reassign 644868 libgnustep-base1.22
retitle 644868 Programs abort on kfreebsd* when arguments exceed 512
affects 644868 gnustep-back-common
block 629477 with 644868
thanks
On Sun, Oct 09, 2011 at 11:02:25PM +0200, Axel Beckert wrote:
> Package: gnustep-back-common
> Version: 0.20.1-2
> Severity: serious
> gnustep-back-common reproducibly fails to configure after upgrade on
> one of my systems (kfreebsd-i386) as follows.
> Error: for some reason, argv not properly set up during GNUstep base
> initialization
> xargs: mknfonts: terminated by signal 6
This is a reincarnation of #593898, when the number of arguments exceeds
512. The attached test program demonstrates the problem:
$ make
$ i=0; d=0; while test $i -eq 0; do for d in `expr $((d + 1))`; do \
./obj/test `seq $d`; i=$?; done; done
Arguments given: 1
Arguments given: 2
...
Arguments given: 512
Error: for some reason, argv not properly set up during GNUstep base
initialization
The condition which leads to the error is
if (argv != 0 && argv[0] != 0)
which for some bizarre reason is false in this case. It's either a
kernel bug or limitation, I hope the GNU/kFreeBSD people have some clue?
There's nothing wrong in the gnustep-base code...
The only way I see as a workaround is to build gnustep-base with
--enable-fake-main, which is already the case on a number of *BSD
systems. However, this change is ABI-incompatible so all GNUstep
packages must be binNMUed _again_ on kfreebsd*. Is this acceptable for
the release team?
#import <Foundation/Foundation.h>
int
main (int argc, char **argv)
{
CREATE_AUTORELEASE_POOL (pool);
NSProcessInfo *procinfo = [NSProcessInfo processInfo];
NSArray *args = [procinfo arguments];
printf ("Arguments given: %d\n", [args count]);
RELEASE (pool);
exit (EXIT_SUCCESS);
}
ifndef GNUSTEP_MAKEFILES
GNUSTEP_MAKEFILES := $(shell gnustep-config --variable=GNUSTEP_MAKEFILES)
endif
include $(GNUSTEP_MAKEFILES)/common.make
TOOL_NAME = test
test_OBJC_FILES = test.m
include $(GNUSTEP_MAKEFILES)/tool.make
