bug#4446: remove ###autoload from ediff-hook.el

[Dan Nicolaescu <dann@ics.uci.edu>]

Stefan Monnier <monnier@IRO.UMontreal.CA> writes:

  > >> BTW, now that I think about it: is it really necessary to skip those
  > >> files?  I mean, what is the impact of having such redundant autoloads in
  > >> loaddefs.el?
  > 
  > >> From memory as I can't verify at the moment: the download size of the
  > > emacs source is increased by > 10K, and the binary is 10KB bigger.
  > 
  > I understand where the 10K of download-size comew from, but do you know
  > why it also impacts the final binary size?  I mean: the objects and
  > variables created by these redundant autoloads should be overwritten by
  > the subsequent load of the actual files, so the only impact they should
  > have is if the GC ends up unable to reuse the corresponding memory cells
  > (and unable to return them to the system), right?  Or are these
  > chunks of autoload-data placed in the pure-space?

Sorry, I don't know.

Another thing that I observed (and hope not to forget until I can
investigate properly): a lot of free conses and strings appear after the
"finding pointers to doc strings" phase in loadup.el.  

It seems that we might have to try to avoid allocating and setting the
doc strings while dumping.