R devel - Sep 2009 - Package that does not work until I re write the exactly the same code

Hi the list,

I am writing a package in S4 and I do not manage to understand a bug.
The "R CMD check" and the "R CMD build" both work. Here is
links to the
package (not on CRAN yet for the raison that I explain bellow):

http://christophe.genolini.free.fr/aTelecharger/kml_0.5.zip
http://christophe.genolini.free.fr/aTelecharger/kml_0.5.tar.gz

Then I install the package and I try an example:

--- 8< --------------
library(kml)
dn <- as.cld(gald())
kml(dn)
# XXX ~ Fast KmL ~
# Erreur dans as.vector(x, mode) : argument 'mode' incorrect
--- 8< --------------


So I make some verifications:
--- 8< ----
class(dn)
# [1] "ClusterizLongData"
# attr(,"package")
# [1] "kml"

getMethod("kml","ClusterizLongData")
# Method Definition:
#
# function (Object, nbClusters = 2:6, nbRedrawing = 20, saveFreq = 100,
#   maxIt = 200, trajMinSize = 2, print.cal = FALSE, print.traj = FALSE,
#    imputationMethod = "copyMean", distance, power = 2, centerMethod
=
meanNA,
#    startingCond = "allMethods", distanceStartingCond =
"euclidean",
#   ...)
#{
#   nbIdFull <- nrow(Object["traj"])
# ...... [[[The full code is available below]]]
# }
# <environment: namespace:kml>
#
#Signatures:
#       Object            
# target  "ClusterizLongData"
# defined "ClusterizLongData"
--- 8< ----

Everything seems fine. The code is correct.
So I copy-and-paste the code that I get with 
getMethods("kml","ClusterizLongData") and I affect it to a
function
"func". Then I define again the method "kml".

Then I run again the example that does not work before and it works...
Any explanations?

Christophe Genolini

--- 8< --------------------------
###
### Affecting to func the code that
getMethod("kml","ClusterizLongData")
delivers
###
func <- function (Object, nbClusters = 2:6, nbRedrawing = 20, saveFreq = 
100,
    maxIt = 200, trajMinSize = 2, print.cal = FALSE, print.traj = FALSE,
    imputationMethod = "copyMean", distance, power = 2, centerMethod =
meanNA,
    startingCond = "allMethods", distanceStartingCond =
"euclidean",
    ...)
{
    nbIdFull <- nrow(Object["traj"])
    convergenceTime <- 0
    noNA <- selectSupTrajMinSize(Object, trajMinSize)
    trajNoNA <- Object["traj"][noNA, ]
    nbTime <- length(Object["time"])
    nbId <- nrow(trajNoNA)
    saveCld <- 0
    scr <- plotAll(Object, print.cal = print.cal, print.traj = print.traj,
        print.sub = FALSE, col = "black", type.mean = "n")
    if (length(startingCond) == 1) {
        if (startingCond == "allMethods") {
            startingCond <- c("maxDist", "randomAll",
rep("randomK",
                nbRedrawing))[1:nbRedrawing]
        }
        else {
            startingCond <- rep(startingCond, nbRedrawing)
        }
    }
    else {
    }
    if (missing(distance)) {
        distance <- "euclidean"
    }
    if (is.character(distance)) {
        distInt <- pmatch(distance, METHODS)
    }
    else {
        distInt <- NA
    }
    if (print.traj) {
        cat(" ~ Slow KmL ~\n")
        fast <- FALSE
        screenPlot <- scr[2]
        if (!is.na(distInt)) {
            distanceSlow <- function(x, y) {
                dist(rbind(x, y), method = distance)
            }
        }
        else {
            distanceSlow <- distance
        }
    }
    else {
        screenPlot <- NA
        if (is.na(distInt)) {
            cat(" ~ Slow KmL ~\n")
            fast <- FALSE
            distanceSlow <- distance
        }
        else {
            cat(" ~ Fast KmL ~\n")
            fast <- TRUE
        }
    }
    nameObject <- deparse(substitute(Object))
    for (iRedraw in 1:nbRedrawing) {
        for (iNbClusters in nbClusters) {
            saveCld <- saveCld + 1
            clustersInit <- partitionInitialise(nbClusters = iNbClusters,
                method = startingCond[iRedraw], lengthPart = nbId,
                matrixDist = as.matrix(dist(trajNoNA, method = 
distanceStartingCond)))
            clust <- rep(NA, nbIdFull)
            if (fast) {
                resultKml <- .C("kml1", as.double(t(trajNoNA)),
                  iNbInd = as.integer(nbId), iNbTime = as.integer(nbTime),
                  iNbCluster = as.integer(iNbClusters), maxIt = 
as.integer(maxIt),
                  distance = as.integer(distInt), power = 
as.numeric(power),
                  vClusterAffectation1 = 
as.integer(clustersInit["clusters"]),
                  convergenceTime = as.integer(convergenceTime),
                  NAOK = TRUE, PACKAGE = "kml")[c(8, 9)]
                clust[noNA] <- resultKml[[1]]
            }
            else {
                resultKml <- trajKmlSlow(traj = trajNoNA, 
clusterAffectation = clustersInit,
                  nbId = nbId, nbTime = nbTime, maxIt = maxIt,
                  screenPlot = scr[2], distance = distanceSlow,
                  centerMethod = centerMethod, ...)
                clust[noNA] <- resultKml[[1]]["clusters"]
            }
            yPartition <- ordered(partition(nbClusters = iNbClusters,
                clusters = clust))
            Object["clusters"] <- clusterization(yLongData =
as(Object,
                "LongData"), xPartition = yPartition, convergenceTime
=
resultKml[[2]],
                imputationMethod = imputationMethod, startingCondition = 
startingCond[iRedraw],
                algorithmUsed = "kml")
            assign(nameObject, Object, envir = parent.frame())
            cat("*")
            if (saveCld >= saveFreq) {
                save(list = nameObject, file = paste(nameObject,
                  ".Rdata", sep = ""))
                saveCld <- 0
                cat("\n")
            }
            else {
            }
            if (print.cal) {
                screen(scr[1])
                plotCriterion(Object, all = TRUE)
            }
            else {
            }
        }
    }
    save(list = nameObject, file = paste(nameObject, ".Rdata",
        sep = ""))
    return(invisible())
}


######
### setting the kml method, using the same code
###
setMethod("kml","ClusterizLongData",func)

#######
### Same example that the one that does not work at the begining of this 
mail
###
kml(dn)

--- 8< --------------------------

Martin Morgan find the solution. Before
   setMethod("kml","ClusterizLongData",func)
kml was in environment kml, After, kml is in environment global.

So, using traceback(), we find that kml use an object Partition that is 
define in another package and that was export to the global environment 
but not to kml environment.
Adding import or importForm in NAMESPACE solve the problem.

Christophe
> Hi the list,
>
> I am writing a package in S4 and I do not manage to understand a bug.
> The "R CMD check" and the "R CMD build" both work. Here
is links to
> the package (not on CRAN yet for the raison that I explain bellow):
>
> http://christophe.genolini.free.fr/aTelecharger/kml_0.5.zip
> http://christophe.genolini.free.fr/aTelecharger/kml_0.5.tar.gz
>
> Then I install the package and I try an example:
>
> --- 8< --------------
> library(kml)
> dn <- as.cld(gald())
> kml(dn)
> # XXX ~ Fast KmL ~
> # Erreur dans as.vector(x, mode) : argument 'mode' incorrect
> --- 8< --------------
>
>
> So I make some verifications:
> --- 8< ----
> class(dn)
> # [1] "ClusterizLongData"
> # attr(,"package")
> # [1] "kml"
>
> getMethod("kml","ClusterizLongData")
> # Method Definition:
> #
> # function (Object, nbClusters = 2:6, nbRedrawing = 20, saveFreq = 100,
> #   maxIt = 200, trajMinSize = 2, print.cal = FALSE, print.traj = FALSE,
> #    imputationMethod = "copyMean", distance, power = 2,
centerMethod
> = meanNA,
> #    startingCond = "allMethods", distanceStartingCond =
"euclidean",
> #   ...)
> #{
> #   nbIdFull <- nrow(Object["traj"])
> # ...... [[[The full code is available below]]]
> # }
> # <environment: namespace:kml>
> #
> #Signatures:
> #       Object            # target  "ClusterizLongData"
> # defined "ClusterizLongData"
> --- 8< ----
>
> Everything seems fine. The code is correct.
> So I copy-and-paste the code that I get with 
> getMethods("kml","ClusterizLongData") and I affect it
to a function
> "func". Then I define again the method "kml".
>
> Then I run again the example that does not work before and it works...
> Any explanations?
>
> Christophe Genolini
>
> --- 8< --------------------------
> ###
> ### Affecting to func the code that 
> getMethod("kml","ClusterizLongData") delivers
> ###
> func <- function (Object, nbClusters = 2:6, nbRedrawing = 20, saveFreq 
> = 100,
>    maxIt = 200, trajMinSize = 2, print.cal = FALSE, print.traj = FALSE,
>    imputationMethod = "copyMean", distance, power = 2,
centerMethod =
> meanNA,
>    startingCond = "allMethods", distanceStartingCond =
"euclidean",
>    ...)
> {
>    nbIdFull <- nrow(Object["traj"])
>    convergenceTime <- 0
>    noNA <- selectSupTrajMinSize(Object, trajMinSize)
>    trajNoNA <- Object["traj"][noNA, ]
>    nbTime <- length(Object["time"])
>    nbId <- nrow(trajNoNA)
>    saveCld <- 0
>    scr <- plotAll(Object, print.cal = print.cal, print.traj =
print.traj,
>        print.sub = FALSE, col = "black", type.mean =
"n")
>    if (length(startingCond) == 1) {
>        if (startingCond == "allMethods") {
>            startingCond <- c("maxDist", "randomAll",
rep("randomK",
>                nbRedrawing))[1:nbRedrawing]
>        }
>        else {
>            startingCond <- rep(startingCond, nbRedrawing)
>        }
>    }
>    else {
>    }
>    if (missing(distance)) {
>        distance <- "euclidean"
>    }
>    if (is.character(distance)) {
>        distInt <- pmatch(distance, METHODS)
>    }
>    else {
>        distInt <- NA
>    }
>    if (print.traj) {
>        cat(" ~ Slow KmL ~\n")
>        fast <- FALSE
>        screenPlot <- scr[2]
>        if (!is.na(distInt)) {
>            distanceSlow <- function(x, y) {
>                dist(rbind(x, y), method = distance)
>            }
>        }
>        else {
>            distanceSlow <- distance
>        }
>    }
>    else {
>        screenPlot <- NA
>        if (is.na(distInt)) {
>            cat(" ~ Slow KmL ~\n")
>            fast <- FALSE
>            distanceSlow <- distance
>        }
>        else {
>            cat(" ~ Fast KmL ~\n")
>            fast <- TRUE
>        }
>    }
>    nameObject <- deparse(substitute(Object))
>    for (iRedraw in 1:nbRedrawing) {
>        for (iNbClusters in nbClusters) {
>            saveCld <- saveCld + 1
>            clustersInit <- partitionInitialise(nbClusters = iNbClusters,
>                method = startingCond[iRedraw], lengthPart = nbId,
>                matrixDist = as.matrix(dist(trajNoNA, method = 
> distanceStartingCond)))
>            clust <- rep(NA, nbIdFull)
>            if (fast) {
>                resultKml <- .C("kml1", as.double(t(trajNoNA)),
>                  iNbInd = as.integer(nbId), iNbTime = as.integer(nbTime),
>                  iNbCluster = as.integer(iNbClusters), maxIt = 
> as.integer(maxIt),
>                  distance = as.integer(distInt), power = 
> as.numeric(power),
>                  vClusterAffectation1 = 
> as.integer(clustersInit["clusters"]),
>                  convergenceTime = as.integer(convergenceTime),
>                  NAOK = TRUE, PACKAGE = "kml")[c(8, 9)]
>                clust[noNA] <- resultKml[[1]]
>            }
>            else {
>                resultKml <- trajKmlSlow(traj = trajNoNA, 
> clusterAffectation = clustersInit,
>                  nbId = nbId, nbTime = nbTime, maxIt = maxIt,
>                  screenPlot = scr[2], distance = distanceSlow,
>                  centerMethod = centerMethod, ...)
>                clust[noNA] <- resultKml[[1]]["clusters"]
>            }
>            yPartition <- ordered(partition(nbClusters = iNbClusters,
>                clusters = clust))
>            Object["clusters"] <- clusterization(yLongData =
as(Object,
>                "LongData"), xPartition = yPartition,
convergenceTime =
> resultKml[[2]],
>                imputationMethod = imputationMethod, startingCondition 
> = startingCond[iRedraw],
>                algorithmUsed = "kml")
>            assign(nameObject, Object, envir = parent.frame())
>            cat("*")
>            if (saveCld >= saveFreq) {
>                save(list = nameObject, file = paste(nameObject,
>                  ".Rdata", sep = ""))
>                saveCld <- 0
>                cat("\n")
>            }
>            else {
>            }
>            if (print.cal) {
>                screen(scr[1])
>                plotCriterion(Object, all = TRUE)
>            }
>            else {
>            }
>        }
>    }
>    save(list = nameObject, file = paste(nameObject, ".Rdata",
>        sep = ""))
>    return(invisible())
> }
>
>
> ######
> ### setting the kml method, using the same code
> ###
> setMethod("kml","ClusterizLongData",func)
>
> #######
> ### Same example that the one that does not work at the begining of 
> this mail
> ###
> kml(dn)
>
> --- 8< --------------------------
>