Sorry that I don't get it but I can't compile the portable OpenSSH since 3.0 up to the SNAP of 27.03.2002 (about a year now, am really in trouble!). I always stumble over libutil not being able to reference atexit. You now offer the option: --disable-libutil but it does not help. At the end of the configure process the list shows up: Libraries: -lutil -lz -lnsl -lcrypto -lcrypt This may be the reason for that: checking for libutil.h... no checking for library containing login... -lutil I don't really get through that. This is ahead of my knowledge. But I am really wondering why it happens because on my system atexit is defined in either stdlib.h or stdio.h (this one, I think). I used atexit and it works fine for me. Because both headers are standards they must be included in your product at all, so why libutil can't find atexit? Is it a problem of autoconf?
This should be fixed in -current CVS, please test
This problem still appears with the newest cvs: openssh-SNAP-20020508.tar.gz I condifured the system using these options: ./configure --prefix=/usr --with-4in6 --with-md5-passwords --disable-libutil Though I chose to not use libutil, it appears in the list of dependencies (configure output): OpenSSH has been configured with the following options: User binaries: /usr/bin System binaries: /usr/sbin Configuration files: /usr/etc Askpass program: /usr/libexec/ssh-askpass Manual pages: /usr/man/catX PID file: /var/run sshd default user PATH: /usr/bin:/bin:/usr/sbin:/sbin Manpage format: cat PAM support: no KerberosIV support: no KerberosV support: no Smartcard support: no AFS support: no S/KEY support: no TCP Wrappers support: no MD5 password support: yes IP address in $DISPLAY hack: no Use IPv4 by default hack: no Translate v4 in v6 hack: yes BSD Auth support: no Random number source: OpenSSL internal ONLY Host: i686-pc-linux-gnu Compiler: gcc Compiler flags: -g -O2 -Wall -Wpointer-arith -Wno-uninitialized Preprocessor flags: Linker flags: Libraries: -lutil -lz -lnsl -lcrypto -lcrypt ere is the error message: /bin/ld -o ssh ssh.o sshconnect.o sshconnect1.o sshconnect2.o sshtty.o readconf.o clientloop.o -L. -Lopenbsd-compat/ -lssh -lopenbsd-compat -lutil -lz -lnsl -lcrypto -lcrypt /bin/ld: warning: cannot find entry symbol _start; defaulting to 0804b208 /usr/lib/libutil.so: undefined reference to `atexit' make: *** [ssh] Fehler 1 Please, I am without OpenSSH for over a year now! I have to keep an old system on a separate partition to log into my sourceforge account. I really need support. thanks
Why is libutil a problem?
Ahem, did you read the message body? Maybe the problem is not because of libutil but OpenSSH stops because of reference problems with libutil. If you know the reason and solution then please tell me. Your qeustion does not make sense to me, I don't understand position. Please explain what you want to say or know!
atexit no longer exists within the code base. How can this be tripping up compile? Can you provide us with the compile time error? Or more details as to why you are saying. "I always stumble over libutil not being able to reference atexit." When.. 20020406 [..] - (bal) We no longer use atexit()/xatexit()/on_exit() Maybe what Linux.. Because I have not seen any problems compiling under Redhat nor Mandrake.
Do you call this support! I already gave the information. Maybe you should learn to read a message to its bottom. To tell you how it looks like for the user of your product: =========================================================== I disabled the use of libutil (--disable-libutil) but got this configuration output: Libraries: -lutil -lz -lnsl -lcrypto -lcrypt So, at least your configuration option does not render very clearly what actually happens. A user is a bit confused by that. Then I get this compilation error: /bin/ld -o ssh ssh.o sshconnect.o sshconnect1.o sshconnect2.o sshtty.o readconf.o clientloop.o -L. -Lopenbsd-compat/ -lssh -lopenbsd-compat -lutil -lz -lnsl -lcrypto -lcrypt /bin/ld: warning: cannot find entry symbol _start; defaulting to 0804b208 /usr/lib/libutil.so: undefined reference to `atexit' make: *** [ssh] Fehler 1 The second line shows the reference to -lutil. So, libutil is still enabled. The forelast line expresses that the reference to atexit was not working. But you write: - (bal) We no longer use atexit()/xatexit()/on_exit() So, if not you who else? How the user shall answer this? To him this is only paradox. So, if not your product uses atexit then at least a product which you refer to. Then, it is still your problem since that product works otherwise. Then you write this: Maybe what Linux.. Because I have not seen any problems compiling under Redhat nor Mandrake. Why you ask me this. If I was telling you that I use SuSE would you then know the solution. Or would you just complement me to my distributor. So, only RedHat users can get your support? Is RedHat the measure for free source? To tell you what I am using: Free Source as mother GNU has created it. Did I loose support now? I don't use any distribution and, though RedHat took over being the measure, just the source still works fine. My system consists of thousands of products like: GNOME, Mozilla, Gnumeric, Evolution, Apache 2.0, PHP4, PERL, Python, MySQL, PostgreSQL, OpenSSL, the compilation environment (as I wrote earlier I tested atexit and my little c prog worked out of the box). So, don't tell me it's my system. Some product that Open SSH refers to seems to need atexit. Since that product works otherwise the culprit could be the autoconf system. Maybe your Makefile does not set environment variables correctly when spitting out gcc commands. I did not append a complete make log but send it to the maintainer directly because bugzilla did everything except sending the message when an attachment was specified. I lost an hour waiting and trying and waiting and trying.
you might want to remember that the developers are volunteers here. they are providing support because they want to. your belligerent attitude will not exactly encourage anyone to look into your problem. you should always provide what OS you are running with every bug report. sometimes others may have come across the same problem. ben's statement that it works on mandrake and redhat wasn't a shot at you, but an attempt to better pigeonhole exactly where the problem is. damien's response looks strange in the order it appears to have been received, but remember that email can cross and he may have sent it quite a while before you sent your response. so, at risk of getting whapped by someone- please chill out. these are volunteers and shouldn't be treated badly when they are trying to figure out a problem you've waited a year on.
You might remember that people now depend on your product because since you offered it for free many sites just force a user like me to use your product. Thus, you carry some responsibility. And, that your supporters are volunteers does not mean that they can write what they want while I have to be quiet. As you stated correctly I am now without OpenSSH since over a year (on my main system, my old SuSE uses an older version which is insecure). And, my first report was posted a month ago now. Since then I only got one creative message stating that the problem was solved which is not true. Now your volunteers do so as if the problem does not even exist. I only got "belligerent" (what is not true, I just tried to get sense into the communication) because your volunteers were, lets say, the opposite of wordy. I am in need and would apprecheate some helpful contact. Maybe you consider this. Or do I have to dance for your volunteers? When they are willing to help?
Here is what ver_linux tells about my system: Linux orgc 2.4.16 #3 Mit Mär 27 16:34:35 CET 2002 i686 unknown Gnu C 2.95.3 Gnu make 3.79.1 binutils 2.11.2 util-linux 2.11i mount 2.11i modutils 2.4.12 e2fsprogs 1.25 PPP 2.4.1 Linux C Library 2.2.4 Dynamic linker (ldd) 2.2.4 Procps 2.0.7 Net-tools 1.60 Kbd 1.06 Sh-utils 2.0
If you edit your Makefile and remove -lutil from the LIBS= line, does it compile?
ermm.. This looks like a glibc issue. Because atexit() is part of glibc. [mouring@newton ~]$nm /usr/lib/libc.a | grep atexit [..] atexit.o: [..] 00000000 T atexit So I'm not sure why libutil would be kicking out that error. This is really not a valid test of atexit(), but can you try compiling this? It would at least show us if atexit() support even exists in glibc. void test(void) { printf("Hello World\n"); } int main(){ atexit(test); } BTW, I don't care to get into distro wars.. I work on servers from Linux/Redhat to Solaris to NeXTStep. To tell you the truth they all piss me off for all different reasons. =) And I'm sorry if you feel we are not solving your problem, but just randomly disabling libraries is the wrong approach to solve this problem. - Ben
Yo ! I have seen this before. It happens when some of your libraries a re built with gcc before < 3.0 and some with gcc >= 3.0. It also is related to the removal of atexit() from the libc.so file around 26 Feb 01. It has nothing to do with openssh. Take this up with the GCC and Glibc folks that created this mess when they removed atexit from libc. Your best fix is to recompile all the libaries you think are being used with the current compiler. That is not always sufficient for some projects like X11 that do strange and unnatural things with dynamic libs. If you search on google you will see this is drving everyone crazy...
This from the glibc-2-2.5 FAQ > 3.23. I get "undefined reference to `atexit'" > > {UD} This means that your installation is somehow broken. The situation is > the same as for 'stat', 'fstat', etc (see question 2.7). Investigate why the > linker does not pick up libc_nonshared.a. It was noted earlier that atexit is in libc.a, but by default the linker uses libc.so and that does NOT contain atexit() in newer versions. RGDS GARY
Some examples why I got angry: First: There was a Gert who flamed me with an email directed to my private address. I already answered. Second: Ben writes (at this place) that I should try a small prog to see if "atexit() support even exists in glibc". He tries to help but he did not read my messages. I already wrote a test prog and atexit works fine on my system. Nevertheless: thanks for being helpful. Third: gem@rellim.com wrote about the problem with changing compilers. Again somebody is creative, thanks. I did not mix compilers, still use the same for all progs on my system (would not mix). He then writes about the newer glibc releases and that they do not include atexit anymore (same problem like with Ben): "It has nothing to do with openssh." This is not true because if the maintainers of glibc change the rules then you have to follow since you, and with you your users, depend on that. He then writes that I shall: "Take this up with the GCC and Glibc folks that created this mess when they removed atexit from libc." I think this was just a call for revolution but, please, your product wants to find something though you already know it is gone. Why I shall struggle with the maintainers of glibc then? And, again, atexit is there and accessible. Fourth: Gary writes: "It was noted earlier that atexit is in libc.a, but by default the linker uses libc.so and that does NOT contain atexit() in newer versions." That is a hint into the right direction but, as I wrote, my atexit works fine by just including stdio.h. So, ok, its good information still. And, in the end, I wonder what you are talking about because you now want me to find atexit though Ben writes that: "We no longer use atexit()/xatexit()/on_exit()". Better is finding the right library which still wants to use atexit, and checking if it's yours or belonging to another project. Then, I could actually bore them. tim@multitalents.net actually helped me most. He wrote: "If you edit your Makefile and remove -lutil from the LIBS= line, does it compile?". I tried out and came a bit more far. Now the error is: /bin/ld -o ssh ssh.o sshconnect.o sshconnect1.o sshconnect2.o sshtty.o readconf.o clientloop.o -L. -Lopenbsd-compat/ -lssh -lopenbsd-compat -lz -lnsl -lcrypto -lcrypt /bin/ld: warning: cannot find entry symbol _start; defaulting to 0804b1f8 /usr/lib/libnsl.so: undefined reference to `atexit' As you can see, still a library of the glibc package wants atexit. And, I think that is the real problem. The glibc library will always want to find atexit. I tell again here, because you seem to forget this quickly, that I already checked for atexit and it is existing and working. Thus, I still believe that there is a problem with the setting of search paths through variables. I am shure you included the stdio.h, otherwise more would not work. One thing on your support really works good now: the frequency ;)
I sent this in private email, but I'll restate this for the record. Working off of Gary's comment.. Try adding /usr/lib/libc_nonshared.a to your LIBC= and see if that resolves it. If it does then I'm unsure what we (as the portable group) can do to find to correct this since that should be linked by default by the compiler. - Ben
Yo All! > Try adding /usr/lib/libc_nonshared.a to your > LIBC= and see if that resolves it. Or link statically. Since you know atexit() is in your libc.a and I bet it is NOT in your libc.so. I'll also bet you updated your libc about the time you first had problems compiling OpenSSH. You said you had problems for about a year and glibc had this bad change about 14 months ago. RGDS GARY
Wow, much going on now :) I already read Ben's message about finishing the flame war. I don't want to flame further, just state what I state now since three messages. Gert, again, wrote me about his "bigger dig", and appended some details about his good relations to RMS, Linus, etc. (He enforced a useless flame war about the term free source and its impact on the user in his previous email). I definately don't agree with: "Always remember: this is volunteer effort. You don't pay for it. The ONLY right you have is "help yourself"." He always sticks to free as in free beer. He should proof his opinion with RMS again. But, the most disturbing point is this "do or dare" opinion. I did not criticize free source, was just angry about YOUR support. This is something individual and not defined by the GPL, FSF, or the community. And, as a dependent user *I have the right* to find out on whom I am dependent. But, ok, have said this, lets skip now. I now see that over months nothing happened here till I got more harsh. Now, everybody takes part with his comments. So, there is never enough time for support but always for flame wars. This is a bad truth not only for this project. I agree to ben: "Gert, if anyone should be getting pissed by his messages it should be me or maybe Damien." There are actually some more helpful people. Ben, I want to excuse because you seem to care. "If you take my original message or this one as being 'condensing' or 'elitest' non-sense then please stop reading into my writing because I'm not. I'm honestly trying to find the right question or bit of information that may help." It wasn't clear at the beginning because the first messages were, say, not that logical. Now, there still is the problem with atexit(). You, and others, asked if I could try appending /usr/lib/libc_nonshared.a to LIBS=. This way of co-operation I seekd for. I tried but still the same referencing error with libutil. As I wrote, the error does not happen to my own prog and thousands of other free source products. Thus, I don't think that glibc messes up something. It still seems to be a referencing problem of the configuration and installation procedure. I did not have to link to any place to let my own prog run. GCC found everything needed and atexit worked. This is my experience, what shall I say. You say: "I've not had anyone bring up this issue and so it leads me to believe there is some uniqueness of your system. Not saying that it is broken, but something unique enough to trigger it. What that means is will be impossible for anyone, but you to resolve it since no one else can reproduce it." I think this is at least a bit incorrect because of my overall good experience. My system is just non-manipulated source, as it should be. It works fine and clean. I was never happier about a linux system. I love it (slime). So, even if this is unique to you, still your product does something that others do not. Means: my system is only unique and triggering to openssh. So, maybe, openssh does something unique under thousands of products (we can now start the next flame war). I still stick to autoconf or similar as the culprit. As a consequence, it does not help to bore the glibc-people. They will send me back to you. I am not a jo-jo and don't want to be pushed from here to there. Maybe you can now start just giving some hints. Maybe I could send config.cache and makefiles to you. Maybe there are other things I could try. Please, you are more experienced in such big projects. You may know better what to do now. I would ask Gert because of his relations to the masters of free source but this is not a good deal, I believe.
broken libc = notabug
Mass change of RESOLVED bugs to CLOSED