Re: Does network data really cons strings?

[Kevin Rodgers]

Jesper Harder wrote:

> Kevin Rodgers <ihs_4664@yahoo.com> writes:
> > Jesper Harder wrote:
> > > I was a bit surprised to see that network data (or any data received
> > > from processes) conses Lisp strings.
> > >
> > > Is it just a quirk of `memory-use-counts' or is it for real?  Why
> > > does it need to cons Lisp strings if the data is only inserted in a
> > > buffer?
> >
> > So that the data is available to the process filter?
>
> Right, it would need to cons strings if there are filters.  But it
> also happens when there are no filters.


That'd be a nice optimization: don't bother allocating strings for processes

that don't have a filter.


> Hmm, I now notice that it also conses cons cells like mad:
>
> (let (m1 process)
>     (with-current-buffer (get-buffer-create " *test*")
>       (erase-buffer))
>     (setq process (open-network-stream "test" " *test*" "news.gmane.org" 119))
>     (setq m1 (memory-use-counts))
>     (process-send-string process "LIST\n")
>     (sit-for 50)
>     (process-send-string process "QUIT\n")
>     (mapcar* '- (memory-use-counts) m1))
>
> => (48055 3 0 0 509059 2 0 2063)
>
> (with-current-buffer " *test*" (buffer-size))
> => 252123
>
> Roughly two char-cells for every byte of data received!  That seems
> rather excessive ... and I wonder what all those cons cells are used
> for.


Yikes, maybe a bug report is in order...


> Emacs 20.5 seems to do _far_ better for the same data (on a faster
> network connection, though):
>
>      (4103 0 0 0 4168 0 0)


--
Kevin Rodgers