[ 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
ScottP.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] ))