bug#34350: 27.0.50; ediff-revision broken with SVN backend + non ascii chars both in directory and in filename

[Stefan Monnier]

>> Why should reading from a pipe be slower than from a file?
> Because we read in small chunks,

I understand that what we currently do, but is it really inevitable?
I know you can't transfer an arbitrarily large text with a single `write`
at one end of a pipe and a single `read` at the other (contrary to what
happens with files), but it should be possible to use chunks large
enough that the efficiency is comparable.

> and need to enlarge the gap each time.

Yes, this used to be a real problems in set-buffer-multibyte (with O(N²)
behavior), but nowadays it should be a relatively small constant factor.
Of course, if needed, we could provide some way for the Elisp code to
"preallocate" space to avoid resizing the buffer too many times.

[ Goes on dreaming about a buffer representation made of a tree of
  immutable text-chunks... ]

> OK, so we basically agree.  I will try to remove the cruft from some
> of that, thanks for the feedback.

Right.   Maybe we could improve the backend API, but that's a side
problem: the real pressing problem is just cruft and should not be too
hard to fix.


        Stefan