Test case:

darcs init R
cd R
mkdir c
darcs record -lam add-c
darcs move c d
darcs record -am move-c-d
touch d/f
darcs add d/f
darcs obliterate -p move-c-d
cat _darcs/patches/pending
# content: addfile ./d/f instead of ./c/f
darcs whatsnew
# result: rmdir ./d (WTF?)

This bug is only in screened, not in 2.16.5. I suspect this patch to be the 
culprit:

patch 4f9ac2caf52b420538798dddc15f4211c1fdcaa8
Author: Ben Franksen <ben.franksen@online.de>
Date:   Tue Mar  2 14:13:40 CET 2021
  * obliterate, rebase suspend: pass NoUpdatePending to tentativelyRemovePatches
  
  This avoids the redundant use of the (unsafe) tentativelyAddToPending to
  remove what gets added to pending with YesUpdatePending.

Indeed I have seen strange things when using rebase pull: after suspending 
conflicting patches it sometimes complains about conflicts with unrecorded 
changes even though there are no unrecorded changes in the repo. When I abort 
the operation at this point, I am left with lots of pending changes.

The problem I mentioned with rebase pull (conflicts with non-existing 
unrecorded changes) is unrelated, but indeed the same bug as with 
obliterate can be observed with rebase suspend.

See patch2285