From Bugzilla Helper:
User-Agent: Mozilla/4.0 (compatible; MSIE 5.5; Windows NT 5.0)
Description of problem:
Intel C/C++ compiler is delivered with shared libraries. It itself needs them
to run. In addition, programs built by Intel compiler with default options need
the shared libraries too. Intel shared libraries are searched in directories
pointed out by the environment variable LD_LIBRARY_PATH. Before building a
package, a user sets this variable properly. During building, the packages may
only add something to the contents of LD_LIBRARY_PATH but not replace it. Using
LD_LIBRARY_PATH to point out some shared libraries is a commonly used method in
Linux, and the packages should treat the variable carefully. Unfortunately,
some packages don't follow the rule and replace LD_LIBRARY_PATH totally during
building (of course, I mean the variable's copy that is local for the building
task). These packages are: libtabe, procmail, openssl and openssl096. I've
already written a bug report on libtabe (# 58092) and I'm going to write bug
reports on others. The following is a more detailed description of the problem
relative to procmail:
While building, the script "src/autoconf" generates and runs several programs
(all named _autotest.c) to determine experimentally some parameters of the
current system. One of the programs is to determine the maximum number of
program arguments. It resets the current environment to just one
variable "PATH". See "src/autoconf", line 1351:
(environ is a global variable pointing to an array of strings with environment
Then the program tries to run itself again by calling execv(argv,...);.
However, the new instance of the program cannot be run because it needs
LD_LIBRARY_PATH in the environment. Unfortunately, I can't propose a bugfix
because I don't quite understand why the program resets the environment.
Certainly resetting environment doesn't influence the program's output much. On
my computer, the original output of the program built by gcc is the following:
#define MAX_argc 7588
When I comment the resetting operator (/**(environ=nenv)="PATH=.";*/ ...), the
output changes but just a little:
#define MAX_argc 7625
Version-Release number of selected component (if applicable):
Steps to Reproduce:
If you don't have Intel compiler (icc) you are unable to reproduce the problem.
This compiler is not free. If you really need it to fix the problem, let me
know and I'll try to help you. If you have the compiler, the steps to reproduce
the problem will be the following:
1. Make sure that all environment necessary for icc is set.
2. Make sure that icc will be called instead of gcc, g++, cc, cpp and c++. I
recommend you make links to icc with these names, put them into some directory
and add this directory to the beginning of the PATH variable.
3. Install procmail from the .src.rpm file (-i option of rpm). Make sure you
have enough permissions to do that.
4. Try to build the .spec file (-bc option of rpm) and get an error message.
Actual Results: The package won't build. In the output (near the end), you may
see the following error message:
error while loading shared libraries: libcxa.so.1: cannot open shared object
file: No such file or directory
Expected Results: The package should build with no problem.
1) Is this still a problem with procmail 3.22 from Rawhide?
2) Why not just put the path of the Intel libraries in /etc/ld.so.conf?
Here are the answers:
1) Yes, this problem presents in procmail 3.22 from Rawhide
2) From compiler's point of view, there is no difference between
LD_LIBRARY_PATH, ld.so.conf or ldconfig using. The reason why we prefer
LD_LIBRARY_PATH is that setting of environment variable does not require root
permissions, when updating of ld.so.conf does. So, using of LD_LIBRARY_PATH
makes compiler or libraries development and usage simpler. Probably we will
take another way for compiler distributions. Nevertheless, procmail uses a
risky programming style while truncates its whole environment.
If you have a non-intrusive clean patch to our package
then I'm happy to accept, but since we don't support icc
I don't have time to fix this.
Closing for now.