[ESS-bugs] ess-mode 15.09-1 [elpa: 20151013.1025]; iESS Performance Degrades Quickly with Buffer Size

[Martin Maechler]

>>>>> Jason Morgan <jason.w.morgan at gmail.com>
>>>>>     on Wed, 14 Oct 2015 09:46:58 -0400 writes:

    > On Wed, Oct 14, 2015 at 6:16 AM, Martin Maechler
    > <maechler at stat.math.ethz.ch> wrote:
    >> 
    >>> I have just updated to ESS 15.09-1 and I am noticing a
    >>> dramatic change in performance. For example, if I print
    >>> a large-ish data.frame or matrix, the iESS process
    >>> becomes very sluggish, almost unusable. I have
    >>> replicated the problem when starting emacs with -q (and
    >>> then manually loading ESS). I thought it may have
    >>> something to do with Ubuntu or GTK, so I also replicated
    >>> the problem from the command line (outside of the
    >>> windowing system). The problem is easily replicated on
    >>> my system: all I have to do is run something like
    >>> matrix(rnorm(1e4)).  Then, every time I do anything in
    >>> the iESS instance, the emacs usage spikes to 25-35%
    >>> CPU. Even simply hitting Enter causes a lag and CPU
    >>> spike. I am on Emacs 24.4.1, but have replicated on
    >>> Emacs 24.3.
    >> 
    >> Interesting.
    >> 
    >> Did you try ESS 15.09 before that ?
    >> 
    >> Honestly, I also had doubts yesterday when working with
    >> the latest ESS version (i.e., 15.09-1) and saw extreme
    >> slowness myself.

    > Thank you for the quick response.

    > I didn't install 15.09 until 15.09-1 was out. I was still
    > on 14.11, I believe. I am willing to try to find the
    > problem, but I'll need guidance on how to profile, etc.

Actually, I just tried myself, also with 15.09, and
very embarrassingly:

     Houston, we'got a problem !

Try our old ESS internal (ess/etc/slow-long-output.R) :

###  Example modified from Paul Johnson's post to ESS-help, Nov.25, 2013
op <- options("max.print")

##' construct (random) character vector of length n,  of about 10 characters each
mkCh <- function(n) {
    if(getOption("max.print") <= n) options(max.print = n + 1) ## <-
    if(n == 10) stop("error's message with apostrophe - work's in developer mode")
    N <- pmin(26,pmax(1, rpois(n, 10)))
    cbind(vapply(N, function(m) paste(sample(letters, m), collapse=""), ""))
}

set.seed(1)
y <- mkCh(9999)
system.time(print(y))

## which in current ESS 15.09-1  took about 13 seconds for me,
but only less 1.9 secondes in  ESS 14.09

Or the *much* more extreme (for ESS 15.09-1):

set.seed(23456)
y <- mkCh(50000)
system.time(print(y))


## which basically freezes emacs after printing an initial part

--> so we are back at the very bad behavior Vitalie may
remember  we had in Decemeber 2013

ESS-core members: do look at

  https://stat.ethz.ch/mailman/private/ess-core/2013/007622.html


----

So indeed, Jason, we have a big problem, already in the release
ESS 15.09.

Thanks a lot for your report!
Martin