ql-dist:dependency-tree signals no-applicable-method

[m-n]

On client 2014-01-30

dependency-tree expects calling ql-dist:find-system on an element of a system's required-systems to return a system object. This expectation is violated for at least "asdf", which leads to dependency-tree being called recursively on nil and signalling. Using asdf-finalizers as an example because it only has the one required system:

;(in-package ql-dist)
(required-systems (find-system "asdf-finalizers"))
("asdf")

;(in-package ql-dist)
(find-system "asdf")
NIL

;(in-package ql-dist)
(dependency-tree "asdf-finalizers") 

There is no applicable method for the generic function:
  #<STANDARD-GENERIC-FUNCTION DEPENDENCY-TREE #x3020008D131F>
when called with arguments:
  (NIL)
   [Condition of type CCL:NO-APPLICABLE-METHOD-EXISTS]
    Restarts:   [snip]

Backtrace:
  0: (#<CCL::STANDARD-KERNEL-METHOD NO-APPLICABLE-METHOD (T)> #<STANDARD-GENERIC-FUN$
      Locals:
        CCL::GF = #<STANDARD-GENERIC-FUNCTION DEPENDENCY-TREE #x3020008D131F>
        CCL::ARGS = (NIL)
  1: (NIL #<Unknown Arguments>)
      [No Locals]
  2: ((:INTERNAL (DEPENDENCY-TREE (SYSTEM))))                                                  Locals:                                                                                #:G8125 = (NIL)
         #:G8126 = (NIL)
        #:G8129 = ("asdf")
        #:G8127 = "asdf"
        SYSTEM = #<SYSTEM asdf-finalizers / asdf-finalizers-20130615-git / quicklisp$

[quicklisp]

dependency-tree is meant only for objects of type ql-dist:system, not ASDF systems. It can't be used on projects not indexed within a dist (e.g. your own projects).

[quicklisp]

It could be fixed to catch that NIL error, though...

[m-n]

To be clear, the above error is signaled from a code path that uses a ql-dist:system (see frame 2 of the trace).

I suspect that the actual bug here is that "asdf" is allowed as a member of a ql system's required-systems, but (ql-dist:find-system "asdf") does not produce a system. I believe changing that would fix dependency-tree.

If that is actually intended, then would the ideal behavior of dependency-tree be to remove the questionable nodes from its output?

[quicklisp]

Ugh. That's an indexing issue. ASDF shouldn't be in depends-on lists. It doesn't make a lick of sense. I'll have to think about it a bit more.