[R-sig-Debian] Debian packaging and openblas related crash when profiling in R

[Dirk Eddelbuettel]

On 11 December 2012 at 10:54, Paul Johnson wrote:
| Hello R-sig-debian and (hopefully) Dirk:
| 
| On Debian wheezy, I have the R packaging that CRAN (you) provide.  I

Not so -- I provide the distro package. Johannes provides the Debian
"backport" build via CRAN, Michael does likewise for Ubuntu.

| run into a little trouble while trying to fiddle with alternative
| BLAS.
| 
| I know you and I went around on this last year and I think perhaps
| I've found something wrong in the framework, or I've just done
| something wrong.
| 
| I installed the packages openblas-base and openblas-dev, and they do
| the "update alternatives" thing so that their shared library is at the
| right spot. I can't tell if R runs any faster, but it does run, which
| is encouraging.
| 
| Now here's the problem I see. Using Rprof causes R to crash, but only
| when openblas stuff is in the picture. I wondered what you think?
| 
| Here's the code to reproduce this:

May I humbly suggest we file this under user-error?  

R is famously single-threaded, OpenBLAS is not and I *strongly* suspect that
by profiling you are getting yourself into callbacks that R is simply not
built for. If you want to profile, constrain yourself to a single-threaded
blas such as reference blas or Atlas.

Dirk
 

| datM <- read.csv("http://pj.freefaculty.org/scraps/profile/missingDat.csv")
| library(Amelia)
| set.seed(12345)
| Rprof("Amelia-1.out")
| impA <- amelia(datM, m = 5, idvars="group")
| Rprof(NULL)
| summaryRprof("Amelia-1.out")
| 
| 
| Here's the crashed session
| 
| > datM <- read.csv("http://pj.freefaculty.org/scraps/profile/missingDat.csv")
| > library(Amelia)
| Loading required package: foreign
| ##
| ## Amelia II: Multiple Imputation
| ## (Version 1.6.3, built: 2012-06-21)
| ## Copyright (C) 2005-2012 James Honaker, Gary King and Matthew Blackwell
| ## Refer to http://gking.harvard.edu/amelia/ for more information
| ##
| > set.seed(12345)
| > Rprof("Amelia-1.out")
| > impA <- amelia(datM, m = 5, idvars="group")
| -- Imputation 1 --
| 
|  1
|  *** caught segfault ***
| address 0x11000088, cause 'memory not mapped'
| Warning: stack imbalance in 'length', 121 then 122
| Warning: stack imbalance in '==', 120 then 121
| Warning: stack imbalance in 'if', 118 then 119
| Warning: stack imbalance in '<-', 120 then 124
| Warning: stack imbalance in '<-', 97 then 94
| Warning: stack imbalance in 'min', 117 then 116
| Warning: stack imbalance in 'min', 117 then 118
| 
| Traceback:
|  1: am.inv(a = g11)
|  2: doTryCatch(return(expr), name, parentenv, handler)
|  3: tryCatchOne(expr, names, parentenv, handlers[[1L]])
|  4: tryCatchList(expr, classes, parentenv, handlers)
|  5: tryCatch(expr, error = function(e) {    call <- conditionCall(e)
|  if (!is.null(call)) {        if (identical(call[[1L]],
| quote(doTryCatch)))             call <- sys.call(-4L)        dcall <-
| deparse(call)[1L]        prefix <- paste("Error in", dcall, ": ")
|   LONG <- 75L        msg <- conditionMessage(e)        sm <-
| strsplit(msg, "\n")[[1L]]        w <- 14L + nchar(dcall, type = "w") +
| nchar(sm[1L], type = "w")        if (is.na(w))             w <- 14L +
| nchar(dcall, type = "b") + nchar(sm[1L],                 type = "b")
|      if (w > LONG)             prefix <- paste0(prefix, "\n  ")    }
|  else prefix <- "Error : "    msg <- paste0(prefix,
| conditionMessage(e), "\n")    .Internal(seterrmessage(msg[1L]))    if
| (!silent && identical(getOption("show.error.messages"),         TRUE))
| {        cat(msg, file = stderr())
| .Internal(printDeferredWarnings())    }    invisible(structure(msg,
| class = "try-error", condition = e))})
|  6: try(am.inv(a = g11), silent = TRUE)
|  7: amsweep(thetareal, c(FALSE, o[ss, ]))
|  8: emfred(x, thetaold, indx$o, indx$m, indx$ivector, indx$icap,
| indx$AMr1, indx$AMr2, AM1stln = AM1stln, returntype = "theta",
| priors = priors, empri = empri, collect = collect)
|  9: emarch(x.stacked$x, p2s = p2s, thetaold = NULL, tolerance =
| tolerance,     startvals = startvals, x.stacked$priors, empri = empri,
| frontend = frontend,     collect = collect, autopri = prepped$autopri,
| emburn = emburn)
| 10: amelia.default(datM, m = 5, idvars = "group")
| 11: amelia(datM, m = 5, idvars = "group")
| Warning: stack imbalance in '<-', 108 then 110
| 
| Possible actions:
| 1: abort (with core dump, if enabled)
| 2: normal R exit
| 3: exit R without saving workspace
| 4: exit R saving workspace
| Selection: 3
| 
| Process R exited abnormally with code 70 at Tue Dec 11 10:36:51 2012
| 
| I double-checked /etc/alternatives and as you can see it does now
| point to openbas
| 
| pauljohn at pjlap-124:alternatives$ ls -la libblas.*
| lrwxrwxrwx 1 root root 36 Dec 11 10:34 libblas.a ->
| /usr/lib/openblas-base/libopenblas.a
| lrwxrwxrwx 1 root root 37 Dec 11 10:34 libblas.so ->
| /usr/lib/openblas-base/libopenblas.so
| lrwxrwxrwx 1 root root 39 Dec 11 10:34 libblas.so.3 ->
| /usr/lib/openblas-base/libopenblas.so.0
| lrwxrwxrwx 1 root root 39 Dec 11 10:34 libblas.so.3gf ->
| /usr/lib/openblas-base/libopenblas.so.0
| 
| -- 
| Paul E. Johnson
| Professor, Political Science      Assoc. Director
| 1541 Lilac Lane, Room 504      Center for Research Methods
| University of Kansas                 University of Kansas
| http://pj.freefaculty.org               http://quant.ku.edu
| 
| _______________________________________________
| R-SIG-Debian mailing list
| R-SIG-Debian at r-project.org
| https://stat.ethz.ch/mailman/listinfo/r-sig-debian

-- 
Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com