Bug#300119: libc6: LD_DEBUG=versions can lead to segmentation fault
On 03/21/05 02:06, GOTO Masanori wrote:
In this bug report, we don't know which vi is used. Your vi links a
lot of libraries. I don't know what the problem is, but I guess (1)
you use non-debian vi which is not linked correctly (2) you use
invalid dynamic linker setting: for example, LD_PRELOAD or
/etc/ld.preload and so on.
No, no, no.
First of all, my system is AFAIK plain old sarge.
The pnglib is, I believe, the first and only thing
I have installed from source, as opposed to just
installing debian pkgs using dselect.
Secondly, as I explained in my original note, I
mentioned vi only as an example of an ordinary
innocuous program. The list of programs that
exhibit the same segfault (when LD_DEBUG=versions
is set) include
vi (which is a just symlink to vim 6.3.58)
Mozilla Firefox 1.0.1
ImageMagick 6.0.6 (including identify, display, etc. etc.)
ESP Ghostscript 7.07.1
... do I really need to go on ... ????
So unless you are going to claim that I _unknowingly_
installed non-Debian versions of all those programs,
you will have to consider that there is a systematic
One thing they all have in common is that the last two
lines before the segfault are always
nnn: checking for version `GLIBC_2.0' in file /lib/tls/libm.so.6
required by file ....
nnn: checking for version `PNG12_0' in file
/usr/lib/libpng12.so.0 required by file ....
For that matter, the following command prints a segfault message
(apparently coming from a subprocess it forked):
LD_DEBUG=versions ldd /usr/lib/libqt-mt.so.3.3.3
It is a typical bash message, i.e.
/usr/bin/ldd: line 1: nnn Segmentation fault
I think this bug report is not essentially related with the current
glibc, so I close it.
I'm shocked. I cannot imagine why you would think that.
If you want to say the bug is not reproducible on your
system, PLEASE SAY THAT. Then we can discuss why not.
Don't go chasing issues (like "vi") that I told you were
Did you even *try* compiling libpng from the source
tarball at libpng.org, to see if you could reproduce
the symptoms? Would it help for me to make available
my already-compiled copy, so you can "ldd" it yourself?
John, if you plan to use versioned symbols, you
possibly want to read Ulrich Drepper's nice document (try to use
I didn't "plan" on using versioned symbols. I didn't
know I had a choice in the matter. I just installed my
system using what I thought were obvious, innocuous
choices using dselect, and what I got is whatever was
in the debian packages.
Also, as I explained in my initial report, it seems to
me turning on LD_DEBUG=versions should not cause a
segfault _no matter what_ the input is. How can you
think even for an instant that this is
"not essentially related with the current glibc"?