If the CVS repository has a directory that did contain some files but is now empty, that is, the files have been 'cvs remove'd, then if the working copy contains an empty directory of that name 'cvs add' gets upset.  Here is how to reproduce:

# First make an empty CVS project, for completeness this is how:
mkdir $HOME/cvsroot
export CVSROOT=$HOME/cvsroot
cd
mkdir scratch
cd scratch
cvs init
mkdir test
cd test
cvs import -m msg test test0 test1
cd ..
rmdir test
cvs checkout test
cd test

# Now we are in the working copy of the empty project.
# Add a directory, add a file in that directory, then remove it.
mkdir a
cvs add a
cd a
touch b
cvs add b
cvs commit -m msg
rm b
cvs remove b
cvs commit -m msg

# Now in our working copy the directory a/ is empty except for
# the CVS special files.  We can remove it from the working copy
# and 'cvs update' will not put it back.
cd ..
rm -r a
cvs update
cvs log

# Here is the bug: if you then recreate the directory a/ then
# 'cvs log' gets upset.  But this empty directory has no effect
# on 'cvs diff', for example, so it shouldn't affect 'cvs log'.
mkdir a
cvs log

Expected result: the same log output as if the directory a/
didn't exist.  Or at least, some reasonable warning of what is
happening and 'cvs log' continues to write a log.

Actual result: got the error message
cvs log: in directory `a':
cvs [log aborted]: there is no version here; do `cvs checkout' first