[Rd] Failed file("foo", "r") keeps file open? (PR#1333)

[hb@maths.lth.se]

On Thu, 28 Feb 2002, Prof Brian Ripley wrote:

> On Thu, 28 Feb 2002 hb@maths.lth.se wrote:
> 
> > Full_Name: Henrik Bengtsson
> > Version: R v1.4.0
> > OS: Solaris 8
> > Submission from: (NULL) (130.235.3.49)
> >
> >
> > When I am trying the following on Rv1.4.0 on Solaris 8, I get that the
> > "cannot close output sink connection":
> >
> >  % R --vanilla
> >
> >  tmpfile <- tempfile()  # Create a filename that does not exist
> >  for(k in 1:100) {
> >    cat(k, ": ");
> >    try(fh <- file(tmpfile, "r"))
> >    close(file("tmp.txt", "w"))   # Opens and closes a file for writing.
> >  }
> >
> > 1 Error in file(tempfile(), "r") : cannot open file `/tmp/fileR70d5S167e'
> > 2 Error in file(tempfile(), "r") : cannot open file `/tmp/fileR70d5S2781'
> > ...
> > 47 Error in file(tempfile(), "r") : cannot open file `/tmp/fileR70d5S345b'
> > Error in close.connection(file("tmp.txt", "w")) :
> >         cannot close output sink connection
> >
> > It always happens after 48 failed file openings.
> >
> >
> > To me it looks like failed 'file(tmpfile, "r")' does not correctly
> > clean up after itself. And since 'fh' is never assigned it can neither
> > be solved by a 'close(fh)'.
> 
> It does correctly clean up: it should leave a connection which is not
> opened.  That may not be the best design, but it is the design.
> 
> Take a look at closeAllConnections.

Accepting the design, calling file(tempfile(), open="r"), will, as
you say, create a "closed" connection as seen by
showConnections(all=TRUE) and it can be deleted by identifying the
connection id from the list and the closing it with

  fh <- getConnection(id)
  close(fh)

or one could call closeAllConnections(), which is not always what you
want. However, if one creates TOO MANY failed connections as above,
neither of the above works:

  % R --vanilla

  K <- 48; # K <- 47 is ok.
  tmpfile <- tempfile()
  for(k in 1:K) try(fh <- file(tmpfile, "r"))

will make [R] stall when closeAllConnections() is called. Doing Ctrl+C and
trying do anything, e.g. "ls() [ENTER]", will generate a "Segmentation
fault" or "0 [ENTER]" will generate a "Bus error". Trying to use

  fh <- getConnection(id)
  close(fh)

gives "Error in close.connection(fh) : cannot close output sink
connection". In this case there is not fatal quitting of [R].

I don't think it is a common thing to end up in this situation, but even
with the design you refer to, I still think something should be fixed
since it is possible to force [R] into an instable and unleavable state,
which is likely to crash [R].

> It's better to call file() and then open(), which will avoid the problem.

I definitely agree, now when you have told me.
 
> > FYI: Doing a cat(file="tmp.txt", k") instead of the close(file(...)) call,
> > will produce a "sink stack is full" error.
> >
> > FYI2: After the failure above, trying to call sink() will result in a
> >       "Segmentation fault" and [R] is stopped.
> >
> > Background: I ran into this problem after have been running the same
> > script several times over several days I eventually ran into the problem.
> > It was hard to identify the problem, but I believe the above example shows
> > the core of it and I could not track the problem further.
> >
> > Unfortunately, I do not have access to R v1.4.1 for Solaris 8.
> >
> >
> > -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
> > r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
> > Send "info", "help", or "[un]subscribe"
> > (in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch
> > _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
> >
> 
> -- 
> Brian D. Ripley,                  ripley@stats.ox.ac.uk
> Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
> University of Oxford,             Tel:  +44 1865 272861 (self)
> 1 South Parks Road,                     +44 1865 272860 (secr)
> Oxford OX1 3TG, UK                Fax:  +44 1865 272595

Thanks

Henrik Bengtsson

Dept. of Mathematical Statistics @ Centre for Mathematical Sciences 
Lund Institute of Technology/Lund University, Sweden (+2h UTC)
Office: P316, +46 46 222 9611 (phone), +46 46 222 4623 (fax)
h b @ m a t h s . l t h . s e, http://www.maths.lth.se/bioinformatics/




-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._