Diagnosing a curious minibuffer problem (proliferating, weird initial contents)

[florian@fsavigny.de (Florian v. Savigny)]

Dear wizards,

I need some assistance on how to pinpoint, diagnose and/or reproduce a
problem with minibuffer initial contents. I tend to think that I
caused this problem myself, but since I have deactivated (i.e. moved to a
file that is not loaded) the function which I would most consider
suspicious, I am at a loss.

The symptoms are as follows:

- When I start Emacs, and for a good deal of activity, everything is
  normal.

- After a while (and I cannot specify it more exactly; the moment when
  it starts seems unpredictable), commands which use completing-read
  (even M-x, but also functions which I have written myself) start to
  have inappropriate stuff as initial content. As far as I can
  remember, this inappropriate stuff is all file names.

- If this goes on longer, more inappropriate stuff is appended to the
  initial content, i.e. after a while, e.g. M-x offers you a long
  string of concatenated file names. You can either choose to delete
  this long monster, or you can get rid of it by typing M-p, which
  gives you shorter minibuffer contents.

- and so on (the monster gets longer and longer)


[speculation]

To me, this looks as if functions which use completing-read consult
some variable to decide what to use as INITIAL-INPUT, and I have
somehow managed to systematically clutter this variable. This is a bit
weird, though, because they usually offer just completion, no intial
contents. And it's just one possibility anyway.

[/speculation]

I have no idea how to get to the root of the problem. I cannot edebug
completing-read, in any case. (That is what I would try otherwise,
anyway.)

Can anybody give me any advice on how to pinpoint, isolate, reproduce,
provoke this problem? I would be awfully grateful, because my "Emacs
system" is complex and I need to run it every day.

Thank you so much!

Best regards,

Florian

-- 

Florian von Savigny
Melanchthonstr. 41
33615 Bielefeld