bug#122: 23.0.60; Slowdown in directory scanning over time.

[Kenichi Handa]

In article <jwvwshc9qfw.fsf-monnier+emacsbugreports@gnu.org>, Stefan Monnier 
<monnier@iro.umontreal.ca> writes:

> > I've just modified the current code to avoid calling
> > Fget_buffer_create when we alreay have
> > Vcode_conversion_reused_workbuf.  But, I think it doens't
> > influence the efficiency that much because
> > Fget_buffer_create doesn't create a new buffer if a buffer
> > of the specified name already exists.

> But the title-problem showed that having many buffers (like 200 or so)
> can cause this buffer-name-lookup to take a long time.  The worst part
> is that having 200 buffers is not particularly excessive: I regularly
> have Emacs session with a hundred buffers or so.  So we should avoid
> this lookup loop whenever necessary.

The slowness came from the inefficiency of
generate-new-buffer-name when there exist many buffers of
the same base name.  To generate the 201th buffer, it scans
the buffer list 200 times (comparing names 200*201/2 = 20100
times), and the following get-buffer-create scans it again.

As there usually exist a few buffers whose name start with a
space, I believe get-buffer is very fast for finding such a
buffer.  But, even if that is so, in genenral, I agree that
we should avoid unnecessary call.

---
Kenichi Handa
handa@ni.aist.go.jp