Re: jar (fastjar) 0.92-gcc corrupts files

[Randall R Schulz <rrschulz at cris dot com>]

Mike,

First of all, a very simple test does not confirm the symptom you're 
seeing. If you'd like to send me (off-list) a BZip2-compressed TAR 
archive containing test files and the series of fastjar invocations 
that produce the problem on your system, I'd be willing to try to 
confirm it that way.


Secondly, it's an odd problem, consider the following:

The obvious hypothesis is that this is a text / binary problem. 
However, a quick look at the fastjar code shows that O_BINARY is used 
uniformly in jartool.c (but not in jargrep.c, which is probably a problem):

% egrep 'open\>[    ]*\(' *.[ch]
jargrep.c:      if((fd = open(jarfile, O_RDONLY)) == -1) {
jartool.c:      jarfd = open(jarfile, O_CREAT | O_BINARY | O_WRONLY | O_TRUNC,
jartool.c:      jarfd = open(jarfile, O_RDONLY | O_BINARY);
jartool.c:      if((jarfd = open(jarfile, O_RDWR | O_BINARY)) < 0) {
jartool.c:    mfd = open(mf_name, O_RDONLY | O_BINARY);
jartool.c:    add_fd = open(file, O_RDONLY | O_BINARY);


Surely the O_BINARY option to the open(2) call overrides a text mount 
or the CYGWIN=nobinmode option, so I don't see how this could be the 
problem. I can find no mention of automode in the local configuration / 
make files (those in the fastjar directory itself).

So I guess it must be something else.

One thing I'm noticing is that among the include files in /usr/include, 
there are disparate definitions of O_BINARY. Presumably this is not an 
error, but rather a difference in the implementation or operation of 
Windows' / MinGW's and Cygwin's open options:

% egrep -R O_BINARY /usr/include
/usr/include/mingw/fcntl.h:#define      _O_BINARY       0x8000 /* Input 
and output is not translated. */
/usr/include/mingw/fcntl.h:#define      _O_RAW          _O_BINARY
/usr/include/mingw/fcntl.h:#define      O_BINARY        _O_BINARY
/usr/include/sys/fcntl.h:#define O_BINARY       _FBINARY
/usr/include/sys/fcntl.h:#define _O_BINARY      O_BINARY
/usr/include/sys/fcntl.h:#define _O_RAW         O_BINARY


% egrep -R _FBINARY /usr/include
/usr/include/sys/fcntl.h:#define _FBINARY        0x10000
/usr/include/sys/fcntl.h:#define O_BINARY       _FBINARY


There were O_BINARY matches from /usr/include/apache, 
/usr/include/kpathsea and /usr/include/postresql but I'll assume those 
are irrelevant. That leaves the mingw/ and sys/ directories. Since 
jartool is a Cygwin program, it should be using sys/fcntl.

% egrep '^[         ]*#[    ]*include\>' jartool.{h,c}
jartool.h:#include "config.h"
jartool.c:#include "config.h"
jartool.c:#include <zlib.h>
jartool.c:#include <stdlib.h>
jartool.c:#include <unistd.h>
jartool.c:#include <stdio.h>
jartool.c:#include <sys/stat.h>
jartool.c:#include <sys/types.h>
jartool.c:#include <sys/param.h>
jartool.c:#include <dirent.h>
jartool.c:#include <fcntl.h>
jartool.c:#include <string.h>
jartool.c:#include <errno.h>
jartool.c:#include <sys/time.h>
jartool.c:#include <time.h>
jartool.c:#include <getopt.h>
jartool.c:#include "jartool.h"
jartool.c:#include "zipfile.h"
jartool.c:#include "dostime.h"
jartool.c:#include "pushback.h"
jartool.c:#include "compress.h"

It looks OK, but without configuring the whole shebang to find out 
exactly what -I (cap I, as in include) or -l (small l, as in library), 
binmode.o / automode.o, -D, etc. options are in effect it's impossible 
to tell for sure.

I'm afraid I'm not that ambitious.


Randall Schulz

P.S. Mike, you need to get your shift key fixed. It appears to be 
malfunctioning.


At 10:14 2003-03-14, Mike Andrews wrote:
> hi there, i've noticed that this cygwin-distributed application 
> ("/bin/jar.exe") often corrupts files it archives, especially 
> binaries, the evidence mainly being that md5sums change after the 
> archiving/dearchiving operation. text files also sometimes get 
> corrupted, but so far i've only seen ignorable whitespace corruption. 
> thus, i've switched to sun microsystem's j2sdk-1.4 jar.exe program 
> until this issue gets resolved. thanks so much -- mike andrews
>
> mike andrews


--
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/