBSD test is nested
> dnl	##### after the Linux test. In that instance, it fails.
> dnl	##### If, on the other hand, it is tested first, it works.
> dnl	#####
> dnl	##### Swap the order of two following test case to see
> dnl	##### how they differ. (Ignore the second case since it
> dnl	##### will use cached data.)
> dnl	#####
>
> dnl FreeBSD style errno location
> WINE_CHECK_ERRNO([__error], [wine_cv_libc_reentrant=__error])
>
> dnl Linux style errno location
> WINE_CHECK_ERRNO([__errno_location], [wine_cv_libc_reentrant=__errno_location],
>   dnl FreeBSD style errno location
>   WINE_CHECK_ERRNO([__error], [wine_cv_libc_reentrant=__error]
> ))

heads up: wine 20020605 broken on FreeBSD, workaround included

[Marcus Meissner]

On Mon, Jul 01, 2002 at 04:18:47PM +0200, Gerald Pfeifer wrote:
> [ I'm now also Cc:ing this to the Wine developers, thus the full-quote.
>   In fact, all non-GLIBC platforms seem to be affected by this bug! ]
>
> Based on your pointer, I believe I found the problem. In aclocal.m4 we
> have:
>
> AC_DEFUN([WINE_CHECK_ERRNO],
>   [AC_CACHE_CHECK([for reentrant libc: $1],[wine_cv_libc_r_$1],
>     [AC_TRY_RUN([int myerrno = 0;
>   char buf[256];
>   int *$1(){return &myerrno;}
>   main(){connect(0,buf,255); exit(!myerrno);}],
>     wine_cv_libc_r_$1=yes, wine_cv_libc_r_$1=no,
>     wine_cv_libc_r_$1=yes)])
>   AS_IF([test "$wine_cv_libc_r_$1" = "yes"],[$2],[$3])])
>
> which, in configure, becomes:
>
> int myerrno = 0;
>   char buf256;
>   int *__error(){return &myerrno;}
>   main(){connect(0,buf,255); exit(!myerrno);}
>
> That is, instead of an array of 256 characters, we have one character,
> and the invocation of connect() is definitely incorrect.
>
> Could someone more familiar with autoconf please have a look at this?

The [256] is evaluated and replaced by 256 by m4. So this is not a glibc,
but an autoconf problem.

What is the autoconf version used there?

Ciao, Marcus

heads up: wine 20020605 broken on FreeBSD, workaround included

[Francois Gouget]

On Mon, 1 Jul 2002, Marcus Meissner wrote:
[...]
> The [256] is evaluated and replaced by 256 by m4. So this is not a glibc,
> but an autoconf problem.
>
> What is the autoconf version used there?

The problem is with the contents of the 'configure' script in Wine CVS:

line 10545 is ok:

char buf[256];

but the following occurences are wrong:

line 10588, 10631, 10674, 10717:

char buf256;

So the problem would be either with the autoconf that Alexandre is
using, or with the configure.ac file.

--
Francois Gouget         [e-mail]        http://fgouget.free.fr/
                      Computers are like airconditioners
                They stop working properly if you open WINDOWS

heads up: wine 20020605 broken on FreeBSD, workaround included

[Francois Gouget]

The bug was in the configure.ac file. Both Marcus and me sent patches
(different approaches ;-) to wine-patches. One of them has to be right
so this should be fixed pretty soon.

--
Francois Gouget         [e-mail]        http://fgouget.free.fr/
The nice thing about meditation is that it makes doing nothing quite respectable
                                  -- Paul Dean

heads up: wine 20020605 broken on FreeBSD, workaround included

[Gerald Pfeifer]

On Mon, 1 Jul 2002, Francois Gouget wrote:
> The bug was in the configure.ac file. Both Marcus and me sent patches
> (different approaches ;-) to wine-patches. One of them has to be right
> so this should be fixed pretty soon.

Yes, it works fine.  Thanks to all that helped in fixing this!

Gerald