Date: Sun, 26 Dec 1999 09:34:36 +0200 (IST) From: Eli Zaretskii X-Sender: eliz AT is To: =?iso-8859-2?Q?Wojciech_Ga=B3=B1zka?= cc: "'djgpp-workers AT delorie DOT com'" Subject: Re: Fwd new release of LFN for Windows NT 4.0 - attached file In-Reply-To: <13288F4408ADD11186FF0060B06A431303341245@MSGWAW1> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Reply-To: djgpp-workers AT delorie DOT com X-Mailing-List: djgpp-workers AT delorie DOT com X-Unsubscribes-To: listserv AT delorie DOT com Precedence: bulk On Thu, 23 Dec 1999, =?iso-8859-2?Q?Wojciech_Ga=B3=B1zka?= wrote: > binutils > not been able to run even configure.bat > test abandoned I had no special problems building Binutils 2.9.1 on plain DOS. If you want, I can send the patches against stock Binutils 2.9.1 that let me build the distribution. > flex > misc/msdod/configur.bat vomplains: > touch: config_status: Input or output error (EIO) > make bigcheck: cannot move 'scan.tmp' to 'scan.c': Invalid argument > (EINVAL) I suspect that these happen because some LFN function doesn't work correctly. `touch' calls function 5705h of Interrupt 21h to set the last-access time. Does the LFN driver support this function? As for `mv', try to reproduce the problem by hand and then post here how to reproduce it (I presume that `mv' generally works, so this case is somehow special). > gawk > make djgpp completed > make test only pid and a few others failed but completes > successfully The Gawk test suite works for me with no problems except with the part where it tests the results of calling math functions with invalid arguments (that part also fails on every platfrom but SunOS). Please look at the failed tests closer and tell which ones failed. > grep > need to comment out lines marked ">>"at the end of each Makefile > #define VERSION "$VERSION" > >>grep _POSIX_VERSION /usr/include/sys/unistd.h >/dev/null > 2>&1 > #define VERSION "$VERSION" > >>s%@VERSION@%$VERSION%g Any idea why did you need to comment these lines? What version of Grep did you try to build? > make > configure passed > make prints 'extended error 183 (file already exists)' but goes on I would suggest to post the details. Where do these problems happen during the build process? > several times prints 'no DOS memory' > make completed > make check not available, 'no make-test-*suite' Try with the latest version 3.78.1: it includes the test suite, and the tests work with the DJGPP tools (you will need Perl). > sed > djgpp\config completed > make completed > make check completed, after first issuing > 'utod testsuite/*.good', 'utod testsuite/*.inp' It is much easier to say "make check CMP='diff -c'", like the file djgpp/README suggests... > tar > djgpp\config completed > make completed > make check completed failed tests are extrac01.sh > extrac02.sh, extrac03.sh, incremen.sh THIS IS REAL BAD > (this is probably when GetShortPathName fails if file not in current > dir) The Tar test suite worked flawlessly for me, so I'd suggest to take a closer look at these failures. Please post the failure descriptions here, if you cannot figure them out. > texinfo > djgpp\config completed > make completed > make check not available Try Texinfo 4.0, it includes a test suite. > gdb > ` with the current port of GDB to DOS it is impossible to use LFN > under NT > for example calls are made to 'libiberty.sed' while 'libibert.sed' > only > exists Are you talking about GDB 4.18? If not, please try the latest version. I don't see libibert.sed or libiberty.sed anywhere in the source distribution of GDB 4.18, but it *is* LFN-clean, so you should have no problems. > gcc > not even though of running build.djg/configure.bat This is GCC 2.7.2.1, I presume. Please try the latest version. > djgpp/src/ > Errors during make > Load error: can't switch mode > Load error: no DOS memory > Load error: No DPMI selectors This is due to Windows NT leaking selectors and DOS memory. Simply restart the build process (by typing "make"), and it will pick up where the previous run left off. > cannot create dependencies for stubXXXX.S in libc/stubs. Why? Can you describe the details of this problem? Also, I sugegst to try building the latest alpha of djlsr203. > creating an foo.tar.gz archive > djtar cvf autoconf.tar autocnf1.22 completes without problems > gzip.exe autoconf.tar creates compressed archive but prints a > error messsage. 'Input or output error EIO' > gzip.exe -d autoconf.tar (in another dir) completes but print > that infamous EIO error > djtar xvf autoconf.tar completes but print 'Could not change access > anf > modification times: Input or output error EIO' I'm guessing that these are all consequences of the same problems that makes `touch' fail: something is wrong in setting access time inside the function `utime'. I suggest to step with a debugger into `utime' and see what exactly fails there. > process_get_Short_name should return short name equivalence even if > file not in current directory Sorry, I don't understand this. Could you please elaborate? > in case a long name is not found should a file with a name > consisting > of first 8 be searched for? This is not how LFN-aware DOS functions work on Windows 9X, so I don't think the NT driver should behave differently. Perhaps you could explain why do you think this would be a good idea. What problems are you trying to solve by this?