This is really just a UI problem.

I did a
  darcs get http://code.haskell.org/GenI --context my-context
and got distinctly bug-like output.

Copying patches, to get lazy repository hit ctrl-C...
darcs: failed to read patch in get_extra:       
Mon Apr 13 11:10:36 BST 2009  Eric Kow <E.Y.Kow@brighton.ac.uk>
  * Remove unused 'grammars' directory.
Mon Apr 13 11:10:36 BST 2009  Eric Kow <E.Y.Kow@brighton.ac.uk>
  * Remove unused 'grammars' directory.
Patch not stored in patch bundle:
Mon Apr 13 11:10:36 BST 2009  Eric Kow <E.Y.Kow@brighton.ac.uk>
  * Remove unused 'grammars' directory.
Perhaps this is a 'partial' repository?

On further inspection, it turns out that the problem was not a bug
but that I was trying to do a darcs get using a context file which
contained patches that I had not even pushed yet.

In other words, I was asking darcs for patches that didn't even exist in the
repo.  The real context file is attached here as "real-context"

It would be nice if darcs could give more helpful output in this
scenario, but I imagine this would be tricky to do.

One thing to look into, though: why do we print the patch name
three times?

That'll teach me to do demos when I'm tired :-)

Note that issue1505 may help, but it won't completely solve the problem for this
particular issue.

I guess in my ideal darcs, we would be able to say something like

  This context file contains patches which are not in the source repository
  ...

Whoever implements this should probably write a regression test first :-)