Re: Release plans

[Stephen J. Turnbull]

Thomas Lord writes:

 > of markers, but because this will be a self-adjusting tree
 > all those operations will be O(1) in the expected case where
 > changes display pretty good locality.

I would advise you not to expect that case.  Experience in XEmacs
shows that some applications (the VM MUA in particular, but also the
old implementation of font-lock, dunno about jit-lock) like to run up
and down the buffer a lot.  AFAICS it's really important to have O(log
N) worst-case behavior.

Sorry I can't be much more precise about why this happens, I just know
that our algorithms that deal with extents (which we use to support
overlay-like behavior and text properties) are tuned for good locality
and lose badly in large buffers; they show up as time hogs in profiling.

It's possible it's something internal to the implementation of
extents, too, but I think a word to the wise is appropriate here.