Segfault in dependency checking...

[James Olin Oden <joden malachi lee k12 nc us>]

Hi Jeff,

I am not sure if this is my bug or yours, so this is why 
I am sending to the list first.

Anyway I was trying to upgrade two rpms (this was as a test 
by the way, to make sure I had not broken anything), and when
I do this I get a core dump.  Now my first thought was that I caused
it except its in the dependcy checking which is before anything I had
touched (actually, it was before it ever got to rpmtsRun()).  Here is 
the output from gdb showing the back trace:

	#0  0x420796b8 in strcmp () from /lib/tls/libc.so.6(gdb) bt 
	#0  0x420796b8 in strcmp () from /lib/tls/libc.so.6
	#1  0x40047bfd in indexcmp (one=0x807b420, two=0x29) at rpmal.c:538
	#2  0x40048364 in rpmalAllSatisfiesDepend (al=0x807b6a0, ds=0x80839a8, keyp=0x0)
	    at rpmal.c:754
	#3  0x400484f0 in rpmalSatisfiesDepend (al=0x29, ds=0x29, keyp=0x29) at rpmal.c:800
	#4  0x4003053f in unsatisfiedDepend (ts=0x80717f0, dep=0x80839a8, adding=0) at depends.c:445
	#5  0x40030bdf in checkPackageDeps (ts=0x80717f0, pkgNEVR=0x807c278 "openssl-0.9.7a-5", 
	    requires=0x80839a8, conflicts=0x0, depName=0x8083b18 "/bin/sh", tscolor=0, adding=0)
	    at depends.c:604
	#6  0x40030e2e in checkPackageSet (ts=0x80717f0, dep=0x8083b18 "/bin/sh", mi=0x8080d08, 
	    adding=0) at depends.c:696
	#7  0x40030ed4 in checkDependentPackages (ts=0x80717f0, dep=0x8083b18 "/bin/sh")
	    at depends.c:723
	#8  0x40032ac7 in rpmtsCheck (ts=0x80717f0) at depends.c:1688
	#9  0x40051ada in rpmInstall (ts=0x80717f0, ia=0x805b9b8, fileArgv=0x8069a18)
   	    at rpminstall.c:628
	#10 0x0804b459 in main ()
	#11 0x420156a4 in __libc_start_main () from /lib/tls/libc.so.6

Now when I go to the last bit of code in rpm that was called before the 
segfault, I see a simple string compare between to strings that are 
comming from availableIndexEntry structures.  My guess is that somehow
one of these is not getting initialized, but when I look at the code in
depends.c and rpmal.c its not exactly obvious why they are not being 
initialed.

Here is the test I am running:

	1) install an rpm works-1-1.i386.rpm.
	2) upgrade to works-1-2.i386.rpm and broken-1-1.i386.rpm. 
	   broken has a scriptlet that returns an error in it to 
	   test my autorollback patch.

Now I am making a copy of an existing rpm database, and setting __dbpath
to the location of this alternate database.  Here are the commands
I am actually running:

	/bin/rpm -Uvh \
		--define '_dbpath /root/rpm/rpm-test-harness/db' \
		/root/work/build/RPM/RPMS/i386/works-1-1.i386.rpm
	/bin/rpm -Uvh \
		--define '_dbpath /root/rpm/rpm-test-harness/db' \
		--define '_repackage_dir /root/rpm/rpm-test-harness/rpdir' \
		/root/work/build/RPM/RPMS/i386/broken-1-1.i386.rpm \
		/root/work/build/RPM/RPMS/i386/works-1-2.i386.rpm


Hmmm...I just figured it out, I was pasting in the spec files for your
perusal, and realized that works-1-1 had:

	Provides: /bin/sh

in it.  This has hung around from when I was not testing with a fully 
populated database.  Anyway, when I take that out the segfault goes away.

So it looks like their may be a general problem when two packages provide 
the same thing, and then someone later goes to upgrade with something that
requires that thing.  Now, I know that someone should not do that, but 
probably this should be handled better.

Anyway, I will bugzilla this now that I figured out what is going on.

Cheers...james